File: std.i

package info (click to toggle)
yorick 2.2.03+dfsg-3
links: PTS, VCS
area: main
in suites: jessie, jessie-kfreebsd
size: 9,620 kB
ctags: 9,317
sloc: ansic: 85,521; sh: 1,665; cpp: 1,282; lisp: 1,234; makefile: 1,034; fortran: 19
file content (6142 lines) | stat: -rw-r--r-- 253,875 bytes
/*
 * $Id: std.i,v 1.46 2010-09-13 03:05:58 dhmunro Exp $
 * Declarations of standard Yorick functions.
 */
/* Copyright (c) 2005, The Regents of the University of California.
 * All rights reserved.
 * This file is part of yorick (http://yorick.sourceforge.net).
 * Read the accompanying LICENSE file for details.
 */

/*
    The Codger automatic code generator program uses this file to
    generate appropriate C code to initialize the various built-in
    functions declared here.
    This file is also used as online documentation for these functions
    by Yorick's help mechanism.

    The "extern" declaration of each function or variable is a no-op,
    but causes Yorick to place the variable in the sourceList for this
    include file, making it available for online help.  The DOCUMENT
    comment is provided in a standard format to simplify manipulation
    of such comments by programs other than Yorick;  it should immediately
    follow the corresponding "extern" so that it will be visible when
    the page containing the "extern" is displayed.

    The Codger code generator finds each "extern" line and creates
    initialization code binding the associated Yorick variable to
    either a BuiltIn function (see ydata.h) Y_variable, or, if a
    "reshape, variable, ..." declaration is found, to a global
    compiled variable y_variable with the compiled data type
    corresponding to the Yorick data type mentioned in the "reshape"
    command.  Codger can generate certain simple Y_variable wrapper
    routines if further information is provided in a PROTOTYPE comment.
 */
/*= SECTION(info) help and information =====================================*/

extern help;
/* DOCUMENT help, topic
         or help
     Prints DOCUMENT comment from include file in which the variable
     TOPIC was defined, followed by the line number and filename.
     By opening the file with a text editor, you may be able to find
     out more, especially if no DOCUMENT comment was found.
     Examples:
       help, set_path
     prints the documentation for the set_path function.
       help
     prints the DOCUMENT comment you are reading.

     This copy of Yorick was launched from the directory:
     **** Y_LAUNCH (computed at runtime) ****
     Yorick's "site directory" at this site is:
     **** Y_SITE (computed at runtime) ****
     You can find out a great deal more about Yorick by browsing
     through these directories.  Begin with the site directory,
     and pay careful attention to the subdirectories doc/ (which
     contains documentation relating to Yorick if the yorick-doc
     package is installed) and i/ (which contain many examples of
     Yorick programs).
     Look for files called README (or something similar) in any
     of these directories -- they are intended to assist browsers.
     The site directory itself contains std.i and graph.i, which
     are worth reading.

     Type:
       help, dbexit
     for help on debug mode.  If your prompt is "dbug>" instead of
     ">", dbexit will return you to normal mode.

     Type:
       quit
     to quit Yorick.

   SEE ALSO: about, quit, info, print, copyright, warranty, legal
 */

local copyright, warranty;
/* DOCUMENT copyright, (no) warranty

     Copyright (c) 2005.  The Regents of the University of California.
                   All rights reserved.

     Yorick is provided "as is" without any warranty, either expressed or
     implied.  For a complete statement, type:

        legal

     at the Yorick prompt.

   SEE ALSO: legal
 */

func legal(void)
/* DOCUMENT legal
     Prints the legal details of Yorick's copyright, licensing,
     and lack of warranty.
   SEE ALSO: copyright, warranty
 */
{
  f = open(Y_HOME+"LICENSE.md", "r", 1);
  if (!f) error, "missing LICENSE file, yorick improperly installed";
  f = rdline(f, 1000);
  write, format="%s\n", f(where(f));
}

func help_worker
/* xxDOCUMENT help_worker (Not for interactive use -- called by help.)
 */
{
  /* help_worker task is pushed by help function -- topic and file
     arguments are left in help_topic and help_file variables */
  topic= help_topic;   help_topic= [];
  file= help_file;     help_file= [];

  if (file) {
    mark= bookmark(file);
    line= rdline(file);

    if (typeof(topic)!="struct_definition") {
      /* non-struct looks for DOCUMENT comment before any blank lines */
      n= 10;   /* read at most 10 lines looking for DOCUMENT comment */
      while (strtok(line)(1) && n--) {
        if (strmatch(line, "/* DOCUMENT")) break;
        line= rdline(file);
      }
      if (strmatch(line, "/* DOCUMENT")) {
        do {
          if (strmatch(line, "**** Y_LAUNCH (computed at runtime) ****"))
            write, "      "+Y_LAUNCH;
          else if (strmatch(line, "**** Y_SITE (computed at runtime) ****"))
            write, "      "+Y_SITE;
          else
            write, line;
          line= rdline(file);
          if (!line) break;
        } while (!strmatch(line, "*/"));
        write, line;
      } else {
        write, "<DOCUMENT comment not found>";
      }

    } else {
      /* struct just prints definition */
      gotopen= 0;
      do {
        if (!gotopen) gotopen= strmatch(line, "{");
        write, line;
        if (gotopen && strmatch(line, "}")) break;
      } while (line= rdline(file));
    }

    mark= print(mark)(2:0);
    line= "";
    for (i=1 ; i<numberof(mark) ; i++) line+= strpart(mark(i),1:-1);
    line+= mark(i);
    write, "defined at:"+line;

  } else if (is_func(topic) == 3) {
    /* autoloaded function */
    buf = print(topic);
    n = numberof(buf);
    str = buf(1);
    escape = "\134"; /* must be octal code to not kill codger */
    newline = "\n";
    for (i=2;i<=n;++i) {
      if (strpart(str, 0:0) == escape) {
        str = strpart(str, 1:-1) + buf(i);
      } else {
        str += newline + buf(i);
      }
    }
    topic_name = file_name = string();
    if (sread(str, format="autoload of: %s from: %[^\n]",
              topic_name, file_name) == 2) {
      include, file_name, 1;
      help_topic = topic_name;
      after, 0.0, _help_auto;
    } else {
      info, topic;
    }
  } else {
    write, "<not defined in an include file, running info function>";
    info, topic;
  }
}

func _help_auto
/* xxDOCUMENT _help_auto (Not for interactive use -- called by help_worker
 *                        for autoloaded function.) */
{
  topic = help_topic;
  help_topic = [];
  if (structof(topic) == string) topic = symbol_def(topic);
  help, topic;
}

func info(args)
/* DOCUMENT info, expr [, expr2, expr3, ...]
     prints the data type and array dimensions of EXPR.  Multiple
     expressions result in multiple descriptions.  You can also
     invoke info as a function to return a string or array of strings
     instead of printing the result.
   SEE ALSO: about, help, print
 */
{
  /* void = use_origins(1); // either worthless or wrong? */
  r = args("_depth");
  prefix = r? string(&array(' ',r+r)) : "";
  value = [];
  s = am_subroutine();
  for (j=1 ; j<=args(0) ; ++j) {
    if (r) line = prefix + args("_name") + " = ";
    else if (args(0)>1) line = (args(-,j)? args(-,j) : "<expr>") + " = ";
    else line = "";
    if (is_array(args(j,:))) {
      line += "array(" + nameof(structof(args(j,:)));
      dims = dimsof(args(j,:));
      orgs = orgsof(args(j,:));
      ndims = dims(1)+1;
      for (i=2 ; i<=ndims ; i++) {
        line += ",";
        if (orgs(i)!=1)
          line += print(orgs(i))(1)+":"+print(orgs(i)+dims(i)-1)(1);
        else
          line += print(dims(i))(1);
      }
      line = swrite(line + ")");
    } else if (is_stream(args(j)) || !is_obj(args(j))) {
      x = print(args(j));
      x(1) = line + x(1);
      line = swrite(format=" %s", x);
    } else {
      o = args(j);
      n = o(*);
      nm = o(*,);
      line = swrite(format=" %sobject with %ld members:", line, n);
      n = min(n, 99);
      for (i=1 ; i<=n ; ++i) {
        k = _name = nm(i);
        if (!k) {
          _name = swrite(format="<%ld>", i);
          k = i;
        }
        grow, line, info(o(noop(k)), _depth=(r?r+1:1), _name=_name);
      }
      n = o(*);
      if (i <= n)
        grow, line, swrite(format=" %s  <%ld more>...", prefix, n-i+1);
    }
    if (s) write, format="%s\n", line;
    else grow, value, line;
  }
  return value;
}
wrap_args, info;

/* need strange variable names to avoid matches with help topic pattern! */
func about(____n____, ____a____)
/* DOCUMENT about, pattern;
         or about, pattern, 1;
     Search and display documentation about functions (or all symbols if
     second argument is true) matching regular expression PATTERN.  If
     multiple matches are found, the user is prompted to select a subject.
     PATTERN may be a string, or a function or structure definition.  If
     PATTERN is a string with a trailing "/i", the other part of the
     regular expression is interpreted so as to ignore case.

     Set
       about_glob = strglob;
     to use UNIX shell style matching (e.g.- ls command line) to match
     PATTERN instead of grep style matching.  The default about_glob
     function is strgrepm.

   SEE ALSO: help, info, strgrep, strglob
*/
{
  if (structof(____n____) != string) {
    ____n____ = nameof(____n____);
    if (structof(____n____) != string)
      error, "expecting a string, a function, or a structure definition";
  }
  /* first get list of all symbol names matching the pattern */
  ____c____ = strpart(____n____, -1:0) == "/i";
  if (____c____) ____n____ = strpart(____n____, 1:-2);
  ____b____ = save(*)(*,);
  ____b____ = ____b____(where(about_glob(____n____,____b____,case=____c____)));
  if (!____a____) {
    /* by default remove all matches which are not functions */
    for (____a____=1 ; ____a____<=numberof(____b____) ; ++____a____)
      if (!is_func(symbol_def(____b____(____a____))))
        ____b____(____a____) = string(0);
  }
  ____a____ = ____b____(where(____b____));
  if (!numberof(____a____)) {
    write, " Sorry no match found.";
    return;
  }
  if (numberof(____a____) == 1)
    ____a____ = ____a____(1);
  else
    ____a____ = select_name(____a____(sort(____a____)), bol=" ",
                            prompt=" Choose one subject: ");
  if (numberof(____a____)) {
    /* Explicitly filter range operators which have a built-in
     * function counterpart to avoid ambiguity in yorick. */
    if (____a____ == "sum") {
      write, format="%s\n%s\n%s\n%s\n",
        "/* DOCUMENT sum(x)",
        "     Returns sum of values in array X.",
        "     Can also be used as a range operator.",
        "   SEE ALSO: avg, max, min. */";
    } else if (____a____ == "avg") {
      write, format="%s\n%s\n%s\n%s\n",
        "/* DOCUMENT avg(x)",
        "     Returns average of values in array X.",
        "     Can also be used as a range operator.",
        "   SEE ALSO: max, min, sum. */";
    } else if (____a____ == "min") {
      write, format="%s\n%s\n%s\n%s\n",
        "/* DOCUMENT min(x)",
        "     Returns minimum value in array X.",
        "     Can also be used as a range operator.",
        "   SEE ALSO: avg, max, sum. */";
    } else if (____a____ == "max") {
      write, format="%s\n%s\n%s\n%s\n",
        "/* DOCUMENT max(x)",
        "     Returns maximum value in array X.",
        "     Can also be used as a range operator.",
        "   SEE ALSO: avg, min, sum. */";
    } else {
      include, ["help, "+____a____];
    }
  }
}

func library(void)
/* DOCUMENT library
     print the Y_SITE/i/README file at the terminal.
 */
{
  f= open(Y_SITE+"i/README");
  while ((line= rdline(f))) write, line;
}

extern get_pkgnames;
/* DOCUMENT get_pkgnames(all)
     returns list of package names, ALL non-zero means to return both
     statically and dynamically loaded packages, otherwise just the
     initial statically loaded packages.
   SEE ALSO: get_path
 */

extern symbol_def;
/* DOCUMENT symbol_def(func_name)(arglist)
         or symbol_def(var_name)
     invokes the function FUNC_NAME with the specified ARGLIST,
     returning the return value.  ARGLIST may be zero or more arguments.
     In fact, symbol_def("fname")(arg1, arg2, arg3) is equivalent to
     fname(arg1, arg2, arg3), so that "fname" can be the name of any
     variable for which the latter syntax is meaningful -- interpreted
     function, built-in function, or array.

     Without an argument list, symbol_def("varname") is equivalent to
     varname, which allows you to get the value of a variable whose name
     you must compute.

     DO NOT OVERUSE THIS FUNCTION.  It works around a specific deficiency
     of the Yorick language -- the lack of pointers to functions -- and
     should be used for such purposes as hook lists (see openb).

   SEE ALSO: symbol_set, symbol_exists
 */

extern symbol_set;
/* DOCUMENT symbol_set, var_name, value
     is equivalent to the redefinition
          varname= value
     except that var_name="varname" is a string which must be computed.

     DO NOT OVERUSE THIS FUNCTION.  It works around a specific deficiency
     of the Yorick language -- the lack of pointers to functions, streams,
     bookmarks, and other special non-array data types.

   SEE ALSO: symbol_def, symbol_exists
 */

extern symbol_exists;
/* DOCUMENT symbol_exists(name)
     Check whether variable/function named NAME exists.  This routine can be
     used prior to symbol_def to check existence of a symbol since symbol_def
     raise an error for non-existing symbol.

   SEE ALSO: symbol_def, symbol_names, symbol_set.
*/

extern symbol_names;
/* DOCUMENT symbol_names()
         or symbol_names(flags)
     Return an  array of  strings with  the names of  all symbols  of given
     type(s) found in  global symbol table.  To select  the type of symbol,
     FLAGS is be the bitwise-or of one or more of the following bits:
         1 - basic array symbols
         2 - structure instance symbols
         4 - range symbols
         8 - nil symbols (i.e. symbols undefined at current scope level)
        16 - interpreted function symbols
        32 - builtin function symbols
        64 - structure definition symbols
       128 - file stream symbols
       256 - opaque symbols (other than the ones below)
       512 - list objects
      1024 - auto-loaded functions

     The special value FLAGS = -1 can be used to get all names found in
     global symbol table.  The default (if FLAGS is nil or omitted) is to
     return the names of all symbols but the nil ones.  Beware that lists,
     hash tables and auto-loaded functions are also opaque symbols (use
     0xffffff7f to get *all* opaque symbols).

   SEE ALSO: symbol_def, symbol_exists, symbol_set.
*/

func split_path(path)
/* DOCUMENT split_path(path)
     splits PATH, a colon or semi-colon delimited list of directories
     as returned by get_path, into a string array with one directory
     per element.
   SEE ALSO: set_path, get_pkgnames
 */
{
  if (numberof(path) > 1) path = sum(";"+path);
  path = strchar(path);
  list = where(path == ';');
  if (numberof(list)) path(list) = '\0';
  list = where(path == ':');
  if (numberof(list)) {
    for (i=j=1 ; j<=numberof(list) ; ++j) {
      for (; i<list(j) ; ++i) if (path(i)) break;
      if ((i==list(j)-1) && ((c=path(i)|'\x20')>='a') && (c<='z')) {
        list(j) = 0;
        if (++j > numberof(list)) break;
      }
      i = list(j) + 1;
    }
    list = list(where(list));
    if (numberof(list)) path(list) = '\0';
  }
  return strchar(path);
}

/*= SECTION(advanced) performance and interface optimizations ==============*/

extern eq_nocopy;
/* DOCUMENT eq_nocopy, y, x
     is the same as
            y= x
     except that if x is an array, it is not copied, even if it is
     not a temporary (i.e.- an expression).  Having multiple variables
     reference the same data can be confusing, which is why the default
     = operation copies the array.  The most important use of eq_nocopy
     involves pointers or lists:
            y= *py
            z= _car(list)
     always causes the data pointed to by py to be copied, while
            eq_nocopy, y, *py
            eq_nocopy, z, _car(list)
     does not copy the data - often more nearly what you wanted.
     Note that scalar int, long, and double variables are always copied,
     so you cannot count on eq_nocopy setting up an "equivalence"
     between variables.

   SEE ALSO: swap, unref.
 */

extern swap;
/* DOCUMENT swap, a, b;
     Exchanges the contents of variables A and B without requiring any
     temporary copy.  The result of the call is identical to:
       tmp = a; a = b; b = tmp;
     which makes a copy of A and then a copy of B.  Another possibility which
     avoids any copy of A nor B is:
       local tmp;
       eq_nocopy, tmp, a; eq_nocopy, a, b; eq_nocopy, b, tmp;

   SEE ALSO: eq_nocopy, unref.
 */

extern unref;
/* DOCUMENT unref(x)
     Returns X, destroying X in the process if it is an array (useful to
     deal with temporary big arrays).

   SEE ALSO: eq_nocopy, swap.
 */

extern wrap_args;
/* DOCUMENT wrap_args, interpreted_function
     converts INTERPRETED_FUNCTION to a wrapped_func object,
     which will accept an arbitrary argument list, then invoke
     INTERPRETED_FUNCTION with a single wrapped_args object as
     its argument.  The INTERPRETED_FUNCTION must be declared as:

     func INTERPRETED_FUNCTION(args)
     {
       ...use args object to retrieve actual arguments...
     }
     wrap_args, INTEPRETED_FUNCTION;

     After wrapping, you invoke the function as usual:

     result = INTERPRETED_FUNCTION(arg1, key1=ka1, arg2, ...);

     Unlike an ordinary interpreted function, a wrapped function
     will accept any number of arguments, and keyword arguments
     of any name.  Furthermore, unlike an ordinary function, you
     can determine the number of arguments passed to the function,
     the names of any simple variable references passed to the
     function, and other useful information about the arguments.
     You can set the external value of any simple variable passed
     as an argument, as if it had been declared func f(..., &x, ...).

     A wrapped_func function call is less efficient and requires
     less transparent coding than an ordinary function call; the
     advantage is that you can write a wrapped function which has
     non-standard semantics, for example, like the save and restore
     built-in functions (which use the names of the arguments passed
     to them), or other special effects, like accepting arbitrary
     keyword names.

     The ARGS object, the single argument actually passed to
     your INTEPRETED_FUNCTION, is a wrapped_args object, which has
     the following Eval methods:

       ARGS(-)   returns [keyname1, keyname2, keyname3, ...]
                   the actual names of the keyword arguments passed
                   or nil [] if no keywords were passed
                   ARGS(*,) is a synonym for ARGS(-) to resemble
                   the object syntax (see help,oxy), although the
                   analogy is not exact.
       ARGS(0)   returns the number of positional arguments passed
                   ARGS(*) is a synonym for ARGS(0) to resemble
                   the object syntax (see help,oxy), although the
                   analogy is not exact.
       ARGS(i)   returns the i-th positional argument
                   i can also be a string to return a keyword argument,
                     or a negative number to return the -i-th keyword
       ARGS, i, value
                 sets the value of argument i, as if it were
                   an output variable declared as func f(..., &x, ...)
       ARGS(-,i) returns the name of argument i if it was
                   passed as a simple variable reference
                   ARGS(*,i) is a synonym for ARGS(-,i) to resemble
                   the object syntax (see help,oxy), although the
                   analogy is not exact.
       ARGS(0,i) returns a flag describing the argument:
         0 if argument is a simple variable reference (set value works)
         1 if argument is an expression (set value will be discarded)
         2 if argument does not exist (as opposed to simply nil)
     For obscure situations, there is also:
       ARGS(i,:) same as ARGS(i), except if the argument is an lvalue,
                   it is not fetched.
     This rather arcane feature permits you to pass an argument of the
     form f.x, where f is a file handle, to functions like dimsof or
     structof without triggering a read of the file.  Do not assign
     the result to a variable; use it only as an argument to another
     function.  The first time you call ARGS(i) for argument i, the
     lvalue is fetched, and ARGS(i,:) will do nothing special.
  SEE ALSO: errs2caller
 */

extern errs2caller;
/* DOCUMENT errs2caller, f1, f2, ...
     makes function F1 (and optionally F2, ...) pass control for dbug
     mode to its caller if a fault occurs inside F1.  This makes F1
     behave more like a compiled function for its caller.  For example,
     if you are writing a mathematical function, you can raise an
     error in its caller rather than in the function itself -- which
     is appropriate if the only errors your function raises are, for
     example, domain errors.  Your function will then respond to a
     domain error in the same way as, for example, asin(1.5).
     If you want to wrap arguments of such a function, you need to call
     errs2caller before wrap_args.
  SEE ALSO: wrap_args
 */

/*= SECTION(query) finding out variable type and properties ================*/

extern structof;
/* DOCUMENT structof(object)
     returns the data type of OBJECT, or nil for non-array OBJECTs.
     Use typeof(object) to get the ASCII name of a the data type.
  SEE ALSO: typeof, dimsof, numberof, sizeof, nameof
 */

extern dimsof;
/* DOCUMENT dimsof(object)
         or dimsof(object1, object2, ...)
     returns a vector of integers describing the dimensions of OBJECT.
     The format of the vector is [number of dims, length1, length2, ...].
     The orgsof function returns the origin of each dimension (normally 1).
     If more than one argument is given, dimsof returns the dimension
     list of the result of binary operations between all the objects,
     or nil if the objects are not conformable.
  SEE ALSO: typeof, structof, numberof, sizeof, orgsof
 */

extern orgsof;
/* DOCUMENT orgsof(object)
     returns a vector of integers describing the dimensions of OBJECT.
     The format of the vector is [number of dims, origin1, origin2, ...].
     By default, dimension origins are ignored, but use_origins changes
     this.  The dimsof function returns the length of each dimension.
     *** NOTE NOTE NOTE ***
     Unless use_origins(1) is in effect, orgsof will always return
     1 for all of the originI in the list.  Thus, whether use_origins(1)
     is in effect or not, you are guaranteed that x(orgsof(x)(2)) is the
     first element of x.

     *** DEPRECATED ***
     Do not use index origins.  Your brain will explode sooner or later.

  SEE ALSO: dimsof, typeof, structof, numberof, sizeof, use_origins
 */

extern use_origins;
/* DOCUMENT dummy= use_origins(dont_force)
     Yorick array dimensions have an origin as well as a length.
     By default, this origin is 1 (like FORTRAN arrays, unlike C
     arrays).  However, the array function and the pseudo-index (-)
     can be used to produce arrays with other origins.

     Initially, the origin of an array index is ignored by Yorick; the
     first element of any array has index 1.  You can change this
     default behavior by calling use_origins with non-zero DONT_FORCE,
     and restore the default behavior by calling use_origins(0).

     When the returned object DUMMY is destroyed, either by return from
     the function in which it is a local variable, or by explicit
     redefintion of the last reference to it, the treatment of array
     index origins reverts to the behavior prior to the call to
     use_origins.  Thus, you can call use_origins at the top of a
     function and not worry about restoring the external behavior
     before every possible return (including errors).

     *** DEPRECATED ***
     Do not use index origins.  Your brain will explode sooner or later.

  SEE ALSO: array, dimsof, orgsof
 */

extern sizeof;
/* DOCUMENT sizeof(object)
     returns the size of the object in bytes, or 0 for non-array objects.
     sizeof(structure_definition) returns the number of bytes per instance.
     sizeof(binary_file) returns the file size in bytes.
  SEE ALSO: dimsof, typeof, structof, numberof
 */

extern numberof;
/* DOCUMENT numberof(object)
     returns the number of elements if object is an array, or 0 if not.
  SEE ALSO: sizeof, dimsof, typeof, structof
 */

extern typeof;
/* DOCUMENT typeof(object)
     returns a string describing the type of object.  For the basic
     data types, these are "char", "short", "int", "long", "float",
     "double", "complex", "string", "pointer", "struct_instance",
     "void", "range", "struct_definition", "function", "builtin",
     "stream" (for a binary stream), and "text_stream".
  SEE ALSO: structof, dimsof, sizeof, numberof, nameof
 */

extern nameof;
/* DOCUMENT nameof(object)
     If OBJECT is a function or a structure definition, returns the
     name of the func or struct as it was defined (not necessarily
     the name of the variable passed to the nameof function).
  SEE ALSO: typeof
 */

extern is_array;
/* DOCUMENT is_array(object)
     returns 1 if OBJECT is an array data type (as opposed to a function,
     structure definition, index range, I/O stream, etc.), else 0.
     An array OBJECT can be written to or read from a binary file;
     non-array Yorick data types cannot.
  SEE ALSO: is_func, is_void, is_range, is_struct, is_stream, is_scalar
 */

extern is_func;
/* DOCUMENT is_func(object)
     returns 1 if OBJECT is a Yorick interpreted function, 2 if OBJECT
     is a built-in (that is, compiled) function, 3 if OBJECT is an
     autoload, 4 if object is a wrap_args function, 5 if object
     is a closure function, else 0.
  SEE ALSO: is_array, is_void, is_range, is_struct, is_stream, autoload,
            closure
 */

extern is_void;
/* DOCUMENT is_void(object)
     returns 1 if OBJECT is nil (the one instance of the void data type),
     else 0.
  SEE ALSO: is_array, is_func, is_range, is_struct, is_stream
 */

extern is_range;
/* DOCUMENT is_range(object)
     returns 1 if OBJECT is an index range (e.g.-  3:5 or 11:31:2),
     else 0.
  SEE ALSO: is_array, is_func, is_void, is_struct, is_stream
 */

extern is_struct;
/* DOCUMENT is_struct(object)
     returns 1 if OBJECT is the definition of a Yorick struct, else 0.
     Thus, is_struct(double) returns 1, but is_struct(1.0) returns 0.
  SEE ALSO: is_array, is_func, is_void, is_range, is_stream
 */

extern is_stream;
/* DOCUMENT is_stream(object)
     returns 1 if OBJECT is a binary I/O stream (usually a file), else 0.
     The _read and _write functions work on object if and only if
     is_stream returns non-zero.  Note that is_stream returns 0 for a
     text stream -- you need the typeof function to test for those.
  SEE ALSO: is_array, is_func, is_void, is_range, is_struct
 */

extern is_list;
/* DOCUMENT is_list(object)
     returns 1 if OBJECT is a list or nil, else 0 (see _lst).
  SEE ALSO: is_array, is_func, is_void, is_range, is_struct, _lst
 */

extern is_scalar;
extern is_vector;
extern is_matrix;
/* DOCUMENT is_scalar(x)
         or is_vector(x)
         or is_matrix(x)
     These functions return true if X is (respectively) a scalar, a vector
     (i.e., a 1-D array), or a matrix (i.e., a 2-D array).

    SEE ALSO: dimsof,
              is_array, is_func, is_hash, is_integer, is_list, is_range,
              is_stream, is_struct, is_void.
 */

extern is_integer;
extern is_real;
extern is_complex;
extern is_numerical;
extern is_string;
extern is_pointer;
/* DOCUMENT is_integer(x)
         or is_real(x)
         or is_complex(x)
         or is_numerical(x)
         or is_string(x)
         or is_pointer(x)
     These functions  return true if  X is an  array of type:  integer, real
     (i.e.  double or  float), complex,  numerical (i.e.  integer,  real or
     complex), string, or pointer.

   SEE ALSO: structof, dimsof,
             is_array, is_func, is_hash, is_list, is_range, is_scalar,
             is_stream, is_struct, is_void.
 */

extern identof;
local Y_CHAR, Y_SHORT, Y_INT, Y_LONG;
local Y_FLOAT, Y_DOUBLE, Y_COMPLEX;
local Y_STRING, Y_POINTER, Y_STRUCT;
local Y_RANGE, Y_LVALUE, Y_VOID;
local Y_FUNCTION, Y_BUILTIN;
local Y_STRUCTDEF, Y_STREAM, Y_OPAQUE;
/* DOCUMENT identof(object)
     Returns type identifier of OBJECT as a long integer:
       0 (Y_CHAR)      for an array of char('s)
       1 (Y_SHORT)     for an array of short('s)
       2 (Y_INT)       for an array of int('s)
       3 (Y_LONG)      for an array of long('s)
       4 (Y_FLOAT)     for an array of float('s)
       5 (Y_DOUBLE)    for an array of double('s)
       6 (Y_COMPLEX)   for an array of complex('s)
       7 (Y_STRING)    for an array of string('s)
       8 (Y_POINTER)   for an array of pointer('s)
       9 (Y_STRUCT)    for a structure object
      10 (Y_RANGE)     for a range object
      11 (Y_LVALUE)    for a lvalue
      12 (Y_VOID)      for a void (undefined) object
      13 (Y_FUNCTION)  for a function array
      14 (Y_BUILTIN)   for a builtin array
      15 (Y_STRUCTDEF) for a data type or structure definition
      16 (Y_STREAM)    for a file stream
      17 (Y_OPAQUE)    for an opaque object

  SEE ALSO: typeof, structof.
 */
Y_CHAR = 0;
Y_SHORT = 1;
Y_INT = 2;
Y_LONG = 3;
Y_FLOAT = 4;
Y_DOUBLE = 5;
Y_COMPLEX = 6;
Y_STRING = 7;
Y_POINTER = 8;
Y_STRUCT = 9;
Y_RANGE = 10;
Y_LVALUE = 11;
Y_VOID = 12;
Y_FUNCTION = 13;
Y_BUILTIN = 14;
Y_STRUCTDEF = 15;
Y_STREAM = 16;
Y_OPAQUE = 17;

extern am_subroutine;
/* DOCUMENT am_subroutine()
     returns 1 if the current Yorick function was invoked as a subroutine,
     else 0.  If am_subroutine() returns true, the result of the current
     function will not be used, and need not be computed (the function
     has been called for its side effects only).
 */

/*= SECTION(math) simple mathematical functions ============================*/

extern sin;
extern cos;
extern tan;
/* DOCUMENT sin(x)
            cos(x)
            tan(x)
     returns the sine, cosine, or tangent of its argument,
     which is in radians.
  SEE ALSO: asin, acos, atan
 */

extern asin;
/* DOCUMENT asin(x)
     returns the inverse sine of its argument, range [-pi/2, pi/2].
  SEE ALSO: sin, cos, tan, asin, acos, atan
 */

extern acos;
/* DOCUMENT acos(x)
     returns the inverse cosine of its argument, range [0, pi].
  SEE ALSO: sin, cos, tan, asin, acos, atan
 */

extern atan;
/* DOCUMENT atan(x)
         or atan(y, x)
     returns the inverse tangent of its argument, range [-pi/2, pi/2].
     In the two argument form, returns the angle from (1, 0) to (x, y),
     in the range (-pi, pi], with atan(1, 0)==pi/2.  (If x>=0, this is
     the same as atan(y/x).)
  SEE ALSO: sin, cos, tan, asin, acos, atan
 */

local pi;
/* DOCUMENT pi
     roughly 3.14159265358979323846264338327950288
 */
pi= 4.0*atan(1.0);    /* to double precision on this machine */

extern sinh;
extern cosh;
extern tanh;
/* DOCUMENT sinh(x)
            cosh(x)
            tanh(x)
     returns the hyperbolic sine, cosine, or tangent of its argument.
  SEE ALSO: sech, csch, asinh, acosh, atanh
 */

func sech(x) { x = exp(_neg_re(x));  return (x+x)/(1.+x*x); }
func csch(x) { y = _neg_re(x,x);  return (4.*x-2.)*exp(y)/expm1(y+y); }
/* DOCUMENT sech(x)
            csch(x)
     returns the hyperbolic secant (1/cosh) or cosecant (1/sinh) of
     its argument, without overflowing for large x.
  SEE ALSO: sinh, cosh, tanh, asinh, acosh, atanh
 */
func _neg_re(x,&m) { m = double(double(x)<0.);  return m*x - (1.-m)*x; }

/* note: factorization in acosh prevents possible overflow
 *       asinh = log(x+sqrt(x*x+1.0)) has both overflow problem
 *       and small x problem */
func asinh(x) { y=-_neg_re(x,x);  return (1.-2.*x)*log1p(y+_sqrt_x2p1m1(y)); }
func acosh(x) { return log(x+sqrt(x+1.)*sqrt(x-1.)); }
func atanh(x) { y=_neg_re(x,x);  return (x-0.5)*log1p((y+y)/(1.0-y)); }
/* DOCUMENT asinh(x)
            acosh(x)
            atanh(x)
     returns the inverse hyperbolic sine, cosine, or tangent of
     its argument.  The range of real acosh is >=0.0.
  SEE ALSO: sinh, cosh, tanh, sech, csch
 */
func _sqrt_x2p1m1(x)
{
  mask = abs(x) > 1.e18;
  b = x(where(mask));
  if (numberof(b)) {   /* avoid overflow for big x */
    s = 1./b;
    b *= sqrt(1.+s*s);
    b = -(_neg_re(b)+1.);
  }
  s = x(where(!mask));
  if (numberof(s)) {   /* avoid rounding error for small x */
    s *= s;
    s /= (sqrt(1.+s) + 1.);
  }
  return merge(b, s, mask);
}

extern exp;
/* DOCUMENT exp(x)
     returns the exponential function of its argument (inverse of log).
  SEE ALSO: expm1, log, log10, sinh, cosh, tanh, sech, csch
 */

extern log;
/* DOCUMENT log(x)
     returns the natural logarithm of its argument (inverse of exp).
  SEE ALSO: log1p, log10, exp, asinh, acosh, atanh
 */

extern log10;
/* DOCUMENT log10(x)
     returns the base 10 logarithm of its argument (inverse of 10^x).
  SEE ALSO: log, exp, asinh, acosh, atanh
 */

func expm1(x, &ex)
/* DOCUMENT expm1(x)
         or expm1(x, ex)
     return exp(X)-1 accurate to machine precision (even for X<<1)
     in the second form, returns exp(x) to EX
   SEE ALSO: exp, log1p
 */
{
  ex = exp(x);
  return (ex-1.) + (x-log(ex+!ex))*ex;
}

func log1p(x)
/* DOCUMENT log1p(x)
     return log(1+X) accurate to machine precision (even for X<<1)
     from Goldberg, ACM Computing Surveys, Vol 23, No 1, March 1991,
       apparently originally from HP-15C Advanced Functions Handbook
   SEE ALSO: expm1, log1p
 */
{
  y = 1.+x;  z = double(y == 1.);
  return x * (log(y)+z)/(y-1.+z);
}

errs2caller, sech, csch, asinh, acosh, atanh, expm1, log1p;

extern sqrt;
/* DOCUMENT sqrt(x)
     returns the square root of its argument.
  SEE ALSO: abs
 */

extern poly;
/* DOCUMENT poly(x, a0, a1, a2, ..., aN)
     returns the polynomial  A0 + A1*x + A2*x^2 + ... + AN*X^N
     The data type and dimensions of the result, and conformability rules
     for the inputs are identical to those for the expression.
 */

extern ceil;
/* DOCUMENT ceil(x)
     returns the smallest integer not less than x (no-op on integers).
  SEE ALSO: floor, round
 */

extern floor;
/* DOCUMENT floor(x)
     returns the largest integer not greater than x (no-op on integers).
  SEE ALSO: ceil, round
 */

extern lround;
extern round;
/* DOCUMENT round(x);
            lround(x);
     These functions return X rounded to the nearest integer.  The result of
     round(X) is a floating point value, while that of lround(X) is a long
     integer.  They are respectively equivalent to: floor(X+0.5) and
     long(floor(X+0.5)).
   SEE ALSO: floor, ceil
*/

extern abs;
/* DOCUMENT abs(x)
         or abs(x, y, z, ...)
     returns the absolute value of its argument.
     In the multi-argument form, returns sqrt(x^2+y^2+z^2+...).
  SEE ALSO: sign, sqrt
 */

extern sign;
/* DOCUMENT sign(x)
     returns algebraic sign of it argument, or closest point on the
     unit circle for complex x.  Guaranteed that x==sign(x)*abs(x).
     sign(0)==+1.
  SEE ALSO: abs
 */

extern conj;
/* DOCUMENT conj(z)
     returns the complex conjugate of its argument.
 */

local re_part;
/* DOCUMENT re_part(z)
     returns the real part of its argument.  (Same as double(z).)
     Unlike z.re, works if z is not complex.
 */
re_part= double;

func im_part(z)
/* DOCUMENT im_part(z)
     returns the imaginary part of its argument.
     Unlike z.im, works if z is not complex (returns zero).
 */
{
  return (structof(z)==complex)? z.im : array(0.0, dimsof(z));
}

extern random;
extern random_seed;
/* DOCUMENT random(dimension_list)
            random_seed, seed
     returns an array of random double values with the given
     DIMENSION_LIST (nil for a scalar result), uniformly distributed
     on the interval from 0.0 to 1.0.
     The algorithm is from Press and Teukolsky, Computers in Physics,
     vol. 6, no. 5, Sep/Oct 1992 (ran2).  They offer a reward of $1000
     to anyone who can exhibit a statistical test that this random
     number generator fails in a "non-trivial" way.
     The random_seed call reinitializes the random number sequence;
     SEED should be between 0.0 and 1.0 non-inclusive; if SEED is
     omitted, nil, or out of range, the sequence is reinitialized as
     when Yorick starts.
     The numbers are actually at the centers of 2147483562 equal width
     bins on the interval [0,1].  Although only these 2 billion numbers
     are possible, the period of the generator is roughly 2.3e18.

   SEE ALSO: randomize
 */

func randomize(void)
/* DOCUMENT randomize
            randomize()
     set the seed for random "randomly" (based on the timer clock
     and the current state of random).  As a function, returns the
     value of the seed passed to random_seed.

   SEE ALSO: random, random_seed
 */
{
  seed= array(0., 3);
  timer, seed;
  seed= pi*sum(abs(seed));
  while (seed > 0.9) seed*= 0.1;
  seed+= 0.05;
  random_seed, seed;
  return seed;
}

/*= SECTION(array) building and manipulating arrays ========================*/

extern array;
/* DOCUMENT array(value, dimension_list)
         or array(type, dimension_list)
     returns an object of the same type as VALUE, consisting of copies
     of VALUE, with the given DIMENSION_LIST appended to the dimensions
     of VALUE.  Hence, array(1.5, 3, 1) is the same as [[1.5, 1.5, 1.5]].
     In the second form, the VALUE is taken as scalar zero of the TYPE.
     Hence, array(short, 2, 3) is the same as [[0s,0s],[0s,0s],[0s,0s]].
     A DIMENSION_LIST is a list of arguments, each of which may be
     any of the following:
        (1) A positive scalar integer expression,
        (2) An index range with no step field (e.g.-  1:10), or
        (3) A vector of integers [number of dims, length1, length2, ...]
            (that is, the format returned by the dimsof function).
  SEE ALSO: reshape, is_array, dimsof, numberof, grow, span, use_origins,
            _lst
 */

extern allof;
extern anyof;
extern noneof;
extern nallof;
/* DOCUMENT allof(x)
            anyof(x)
            nallof(x)
            noneof(x)
     Respectively:
      returns 1 if every element of the array x is non-zero, else 0.
      returns 1 if at least one element of the array x is non-zero, else 0.
      returns 1 if at least one element of the array x is zero, else 0.
      returns 1 if every element of the array x is zero, else 0.
  SEE ALSO: allof, anyof, noneof, nallof, where, where2
 */

extern where;
/* DOCUMENT where(x)
     returns the vector of longs which is the index list of non-zero
     values in the array x.  Thus, where([[0,1,3],[2,0,4]]) would
     return [2,3,4,6].  If noneof(x), where(x) is a special range
     function which will return a nil value if used to index an array;
     hence, if noneof(x), then x(where(x)) is nil.
     If x is a non-zero scalar, then where(x) returns a scalar value.
     The rather recondite behavior for scalars and noneof(x) provides
     maximum performance when the merge function to be used with the
     where function.
  SEE ALSO: where2, merge, merge2 allof, anyof, noneof, nallof, sort
 */

func where2(x)
/* DOCUMENT where2(x)
     like where(x), but the returned list is decomposed into indices
     according to the dimensions of x.  The returned list is always
     2 dimensional, with the second dimension the same as the dimension
     of where(x).  The first dimension has length corresponding to the
     number of dimensions of x.  Thus, where2([[0,1,3],[2,0,4]]) would
     return [[2,1],[3,1],[1,2],[3,2]].
     If noneof(x), where2 returns [] (i.e.- nil).
  SEE ALSO: where, merge, merge2, allof, anyof, noneof, nallof, sort
 */
{
  w= where(x);
  /* Since the result of where2 cannot be used as an index list, the
     case noneof(x) can be disposed of more easily than with where.  */
  if (!is_array(w)) return [];
  d= dimsof(x);
  n= d(1);
  if (!n) return w;  /* catcall for passing a scalar */
  d= d(2:);
  o= orgsof(x)(2:);
  w2= w(-:1:n,);
  w-= o(1);
  for (i=1 ; i<=n ; i++) {
    w2(i,)= w%d(i) + o(i);
    w/= d(i);
  }
  return w2;
}

extern merge;
/* DOCUMENT merge(true_expr, false_expr, condition)
     returns the values TRUE_EXPR or FALSE_EXPR where CONDITION is
     non-zero or zero, respectively.  The result has the data type of
     TRUE_EXPR or FALSE_EXPR, promoted to the higher arithmetic type
     if necessary.  The result has the dimensions of CONDITION.
     The number of elements in TRUE_EXPR must match the number of
     non-zero elements of CONDITION, and the number of elements in
     FALSE_EXPR must match the number of zero elements of CONDITION.
     (TRUE_EXPR or FALSE_EXPR should be nil if there are no such
     elements of CONDITION.  Normally, TRUE_EXPR and FALSE_EXPR should
     be 1-D arrays if they are not nil.)
     This function is intended for vectorizing a function whose
     domain is divided into two or more parts, as in:
        func f(x) {
          big= (x>=threshhold);
          wb= where(big);
          ws= where(!big);
          if (is_array(wb)) {
            xx= x(wb);
            fb= <function of xx>
          }
          if (is_array(ws)) {
            xx= x(ws);
            fs= <function of xx>
          }
          return merge(fb, fs, big);
        }
   SEE ALSO: mergef, merge2, where
 */

func merge2(t, f, c)
/* DOCUMENT merge2(true_expr, false_expr, condition)
     returns the values TRUE_EXPR or FALSE_EXPR where CONDITION is
     non-zero or zero, respectively.  The result has the data type of
     TRUE_EXPR or FALSE_EXPR, promoted to the higher arithmetic type
     if necessary.  Unlike the merge function, TRUE_EXPR and FALSE_EXPR
     must be conformable with each other, and with the CONDITION.
   SEE ALSO: merge, where, mergef
 */
{
  dims= dimsof(t, f, c);
  if (dims(1)) {
    c+= array(structof(c), dims);
    tt= array(structof(t), dims);  tt(..)= t;
    ff= array(structof(f), dims);  ff(..)= f;
  } else {
    tt= t;
    ff= f;
  }
  return merge(tt(where(c)), ff(where(!c)), c);
}

func mergei(__x, ..)
/* DOCUMENT y = mergei(x, f0, x1, f1, x2, ... xN, fN)
 *   Evaluate F1 where X<X1, F2 where X1<=X<X2, ..., FN where X>=XN,
 *   and merge all the results back into an array Y with the
 *   same dimensions as X.
 *
 *   During the evaluation of Fi, all of the local variables of
 *   the caller of mergei are available.  The Fi are called in
 *   order, skipping any for which no X is in the specified interval.
 *   Each Fi must return a value with the same dimensions as
 *   its input.
 *
 *   Additional input and output variables can be constructed using
 *   the mergel index list employed by mergei, and using the mergeg
 *   function.  For example, let w be an additional input to and z be
 *   an additional output from the function:
 *     func myfunc(x, w, &z) {
 *       z = array(0.0, dimsof(x, w));
 *       x += z;
 *       w += z;
 *       return mergei(x, _myfunc_lo, 1.234, _myfunc_hi);
 *     }
 *     func _myfunc_lo(x) {
 *       wp = w(mergel);  // part of w for this function
 *       z = mergeg(z, <expression using x and wp>);
 *       return <expression using x and wp>;
 *     }
 *     func _myfunc_hi(x) {
 *       wp = w(mergel);  // part of w for this function
 *       z = mergeg(z, <expression using x and wp>);
 *       return <expression using x and wp>;
 *     }
 *
 * SEE ALSO: mergef, merge
 */
{
  /**/ local __r;
  _1_ = !dimsof(__x)(1);
  __n = dimsof(__x)(1)? indgen(numberof(__x)) : 1;
  while (numberof(__x)) {
    __f = next_arg();
    __m = next_arg();
    __m = is_void(__m)? (__n>0) : (__x < __m);
    __l = where(__m);
    if (numberof(__l)) {
      __m = where(!__m);
      mergel = __n(__l);
      if (_1_) return __f(__x);
      __f = __f(__x(__l));
      if (is_void(__r)) __r = array(structof(__f), dimsof(__x));
      __r(mergel) = __f;
      __x = __x(__m);
      __n = __n(__m);
    }
  }
  return __r;
}

func mergef(__x, ..)
/* DOCUMENT y = mergef(x, f1, cond1, f2, cond2, ... felse)
 *   Evaluate F1(X(where(COND1))), F2(X(where(COND2))),
 *   and so on, until FELSE(X(where(!(COND1 | COND2 | ...))))
 *   and merge all the results back into an array Y with the
 *   same dimensions as X.  Each of the CONDi must have the
 *   same dimensions as X, and they must be mutally exclusive.
 *
 *   During the evaluation of Fi, all of the local variables of
 *   the caller of mergei are available.  The Fi are called in
 *   order, skipping any for which no X is in the specified interval.
 *   Each Fi must return a double value with the same dimensions as
 *   its input.
 *
 *   Additional input and output variables can be constructed using
 *   the mergel index list employed by mergei, and using the mergeg
 *   function.  For example, let w be an additional input to and z be
 *   an additional output from the function:
 *     func myfunc(x, w, &z) {
 *       z = array(0.0, dimsof(x, w));
 *       x += z;
 *       w += z;
 *       return mergef(x, _myfunc_lo, x<1.234, _myfunc_hi);
 *     }
 *     func _myfunc_lo(x) {
 *       wp = w(mergel);  // part of w for this function
 *       z = mergeg(z, <expression using x and wp>);
 *       return <expression using x and wp>;
 *     }
 *     func _myfunc_hi(x) {
 *       wp = w(mergel);  // part of w for this function
 *       z = mergeg(z, <expression using x and wp>);
 *       return <expression using x and wp>;
 *     }
 *
 * SEE ALSO: mergei, merge
 */
{
  /**/ local __r;
  _1_ = !dimsof(__x)(1);
  __n = dimsof(__x)(1)? indgen(numberof(__x)) : 1;
  while (numberof(__x)) {
    __f = next_arg();
    __m = next_arg();
    __m = is_void(__m)? (__n>0) : __m(__n);
    __l = where(__m);
    if (numberof(__l)) {
      __m = where(!__m);
      mergel = __n(__l);
      if (_1_) return __f(__x);
      __f = __f(__x(__l));
      if (is_void(__r)) __r = array(structof(__f), dimsof(__x));
      __r(mergel) = __f;
      __x = __x(__m);
      __n = __n(__m);
    }
  }
  return __r;
}

func mergeg(&z, value)
/* DOCUMENT z = mergeg(z, value)
 *       or z = mergeg(z)
 *   If secondary results are to be returned from a mergef, besides
 *   its return value, the Fi may construct them using mergeg.
 *     z = mergeg(z, value)
 *   where z is a variable in the original caller of mergef,
 *   and value is its value.
 *
 *   z = [];    or   z = <something with shape of x>;
 *   y = mergef(x, f1, cond, f2);
 *   z = mergeg(z);  // this can now be omitted, but does no harm
 *   ...
 *   func f1(x) { <exprz(x) computes z(x), expry(x) computes y(x)>
 *     z = mergeg(z, exprz(x));
 *     return expry(x);
 *   }
 *   func f2(x) { <exprz(x) computes z(x), expry(x) computes y(x)>
 *     z = mergeg(z, exprz(x));
 *     return expry(x);
 *   }
 *
 * SEE ALSO: mergef, merge
 */
{
  if (is_void(value)) {
    return z;  /* for backward compatibility */
  } else {
    /* void z case for backward compatibility */
    if (is_void(z)) z = array(structof(value), dimsof(__x));
    if (_1_) z = value;
    z(mergel) = value;
    return z;
  }
}

extern grow;
extern _;
/* DOCUMENT grow, x, xnext1, xnext2, ...
         or grow(x, xnext1, xnext2, ...)
         or    _(x, xnext1, xnext2, ...)
     lengthens the array X by appending XNEXT1, XNEXT2, etc. to its
     final dimension.  If X is nil, X is first redefined to the first
     non-nil XNEXT, and the remainder of the XNEXT list is processed
     normally.  Each XNEXT is considered to have the same number of
     dimensions as X, by appending unit-length dimensions if necessary.
     All but this final dimension of each XNEXT must be right-conformable
     (that is, conformable in the sense of the right hand side of an
     assignment statement) with all but the final dimension of X.
     The result has a final dimension which is the sum of the final
     dimension of X and all the final dimensions of the XNEXT.  Nil
     XNEXT are ignored.  The value of the result is obtained by
     concatenating all the XNEXT to X, after any required broadcasts.

     If invoked as a function, grow returns the new value of X; in
     this case, X may be an expression.  X must be a simple variable
     reference for the subroutine form of grow; otherwise there is
     nowhere to return the result.  The subroutine form is slightly
     more efficient than the function form for the common usage:
          x= grow(x, xnext1, xnext2)           is the same as
          grow, x, xnext1, xnext2              the preferred form

     The _ function is a synonym for grow, for people who want this
     operator to look like punctuation in their source code, on analogy
     with the array building operator [a, b, c, ...].

     The _cat function is sometimes more appropriate than grow.

     Usage note:
     Never do this:
       while (more_data) grow, result, datum;
     The time to complete this loop scales as the SQUARE of the number
     of passes!  Instead, do this:
       for (i=1,result=array(things,n_init) ; more_data ; i++) {
         if (i>numberof(result)) grow, result, result;
         result(i) = datum;
       }
       result = result(1:i-1);
     The time to complete this loop scales as n*log(n), because the
     grow operation doubles the length of the result each time.

   SEE ALSO: _cat, array
 */

extern indgen;
/* DOCUMENT indgen(n)
         or indgen(start:stop)
         or indgen(start:stop:step)
     returns "index generator" list -- an array of longs running from
     1 to N, inclusive.  In the second and third forms, the index
     values specified by the index range are returned.
   SEE ALSO: span, spanl, array
 */

extern span;
/* DOCUMENT span(start, stop, n)
         or span(start, stop, n, which)
     returns array of N doubles equally spaced from START to STOP.
     The START and STOP arguments may themselves be arrays, as long as
     they are conformable.  In this case, the result will have one
     dimension of length N in addition to dimsof(START, STOP).
     By default, the result will be N-by-dimsof(START, STOP), but
     if WHICH is specified, the new one of length N will be the
     WHICHth.  WHICH may be non-positive to position the new
     dimension relative to the end of dimsof(START, STOP); in
     particular WHICH of 0 produces a result with dimensions
     dimsof(START, STOP)-by-N.
   SEE ALSO: spanl, indgen, array
 */

func spanl(start, stop, n, which)
/* DOCUMENT spanl(start, stop, n)
         or spanl(start, stop, n, which)
     similar to the span function, but the result array have N points
     spaced at equal ratios from START to STOP (that is, equally
     spaced logarithmically).  See span for discussion of WHICH argument.
     START and STOP must have the same algebraic sign for this to make
     any sense.
   SEE ALSO: span, indgen, array
 */
{
  if (is_void(which)) which = 1;
  s = sign(start+0*stop);
  return exp(span(log(abs(start)), log(abs(stop)), n, which)) *
             span(s,s,n,which);
}
errs2caller, spanl;

extern digitize;
/* DOCUMENT digitize(x, bins)
     returns an array of longs with dimsof(X), and values i such that
     BINS(i-1) <= X < BINS(i) if BINS is monotonically increasing, or
     BINS(i-1) > X >= BINS(i) if BINS is monotonically decreasing.
     Beyond the bounds of BINS, returns either i=1 or i=numberof(BINS)+1
     as appropriate.
   SEE ALSO: histogram, interp, integ, sort, where, where2
 */

extern histogram;
/* DOCUMENT histogram(list)
         or histogram(list, weight)
     returns an array hist which counts the number of occurrences of each
     element of the input index LIST, which must consist of positive
     integers (1-origin index values into the result array):
          histogram(list)(i) = number of occurrences of i in LIST

     A second argument WEIGHT must have the same shape as LIST; the result
     will be the sum of WEIGHT:
          histogram(list)(i) = sum of all WEIGHT(j) where LIST(j)==i

     The result of the single argument call will be of type long; the
     result of the two argument call will be of type double (WEIGHT is
     promoted to that type).  The input argument(s) may have any number
     of dimensions; the result is always 1-D.

   KEYWORD: top=max_list_value
     By default, the length of the result is max(LIST).  You may
     specify that the result have a larger length by means of the TOP
     keyword.  (Elements beyond max(LIST) will be 0, of course.)

   SEE ALSO: digitize, sort, histinv
 */

func histinv(hist)
/* DOCUMENT list = histinv(hist)
     returns a list whose histogram is HIST, hist = histogram(list),
     that is, hist(1) 1's followed by hist(2) 2's, followed by hist(3)
     3's, and so on.  The total number of elements in the returned
     list is sum(hist).  All values in HIST must be non-negative;
     if sum(hist)==0, histinv returns [].  The input HIST array may
     have any number of dimensions; the result will always be either
     nil or a 1D array.

   SEE ALSO: histogram
 */
{
  if (anyof(hist < 0)) error, "histinv argument must be non-negative";
  hist = hist(*)(psum);
  return hist(0)? histogram(hist+1)(psum:1:-1) + 1 : [];
}

extern reshape;
/* DOCUMENT reshape, reference, address, type, dimension_list
         or reshape, reference, type, dimension_list
         or reshape, reference
     The REFERENCE must be an unadorned variable, not an expression;
     reshape sets this variable to an LValue at the specified ADDRESS
     with the specified TYPE and DIMENSION_LIST.  (See the array
     function documentation for acceptable DIMENSION_LIST formats.)
     If ADDRESS is an integer (e.g.- a long), the programmer is
     responsible for assuring that the data at ADDRESS is valid.
     If ADDRESS is a (Yorick) pointer, Yorick will assure that the
     data pointed to will not be discarded, and the reshape will
     fail if TYPE and DIMENSION_LIST extend beyond the pointee
     bounds.  In the second form, ADDRESS is taken to be &REFERENCE;
     that is, the TYPE and DIMENSION_LIST of the variable are changed
     without doing any type conversion.  In the third form, REFERENCE
     is set to nil ([]).  (Simple redefinition will not work on a
     variable defined using reshape.)
     WARNING: There are almost no situations for which reshape is
       the correct operation.  Use reform instead.
  SEE ALSO: reform, array, dimsof, numberof, is_array, eq_nocopy
 */

func reform(x, ..)
/* DOCUMENT reform(x, dimlist)
 *    returns array X reshaped according to dimension list DIMLIST.
 * SEE ALSO: array, dimsof, accum_dimlist
 */
{
  dims = [0];
  while (more_args()) accum_dimlist, dims, next_arg();
  if (dims(1)) {
    y = array(structof(x), dims);
    y(*) = x(*);   /* will blow up if lengths differ */
  } else {
    if (numberof(x)>1) error, "X longer than specified DIMLIST";
    y = x(1);
  }
  return y;
}

func accum_dimlist(&dims, d)
/* DOCUMENT accum_dimlist, dims, d
     accumulate a dimension argument D onto a dimension list DIMS.
     This can be used to emulate the dimension lists supplied to the
     array function.  For example:
       func myfunc(arg1, arg2, ..) {
         local dims;
         while (more_args()) accum_dimlist, dims, next_arg();
         ...
       }
   SEE ALSO: array, reform
 */
{
  if (is_void(d)) return;
  if (is_range(d)) {  /* yuck */
    mn = mx = 0;
    if (sread(print(d)(1),format="%ld:%ld",mn,mx) != 2)
      error, "only min:max ranges allowed in dimension list";
    d = mx - mn + 1;
  }
  if (!dimsof(d)(1)) d = [1, d];
  if (is_void(dims)) {
    dims = d;
  } else {
    grow, dims, d(2:1+d(1));
    dims(1) += d(1);
  }
}

extern interp;
/* DOCUMENT interp(y, x, xp)
         or interp(y, x, xp, which)
     returns yp such that (XP, yp) lies on the piecewise linear curve
     (X(i), Y(i)) (i=1, ..., numberof(X)).  Points beyond X(1) are set
     to Y(1); points beyond X(0) are set to Y(0).  The array X must be
     one dimensional, have numberof(X)>=2, and be either monotonically
     increasing or monotonically decreasing.  The array Y may have more
     than one dimension, but dimension WHICH must be the same length as
     X.  WHICH defaults to 1, the first dimension of Y.  WHICH may be
     non-positive to count dimensions from the end of Y; a WHICH of 0
     means the final dimension of Y.  The result yp has dimsof(XP)
     in place of the WHICH dimension of Y (if XP is scalar, the WHICH
     dimension is not present).  (The dimensions of the result are the
     same as if an index list with dimsof(XP) were placed in slot
     WHICH of Y.)
   SEE ALSO: integ, digitize, span
 */

extern integ;
/* DOCUMENT integ(y, x, xp)
         or integ(y, x, xp, which)
     See the interp function for an explanation of the meanings of the
     arguments.  The integ function returns ypi which is the integral
     of the piecewise linear curve (X(i), Y(i)) (i=1, ..., numberof(X))
     from X(1) to XP.  The curve (X, Y) is regarded as constant outside
     the bounds of X.  Note that X must be monotonically increasing or
   SEE ALSO: interp, digitize, span
 */

extern sort;
/* DOCUMENT sort(x)
         or sort(x, which)
     returns an array of longs with dimsof(X) containing index values
     such that X(sort(X)) is a monotonically increasing array.  X can
     contain integer, real, or string values.  If X has more than one
     dimension, WHICH determines the dimension to be sorted.  The
     default WHICH is 1, corresponding to the first dimension of X.
     WHICH can be non-positive to count dimensions from the end of X;
     in particular a WHICH of 0 will sort the final dimension of X.

     WARNING: The sort function is non-deterministic if some of the
              values of X are equal, because the Quick Sort algorithm
              involves a random selection of a partition element.

     For information on sorting with multiple keys (and on making
     sort deterministic), type the following:  help, msort

   SEE ALSO: median, digitize, interp, integ, histogram
 */

extern min;
/* DOCUMENT min(x)
         or min(x, y, z, ...)
     returns the scalar minimum value of its array argument, or, if
     more than one argument is supplied, returns an array of the
     minimum value for each array element among the several arguments.
     In the multi-argument case, the arguments must be conformable.
  SEE ALSO: max, sum, avg
 */

extern max;
/* DOCUMENT max(x)
         or max(x, y, z, ...)
     returns the scalar maximum value of its array argument, or, if
     more than one argument is supplied, returns an array of the
     maximum value for each array element among the several arguments.
     In the multi-argument case, the arguments must be conformable.
  SEE ALSO: min, sum, avg
 */

extern sum;
/* DOCUMENT sum(x)
     returns the scalar sum of all elements of its array argument.
     If X is a string, concatenates all elements.
  SEE ALSO: avg, min, max
 */

extern avg;
/* DOCUMENT avg(x)
     returns the scalar average of all elements of its array argument.
  SEE ALSO: sum, min, max
 */

func median(x, which)
/* DOCUMENT median(x)
         or median(x, which)
     returns the median of the array X.  The search for the median takes
     place along the dimension of X specified by WHICH.  WHICH defaults
     to 1, meaning the first index of X.  The median function returns an
     array with one fewer dimension than its argument X (the WHICH
     dimension of X is missing in the result), in exact analogy with
     rank reducing index range functions.  If dimsof(X)(WHICH) is
     odd, the result will have the same data type as X; if even, the
     result will be a float or a double, since the median is defined
     as the arithmetic mean between the two central values in that
     case.
   SEE ALSO: sort
 */
{
  if (is_void(which)) which= 1;
  list= sort(x, which);
  dims= dimsof(x);
  if (which<1) which= dims(1)-which;
  n= dims(1+which);
  odd= n%2;
  n/= 2;         /* index with half above, half below... */
  n+= 1;         /* ...corrected for 1-origin */
  stride= 1;
  for (i=1 ; i<which ; i++) stride*= dims(1+i);
  ldims= dims(1)-which+1;
  /**/ local l;
  reshape, l, &list, long, stride, grow(ldims, dims(1+which:));
  lm= l(,n,..);
  if (which<dims(1)) dims(1+which:-1)= dims(2+which:0);
  --dims(1);
  reshape, lm, long, dims;
  xm= x(lm);
  if (!odd) {     /* even length dimensions have more complicated median */
    reshape, lm;  /* undefine the LValue lm so following define works */
    lm= l(,n-1,..);
    reshape, lm, long, dims;
    xm= 0.5f*(xm+x(lm));
  }
  return xm;
}

extern transpose;
/* DOCUMENT transpose(x)
         or transpose(x, permutation1, permutation2, ...)
     transpose the first and last dimensions of array X.  In the second
     form, each PERMUTATION specifies a simple permutation of the
     dimensions of X.  These permutations are compounded left to right
     to determine the final permutation to be applied to the dimensions
     of X.  Each PERMUTATION is either an integer or a 1D array of
     integers.  A 1D array specifies a cyclic permutation of the
     dimensions as follows: [3, 5, 2] moves the 3rd dimension to the
     5th dimension, the 5th dimension to the 2nd dimension, and the 2nd
     dimension to the 3rd dimension.  Non-positive numbers count from the
     end of the dimension list of X, so that 0 is the final dimension,
     -1 in the next to last, etc.  A scalar PERMUTATION is a shorthand
     for a cyclic permutation of all of the dimensions of X.  The value
     of the scalar is the dimension to which the 1st dimension will move.

     Examples:  Let x have dimsof(x) equal [6, 1,2,3,4,5,6] in order
        to be able to easily identify a dimension by its length. Then:
        dimsof(x)                          == [6, 1,2,3,4,5,6]
        dimsof(transpose(x))               == [6, 6,2,3,4,5,1]
        dimsof(transpose(x,[1,2]))         == [6, 2,1,3,4,5,6]
        dimsof(transpose(x,[1,0]))         == [6, 6,2,3,4,5,1]
        dimsof(transpose(x,2))             == [6, 6,1,2,3,4,5]
        dimsof(transpose(x,0))             == [6, 2,3,4,5,6,1]
        dimsof(transpose(x,3))             == [6, 5,6,1,2,3,4]
        dimsof(transpose(x,[4,6,3],[2,5])) == [6, 1,5,6,3,2,4]
 */

/*= SECTION(string) string manipulation ====================================*/
/*
 * yorick string manipulation functions
 *
 * Inspired by the regexp package of Francois Rigaut 2004/Oct/6.
 * This API designed by David Munro January, 2005.
 *
 * compiled regular expression engine from Henry Spencer (Univ. of Toronto)
 *   see yregexp.c source
 *     ftp://ftp.zoo.toronto.edu/pub/regex.shar
 * compiled globbing engine from Guido van Rossum (for Univ. of California)
 *   see yfnmatch.c source
 *     http://www.freebsd.org/cgi/cvsweb.cgi/src/sys/libkern/fnmatch.c
 *     http://www.openbsd.org/cgi-bin/cvsweb/src/lib/libc/gen/fnmatch.c
 */

local string;
/* DOCUMENT string
 *
 * The yorick string datatype is a character string, e.g.- "Hello, world!".
 * Internally, strings are stored as 0-terminated sequences of characters,
 * which are 8-bit bytes, the same as the char datatype..
 *
 * Like numeric datatypes, string behaves as a function to convert objects
 * to the string datatype.  There are only two interesting conversions:
 *   string(0) is the nil string, like a 0 pointer
 *     This is the only string which is "false" in an if test.
 *   string(pc) where pc is an array of type pointer where each pointer
 *     is either 0 or points to an array of type char, copies the chars
 *     into an array of strings, adding a trailing '\0' if necessary
 *   pointer(sa) where sa is an array of stringa is the inverse
 *     conversion, copying each string to an array of char (including the
 *     terminal '\0') and returning an array of pointers to them
 * The strchar() function may be a more convenient way to convert from
 * string to char and back.
 *
 * Yorick provides the following means of manipulating string variables:
 *
 * s+t         when s and t are strings, + means concatentation
 *             (this is not perfect nomenclature, since t+s != s+t)
 * s(,sum,..)  the sum index range concatentates along a dimension of
 *             an array of strings
 * sum(s)      concatenates all the strings in an array (in storage order)
 *
 * strlen(s)          returns length(s) of string(s) s
 * strcase(upper, s)  converts s to upper or lower case
 * strchar(s_or_c)    converts between string and char arrays
 *                    (quick and dirty alternative to string<->pointer)
 * strpart(s, m:n)
 * strpart(s, sel)    extracts substrings (sel is a [start,end] list)
 *   string search functions:
 * strglob(pat, s)    shell-like wildcard pattern match, returns 0 or 1
 * strword(s, delim)  parses s into word(s), returns a sel
 * strfind(pat, s)    simple pattern match, returns a sel
 * strgrep(pat, s)    regular expression pattern match, returns a sel
 * streplace(s, sel, t)  replaces sel in s by t
 *
 * strtrim trims leading and/or trailing blanks (based on strword)
 * strmatch is a wrapper for strfind that simply returns whether there
 *   was a match or not rather than its exact offset
 * strtok is a variant of strword that calls strpart in order to
 *   return the substrings rather than an sel index list
 *
 * The strword, strfind, and strgrep functions produce a sel, that is,
 * a list of [start,end] offsets into an array of strings.
 * These sel indicate portions of a string to be operated on for the
 * strpart and streplace functions.
 *
 * The sread, swrite, and print functions operate on or produce strings.
 * The rdline, rdfile, read, and write functions perform I/O on strings
 * to text files.
 */
{}
/* (previous line is kludge to halt help,string interactive printout) */

extern strlen;
/* DOCUMENT strlen(string_array)
     returns an long array with dimsof(STRING_ARRAY) containing the
     lengths of the strings.  Both string(0) and "" have length 0.
   SEE ALSO: string, strchar, strcase, strpart, strfind, strword
 */

extern strchar;
/* DOCUMENT strchar(string_array)
         or strchar(char_array)
     converts STRING_ARRAY to an array of characters, or CHAR_ARRAY
     to an array of strings.  The return value is always a 1D array,
     except in the second form if CHAR_ARRAY contains only a single
     string, the result will be a scalar string.  Each string is
     stored in sequence including its trailing '\0' character, with
     any string(0) elements treated as if they were "".  Going in
     the opposite direction, a '\0' before any non-'\0' characters
     produces string(0), so that "" can never be an element of
     the result, and if the final char (of the leading dimension)
     is not '\0', an implicit '\0' is assumed beyond the end of the
     input char array.  For example,
        strchar(["a","b","c"]) --> ['a','\0','b','\0','c','\0']
        strchar([['a','\0','b'],['c','\0','\0']]) --> ["a","b","c",string(0)]
     The string and pointer data types themselves also convert between
     string and char data, avoiding the quirks of strchar.
   SEE ALSO: string, strpart, strword, strfind
 */

extern strpart;
/* DOCUMENT strpart(string_array, m:n)
         or strpart(string_array, start_end)
         or strpart, string_array, start_end
    returns another string array with the same dimensions as
    STRING_ARRAY which consists of characters M through N of
    the original strings.  M and N are 1-origin indices; if
    M is omitted, the default is 1; if N is omitted, the default
    is the end of the string.  If M or N is non-positive, it is
    interpreted as an index relative to the end of the string,
    with 0 being the last character, -1 next to last, etc.
    Finally, the returned string will be shorter than N-M+1
    characters if the original doesn't have an Mth or Nth
    character, with "" (note that this is otherwise impossible)
    if neither an Mth nor an Nth character exists.  A 0
    is returned for any string which was 0 on input.

    In the second form, START_END is an array of [start,end] indices.
    A single pair [start,end] is equivalent to the range start+1:end,
    that is, start is the index of the character immediately before
    the substring (which is to say start is the number of characters
    skipped at the beginning of the string).  If end<start, or if
    either start or end are <0 or >length, or if the original string
    is string(0), strpart returns string(0); otherwise, if end==start,
    strpart returns "".

    However, the START_END array may have any additional dimensions
    (beyond the leading dimension of length 2) which are conformable
    with the dimensions of the STRING_ARRAY.  The result will be a
    string array with dimensions dimsof(STRING_ARRAY,START_END(1,..)).
    Furthermore, the leading dimension of START_END may have any
    even length, say 2*n, in which case the leading dimension of
    the result will be n.  For example,
      strpart(a, [s1,e1,s2,e2,s3,e3,s4,e4])
    is equivalent to (or shorthand for)
      strpart(a(-,..), [[s1,e1],[s2,e2],[s3,e3],[s4,e4]])(1,..)

    In the third form, called a subroutine, strpart operates on
    STRING_ARRAY in place.  In this case START_END must have
    leading dimension of length 2, although it may have trailing
    dimensions as usual.

    Examples:
    strpart("Hello, world!", 4:6) --> "lo,"
    strpart("Hello, world!", [3,6]) --> "lo,"
      -it may help to think of [start,end] as the 0-origin offset
       of a "cursor" between the characters of the string
    strpart("Hello, world!", [3,3]) --> ""
    strpart("Hello, world!", [3,2]) --> string(0)
    strpart("Hello, world!", [3,20]) --> string(0)
    strpart("Hello, world!", [3,6,7,9]) --> ["lo,","wo"]
    strpart(["one","two"], [[1,2],[0,1]]) --> ["n","t"]
    strpart(["one","two"], [1,2,0,1]) --> [["n","o"],["w","t"]]

   SEE ALSO: string, strcase, strlen, strfind, strword
 */

extern strcase;
/* DOCUMENT strcase(upper, string_array)
         or strcase, upper, string_array
     returns STRING_ARRAY with all strings converted to upper case
     if UPPER is non-zero.  If UPPER is zero, result is lower case.
     (For characters >=0x80, the case conversion assumes the ISO8859-1
      character set.)
     Called as a subroutine, strcase converts STRING_ARRAY in place.
   SEE ALSO: string, strlen, strpart, strglob, strfind, strgrep, strword
 */

extern strword;
/* DOCUMENT strword(string_array)
         or strword(string_array, delim)
         or strword(string_array, delim, n)
         or strword(string_array, off, delim, n)
     scans to the first character in STRING_ARRAY which is not in
     the DELIM list.  DELIM defaults to " \t\n", that is, whitespace.
     The return value is a [start,end] offset pair, with trailing
     dimensions matching the dimensions of the given STRING_ARRAY.
     Note that this return value is suitable for use in the strpart
     or streplace functions.

     If the first character of DELIM is "^", the sense is reversed;
     strword scans to the first character in DELIM.  (Except that
     if DELIM is the single character "^", it has its usual meaning.)
     Also, a "-" which is not the first (or second after "^") or last
     character of DELIM indicates a range of characters.  Finally,
     if DELIM is "" or string(0), the scan stops immediately, since
     the first character (no matter what it is) is not in DELIM.

     Furthermore, DELIM can be a list of delimiter sets, where each
     element of the list delimits a new word, so the return value will
     be [start1,end1, ..., startN,endN], where N=numberof(DELIM),
     and start1 is the offset of the first character not in DELIM(1),
     characters with offset between end1 and start2 are in DELIM(2),
     characters with offset between end2 and start3 are in DELIM(3),
     and so on.  If endM is the length of the string for some M<N,
     then all subsequent start,end pairs will be endM,-1.  Note that
     endN is always the length of the string.

     If N is specified, the final DELIM will be repeated N times, so
     the leading dimension of the result is 2*(numberof(DELIM)+N-1).
     Thus, instead of DELIM = [d1,d2,d3,d3,d3,d3,d3], you may specify
     DELIM = [d1,d2,d3] and N = 5.  (By default, N=1.)

     If OFF is supplied (it is distinguished from DELIM by the
     fact that DELIM is a string or [], while OFF is an integer),
     it represents the offset (0-origin character index) in
     STRING_ARRAY at which the word search is to begin.  (DELIM
     or N may be omitted to get their default values.)  The OFF
     may be a scalar or an array conformable with STRING_ARRAY.
     The returned offsets include OFF; you do not need to add OFF
     to the return value to get the offsets into STRING_ARRAY.

     The action of strword can be described as follows: If d1 is
     the first delimiter, d2 is the second delimiter, and so on,
     then the input element of STRING_ARRAY looks like this:
          d1 word1 d2 word2 d3 word3 d4 word4 ...
     The offset array returned by strword will select word1, word2,
     word3, etc., if used in strpart.  One defect of this system is
     that the final word has no trailing delimiter - it always
     extends to the end of the string.

     To remedy this defect, N<0 is interpreted as follows: It is
     the same as abs(N), except that the final [start,end] pair is
     supressed; that is, if you have specified N delimiters, only
     N-1 words will be returned.

     As a second special case, N=0 forces the final delimeter to be
     at the end of the string, rather than as soon after the previous
     endM as possible.  Like N<0, this reduces the number of start,end
     pairs by one.  (N=0 is useful, for example, in order to trim
     trailing blanks.)

     Examples:
     strword("  Hello, world!") --> [2,15]
     strword("Hello, world!") --> [0,13]
     strword("Hello, world!", , 2) --> [0,6,7,13]
     strword("Hello, world!", , -2) --> [0,6]
     strword("Hello, world!", ".!, \t\n", -2) --> [0,5]
     strword("Hello, world!", [string(0), ".!, \t\n"], 0) --> [0,12]
     strword("Hello, world!", "A-Za-z", 2) --> [5,7,12,13]
     strword("Hello, world!", "^A-Za-z", 2) --> [0,5,7,13]
     strword("Hello, world!", "^A-Za-z", 3) --> [0,5,7,12,13,-1]
     strword("  Hello, world!", [" \t\n",".!, \t\n"]) --> [2,7,9,15]
     strword("  Hello, world!", [" \t\n",".!, \t\n"], 2) --> [2,7,9,14,15,-1]

   SEE ALSO: string, strlen, strpart, strfind, strtok, strtrim
 */

func strtrim(s, which, blank=)
/* DOCUMENT strtrim(string_array)
         or strtrim(string_array, which)
         or strtrim, string_array, which
     returns STRING without leading and/or trailing blanks.  WHICH=1
     means to trim leading blanks only, WHICH=2 trims trailing blanks
     only, while WHICH=3 (the default) trims both leading and trailing
     blanks.  Called as a subroutine, strtrim performs this operation
     in place.
     The blank= keyword, if present, is a list of characters to be
     considered "blanks".  Use blank=[lead_delim,trail_delim] to get
     different leading and trailing "blanks" definitions.  By default,
     blank=" \t\n".  (See strword for more about delim syntax.)
   SEE ALSO: string, strpart, strword
 */
{
  which = is_void(which)? 3 : (which&3);
  if (!which) return s;
  if (which == 1) {
    if (numberof(blank)>1) blank = blank(1);
    b = strword(s, blank);
  } else {
    if (which == 2) {
      if (!numberof(blank)) blank = [string(0), " \t\n"];
      else blank = [string(0), blank(numberof(blank))];
    } else {
      if (!numberof(blank)) blank = [" \t\n", " \t\n"];
      else blank = [blank(1), blank(numberof(blank))];
    }
    b = strword(s, blank, 0);
  }
  if (am_subroutine()) strpart, s, b;
  else return strpart(s, b);
}

func strtok(s, delim, n)
/* DOCUMENT strtok(string_array, delim)
         or strtok(string_array)
         or strtok(string_array, delim, n)
     strips the first token off of each string in STRING_ARRAY.
     A token is delimited by any of the characters in the string
     DELIM.  If DELIM is blank, nil, or not given, the
     default DELIM is " \t\n" (blanks, tabs, or newlines).
     The result is a string array ts with dimensions
     2-by-dimsof(STRING_ARRAY); ts(1,) is the first token, and
     ts(2,) is the remainder of the string (the character which
     terminated the first token will be in neither of these parts).
     The ts(2,) part will be 0 (i.e.- the null string) if no more
     characters remain after ts(1,); the ts(1,) part will be 0 if
     no token was present.  A STRING_ARRAY element may be 0, in
     which case (0, 0) is returned for that element.

     With yorick-1.6, strtok has been extended to accept multiple
     delimiter sets DELIM for successive words, and a repeat count
     N for the final DELIM set.  The operation is the same as for
     strword, except that the N<=0 special cases are illegal, and
     if DELIM consists of only a single set, N=2 is the default
     rather than N=1.  The dimensions of the return value are thus
     min(2,numberof(DELIM)+N-1)-by-dimsof(STRING_ARRAY).

   SEE ALSO: string, strword, strmatch, strpart, strlen
 */
{
  return strpart(s, (is_void(n)?_strtok(s, delim):_strtok(s, delim, n)));
}

extern _strtok;  /* worker for strtok, a variant on strword */

extern strglob;
/* DOCUMENT strglob(pat, string_array)
         or strglob(pat, string_array, off)
     test if pattern PAT matches STRING_ARRAY.  Optional OFF is an integer
     array conformable with STRING_ARRAY or 0-origin offset(s) within
     the string(s) at which to begin the search(es).  The return value
     is an int with the same dimensions as STRING_ARRAY, 1 for a match,
     and 0 for no match.

     PAT can contain UNIX shell wildcard or "globbing" characters:
     *   matches any number of characters
     ?   matches any single character
     [abcd]  matches any single character in the list, which may
             contain ranges such as [a-z0-9A-Z]
     \c  matches the character c (useful for c= a special character)
         (note that this is "\\c" in a yorick string)

     The strglob function is mostly intended for matching lists of
     file names.  Note, in particular, that unlike strfind or strgrep,
     the entire string must match PAT.

     Keywords:
     case=  (default 1) zero for case-insensitive search
     path=  (default 0) 1 bit set means / must be matched by /
                        2 bit set means leading . must be matched by .
     esc=   (default 1) zero means \ is not treated as an escape

     The underlying compiled routine is based on the BSD fnmatch
     function, contributed by Guido van Rossum.

     Examples:
     return all files in current directory with .pdb extension:
       d=lsdir("."); d(where(strglob("*.pdb", d)));
     return all subdirectories of the form "hackNN", case insensitive:
       d=lsdir(".",1);
       d(where(strglob("hack[0-9][0-9]", d, case=0)));

   SEE ALSO: string, strfind, strgrep, strword, strpart, streplace
 */

extern strfind;
/* DOCUMENT strfind(pat, string_array)
         or strfind(pat, string_array, off)
     finds pattern PAT in STRING_ARRAY.  Optional OFF is an integer
     array conformable with STRING_ARRAY or 0-origin offset(s) within
     the string(s) at which to begin the search(es).  The return value
     is a [start,end] offset pair specifying the beginning and end
     of the first match, or [len,-1] if none, with trailing dimensions
     the same as the dimensions of STRING_ARRAY.  This return value
     is suitable as an input to the strpart or streplace functions.

     The strfind function is the simpler string pattern matcher:
     strfind - just finds a literal pattern (possibly case insensitive)
     strgrep - matches a pattern containing complex regular expressions
     Additionally, the strglob function does filename wildcard matching.

     Keywords:
     n=  (default 1) returns list of first n matches, so leading
         dimension of result will be 2*n
     case=  (default 1) zero for case-insensitive search
     back=  (default 0) non-zero for backwards search
             If back!=0 and n>1, the last match is listed as the
             last start-end pair, so the output pairs still appear
             in increasing order, and the first few may be 0,-1
             to indicate no match.

     Examples:
     s = ["one two three", "four five six"]
     strfind("o",s)  -->  [[0,1], [1,2]]
     strfind(" t",s)  -->   [[3,5], [13,-1]]
     strfind(" t",s,n=2)  -->   [[3,5,7,9], [13,-1,13,-1]]
     strfind("e",s,n=2,back=1)  -->   [[11,12,12,13], [0,-1,8,9]]

     SEE ALSO: string, strglob, strgrep, strword, strpart, streplace
 */

extern strgrep;
/* DOCUMENT strgrep(pat, string_array)
         or strgrep(pat, string_array, off)
    finds pattern PAT in STRING_ARRAY.  Optional OFF is an integer
    array conformable with STRING_ARRAY or 0-origin offset(s) within
    the string(s) at which to begin the search(es).  The return value
    is a [start,end] offset pair specifying the beginning and end
    of the first match, or [len,-1] if none, with trailing dimensions
    the same as the dimensions of STRING_ARRAY.  This return value
    is suitable as an input to the strpart or streplace functions.

    The underlying compiled routine is based on the regexp package
    written by Henry Spencer (copyright University of Toronto 1986),
    slightly modified for yorick.

    PAT is a regular expression, simliar to the UNIX grep utility.
    Every "regular expression" syntax is slightly different; here is
    the syntax supported by strgrep:

    The following characters in PAT have special meanings:

    '[' followed by any sequence of characters followed by ']' is a
        "range", which matches any single one of those characters
        '^' first means to match any character NOT one in the sequence
        '-' in such a sequence indicates a range of characters
          (e.g.- "[A-Za-z0-9_]" matches any alphanumeric character
           or underscore, while "[^A-Za-z0-9_]" matches anything else)
        to include ']' in the sequence, place it first,
        to include '-' in the sequence, place it first or last
          (or first after a leading '^' in either case)
        Note that the following special characters lose their special
        meanings inside a range.
    '.' matches any single character
    '^' matches the beginning of the string (but no characters)
    '$' matches the end of the string (but no characters)
        (that is, ^ and $ serve to anchor a search so that it will
         only find a match at the beginning or end of the string)
    '\' (that is, a single backslash, which can only be entered
         into a yorick string by a double backslash "\\")
        followed by any single character eliminates any special
        meaning for that character, for example "\\." matches
        period, rather than any single character (its special meaning)
    '(' followed by a regular expression followed by ')' matches the
        regular expression, creating a sub-pattern, which is a type
        of atom (see below)
    '|' means "or"; it separates branches in a regular expression
    '*' after an atom matches 0 or more matches of the atom
    '+' after an atom matches 1 or more matches of the atom
    '?' after an atom matches 0 or 1 matches of the atom

    The definitions of "atom", "branch", and "regular expression" are:

    A "regular expression" (which is what PAT is) consists of zero
    or more "branches" separated by '|'; it matches anything that
    matches one of the branches.

    A "branch" consists of zero or more "pieces", concatenated; it
    matches a match for the first followed by a match for the second,
    etc.

    A "piece" is an "atom", optionally followed by '*', '+', or '?';
    it matches the atom, or zero or more repetitions of the atom, as
    specified by the optional suffix.

    Finally, an "atom" is an ordinary single character, or a
    '\'-escaped single character (matching that character), or
    one of the special characters '.', '^', or '$', or a
    []-delimited range (matching any single character in the range),
    or a sub-pattern enclosed in () (matching the sub-pattern).

    A maximum of nine sub-patterns is allowed in PAT; these are
    numbered 1 through 9, in order of their opening '(' in PAT.

    This recursive definition of regular expressions often leads to
    ambiguities, both subtle and glaring.  Here is Henry Spencer's
    synopsis of how his routines behave:

    -------------------------------------------------------------------
    If a regular expression could match two different parts of the
    input string, it will match the one which begins earliest.  If both
    begin in the same place but match different lengths, or match the
    same length in different ways, life gets messier, as follows.

    In general, the possibilities in a list of branches are considered
    in left-to-right order, the possibilities for `*', `+', and `?' are
    considered longest-first, nested constructs are considered from the
    outermost in, and concatenated constructs are considered leftmost-
    first.  The match that will be chosen is the one that uses the
    earliest possibility in the first choice that has to be made.  If
    there is more than one choice, the next will be made in the same
    manner (earliest possibility) subject to the decision on the first
    choice.  And so forth.

    For example, `(ab|a)b*c' could match `abc' in one of two ways. The
    first choice is between `ab' and `a'; since `ab' is earlier, and
    does lead to a successful overall match, it is chosen. Since the
    `b' is already spoken for, the `b*' must match its last possibility
    -the empty string- since it must respect the earlier choice.

    In the particular case where no `|'s are present and there is only
    one `*', `+', or `?', the net effect is that the longest possible
    match will be chosen.  So `ab*', presented with `xabbbby', will
    match `abbbb'.  Note that if `ab*' is tried against `xabyabbbz', it
    will match `ab' just after `x', due to the begins-earliest rule.
    (In effect, the decision on where to start the match is the first
    choice to be made, hence subsequent choices must respect it even if
    this leads them to less-preferred alternatives.)
    -------------------------------------------------------------------

    When PAT contains parenthesized sub-patterns, strgrep returns
    the [start,end] of the entire match by default, but you can
    also get the [start,end] of any or all of the sub-patterns
    using the sub= keyword (see below).

    If PAT does not contain any regular expression constructs, you
    should use the strfind function instead of strgrep.  The strglob
    function, if appropriate, will also be faster than strgrep.

    Keywords:
    n=  (default 1) returns list of first n matches, so leading
        dimension of result will be 2*n
    sub=[n1,n2,...] is a list of the sub-pattern [start,end] pairs
        to be returned.  Thus 0 is the whole PAT, 1 is the first
        parenthesized sub-pattern, and so on.  The leading
        dimension of the result will be 2*numberof(sub)*n.  The
        sequence n1,n2,... must strictly increase: n1<n2<...
        The default is sub=0, which returns only the whole match.

    Examples:
    s = "Hello, world!"
    strgrep("(Hello|Goodbye), *([a-zA-Z]+)!", s)
      --> [0,13]
    strgrep("(Hello|Goodbye), *([a-z]*|[A-Z]*)!", s, sub=[1,2])
      --> [0,5,7,12]
    strgrep("(Hello|Goodbye), *([a-z]*|[A-Z]*)!", s, sub=[0,2])
      --> [0,13,7,12]
    strgrep("(Hello|Goodbye), *(([A-Z]*)|([a-z]*))!", s, sub=[0,2,3,4])
      --> [0,13,7,12,13,-1,7,12]

   SEE ALSO: string, strglob, strfind, strword, strpart, streplace,
             strgrepm
 */

func strgrepm(pat, x, off, case=)
/* DOCUMENT strgrepm(pat, string_array)
 *       or strgrepm(pat, string_array, off)
 *   call strgrep, but simply return mask of same dimensions as STRING_ARRAY
 *   set to 1 where it matches the PAT, and 0 where it does not match.
 *   The strgrepm function does not accept any of the strgrep keywords,
 *   but it does accept the strglob case= keyword to indicate a case
 *   insensitive search.
 * SEE ALSO: strglob, strgrep
 */
{
  if (case) {
    pat = strcase(0, pat);
    x = strcase(0, x);
  }
  return strgrep(pat, x, off)(2,..) >= 0;
}

about_glob = strgrepm;

func strmatch(s, pat, case)
/* DOCUMENT strmatch(string_array, pattern)
         or strmatch(string_array, pattern, case_fold)
         or strmatch(string_array, pattern, case_fold)
     returns an int array with dimsof(STRING_ARRAY) with 0 where
     PATTERN was not found in STRING_ARRAY and 1 where it was found.
     If CASE_FOLD is specified and non-0, the pattern match is
     insensitive to case, that is, an upper case letter will match
     the same lower case letter and vice-versa.
     (Consider using strfind directly.)
   SEE ALSO: string, strfind, strpart, strlen
 */
{
  if (is_void(case)) i = strfind(pat, s);
  else i = strfind(pat, s, case=(case==0));
  return i(2,..) >= 0;
}

extern streplace;
/* DOCUMENT streplace(string_array, start_end, to_string)
     replaces the part(s) START_END of STRING_ARRAY by TO_STRING.
     The leading dimension of START_END must be a multiple of 2,
     while any trailing dimensions must be conformable with the
     dimensions of STRING_ARRAY.  The TO_STRING must be conformable
     with STRING_ARRAY if the leading dimension of START_END is 2.
     An element of START_END may represent "no match" (for example,
     when end<start), in which case streplace silently skips the
     replacement.  Note that when start==end, streplace functions
     as text insertion.

     However, if the leading dimension of START_END = 2*n > 2, then
     TO_STRING must have a leading dimension conformable with n
     (that is, of length either 1 or n).  In this case, streplace
     performs multiple replacements within each string.  In order
     for multiple replacements to be meaningful, the START_END
     must be disjoint and sorted, as returned by strfind or
     strgrep with a repeat count, or by strword.  In other words,
     the first dimension of START_END should be non-decreasing,
     except where end<start indicates no replacement.

     Invoked as a subroutine, streplace operates on STRING_ARRAY
     in place.

     Examples:
     s = "Hello, world!"
     streplace(s,[0,5], "Goodbye")
       -->  "Goodbye, world!"
     streplace(s,[0,5,7,7], ["Goodbye","cruel "])
       -->  "Goodbye, cruel world!"
     streplace(s,[0,5,7,7,12,13], ["Goodbye","cruel ","?"])
       -->  "Goodbye, cruel world?"
     streplace(s,[0,5,0,-1,12,13], ["Goodbye","cruel ","?"])
       -->  "Goodbye, world?"
     streplace([s,s],[0,5], ["Goodbye", "Good bye"])
       -->  ["Goodbye, world!", "Good bye, world!"]
     streplace([s,s],[0,5,7,7], [["Goodbye","cruel "], ["Good bye",""]])
       -->  ["Goodbye, cruel world!", "Good bye, world!"]

     SEE ALSO: string, strfind, strgrep, strword, strpart
 */

/*= SECTION(fileio) generic file i/o =======================================*/

extern open;
/* DOCUMENT f= open(filename)
         or f= open(filename, filemode)
         or f= open(filename, filemode, errmode)
     opens the file FILENAME according to FILEMODE (both are strings).
     If ERRMODE is non-nil and non-zero, fail by returning nil F,
     otherwise failure to open or create the file is a runtime error.

     To use ERRMODE to check for the existence of a file:
        if (open(filename,"r",1)) file_exists;
        else file_does_not_exist;

     The return value F is an IOStream (or just stream for short).  When
     the last reference to this return value is discarded, the file will
     be closed.  The file can also be explicitly closed with the close
     function.  The FILEMODE determines whether the file is to be
     opened in read, write, or update mode, and whether writes are
     restricted to the end-of-file (append mode).  FILEMODE also
     determines whether the file is opened as a text file or as a
     binary file.  FILEMODE can have the following values, which are
     the same as for the ANSI standard fopen function:
        "r"     - read only
        "w"     - write only, random access, existing file overwritten
        "a"     - write only, forced to end-of-file,
                existing file preserved
        "r+"    - read/write, random access, existing file preserved
        "w+"    - read/write, random access, existing file overwritten
        "a+"    - read/write, reads random access,
                writes forced to end-of-file, existing file preserved
        "rb"  "wb"  "ab"  "r+b"  "rb+"  "w+b"  "wb+"  "a+b"  "ab+"
                without b means text file, with b means binary file
     The default FILEMODE is "r" -- open an existing text file for
     reading.

     The read and write functions perform I/O on text files.
     I/O to binary files may be performed explicitly using the save
     and restore functions, or implicitly by using the stream variable
     F as if it were a data structure instance (e.g.- f.x refers to
     variable x in the binary file f).
  SEE ALSO: create, close, read, write, rdline, bookmark, backup, popen
            vopen, rename, remove, save, restore
 */

extern vopen;
/* DOCUMENT f = vopen(source)
         or f = vopen(source, 1)
     opens SOURCE, which can be a string or char array, as if it were a
     file, returning a file handle.  The file handle will be a text file
     unless the optional second argument is non-nil and non-zero, as in
     the second form.  For the case of a binary file, SOURCE must be a
     char array.  Any dimensions of a char array are ignored in either
     case.  For a text file, if SOURCE is a string array, each array element
     is treated as one line of text.  For a text file char array, "\n",
     "\r", "\r\n", or "\0" are all recognized as newline markers.  These
     are read only files.

     If SOURCE is nil, the file handle will be read-write.  After writing
     the in-memory file, you can retrieve the finished array with the
     vclose function.  If the file is text, the array will be an array of
     strings, one per line.  If the file is binary, the array will be an
     array of char.

  SEE ALSO: vsave, vclose, open, system
 */

extern vclose;
/* DOCUMENT contents = vclose(handle)
     closes a file handle opened with vopen, returning the contents as
     an array.  As a side effect, the handle is set to nil [], as in
     close.  For a read-only handle, the contents will be the same as
     the array passed to the vopen call which returned the handle.
     For a read-write handle, vclose is the only way to get back what
     you have written to the file; if you close such a file using the
     ordinary close function, you will lose what you have written.
  SEE ALSO: vsave, vopen, close
 */

func vsave(args)
/* DOCUMENT c = vsave(var1, var2, ...);
         or c = vsave(var1, ..., string(namea), vara, ...);
         or vfile = createb(char);
            vsave, vfile, var1, var2, ...;
            vsave, vfile, var3, var4, ...;
            ...
            c = vsave(vfile);
     save the array variables VAR1, VAR2, ..., in the char array that is
     returned.  Any of the variables may instead be a string expression
     NAMEA followed by the value VARA of the variable.  The NAMEA argument
     is recognized as the name of the following argument by being an
     expression; arguments that are to be stored in f must be simple
     variable references.  You can achieve this as shown by placing the
     argument inside a call to string(), or by adding "", or simply by
     passing a constant string value like "myvarname".

     If you wish to build up a char array over several calls to vsave,
     pass the first argument VFILE, which you create with createb(char).
     A final call with no variables returns the char array and closes
     VFILE.

     You can pass the returned char array to openb, f=openb(c), to get
     an in-memory file handle f like any other binary file handle,
     allowing you to use the restore function, or the f.var1 syntax,
     or the get_member function.

     You can set the internal primitives using the prims= keyword; see
     createb for details.
  SEE ALSO: openb, vopen, vpack, restore, get_member, wrap_args
 */
{
  /**/ local a, name;
  prims = args(-);
  if (numberof(prims)) {
    if (numberof(prims)!=1 || prims(1)!="prims")
      error, "unrecognized keyword argument";
    eq_nocopy, prims, args(-1);
  }
  eq_nocopy, a, args(1);
  handle = is_stream(a);
  if (handle) f = a;
  else f = createb(char, prims);
  n = args(0);
  if (handle && n==1) {
    a = vclose(f);
    args, 1, f;
    return a;
  }
  handle = (handle || is_range(a));
  for (i=1+handle ; i<=n ; ++i) {
    eq_nocopy, a, args(i);
    name = args(-,i);
    if (!name) {
      eq_nocopy, name, a;
      eq_nocopy, a, args(++i);
    }
    add_variable, f, -1, name, structof(a), dimsof(a);
    get_member(f, name) = a;
  }
  return handle? f : vclose(f);
}
wrap_args, vsave;

func vpack(args)
/* DOCUMENT bytes = vpack(var1, var2, ...);
         or vfile = vopen(,1);
            vpack, vfile, var1, var2, ...;
            vpack, vfile, var3, var4, ...;
            ...
            bytes = vpack(vfile);
     pack variables into a byte stream, preserving data types and dimensions.
     If the first argument is an in-memory file created by vopen(,1), then
     vpack appends the variables to the file; to close the file, supply
     no new variables to pack.  The VARi must be arrays, and may not be
     pointers or struct instances.  If you want to store pointers or struct
     instances and preserve variable names, use vsave.
     The returned byte stream contains the primitive data formats (as
     returned by get_primitives), so it can be used on a platform other
     than the one on which vpack was run.
  SEE ALSO: vunpack, vsave
 */
{
  if (numberof(args(-)))
    error, "unexpected keyword argument";
  narg = args(0);
  iarg = 1;
  var = args(iarg);
  if (is_stream(var)) {
    f = var;
    if (narg == 1) {
      var = vclose(f);
      args, 1, f;
      return var;
    }
    var = args(++iarg);
    handle = 1;
  } else if (is_range(var)) {
    f = vopen(,1);
    var = args(++iarg);
    handle = 1;
  } else {
    f = vopen(,1);
  }

  addr = sizeof(f);
  prims = get_primitives(f);
  sz = prims(1:22:3);
  sz(7:8) = [2*sz(6), 1];
  if (!addr) {
    p = array(char, 4, 32);
    for (i=0 ; i<4 ; ++i) p(i+1,) = (prims >> (8*i)) & 0xff;
    _write, f, 0, p;     /* first 128 bytes are primitive formats */
    p = array(char, 8);  /* leave 8 bytes of scratch space at byte 128 */
    _write, f, 128, p;
    addr = 136;
    save, f, complex;
  }

  for (;;) {
    s = structof(var);
    d = dimsof(var);
    if (is_void(s)) s = 0;
    else if (s == char) s = 1;
    else if (s == short) s = 2;
    else if (s == int) s = 3;
    else if (s == long) s = 4;
    else if (s == float) s = 5;
    else if (s == double) s = 6;
    else if (s == complex) s = 7;
    else if (s == string) s = 8;
    else error, "vpack supports only string and numeric arrays";

    _write, f, addr, s+16*numberof(d);
    addr += sz(4);
    if (s) {
      _write, f, addr, d;
      addr += sz(4)*numberof(d);
      if (s == 8) {
        list = where(var == "");
        if (numberof(list)) {  /* strchar does not handle "" properly */
          var = grow(array(string, dimsof(var)), var);
          var(list) = ".";
        }
        var = strchar(var);
        _write, f, addr, numberof(var);
        addr += sz(4);
      }
      _write, f, addr, var;
      addr += sz(s)*numberof(var);
    }

    if (iarg >= narg) break;
    var = args(++iarg);
  }

  if (handle) return f;
  return vclose(f);
}
wrap_args, vpack;

func vunpack(args)
/* DOCUMENT eof = vunpack(bytes, var1, var2, ...);
         or nextvar = vunpack(bytes,-);
         or vunpack, bytes;
     unpack variables VAR1, VAR2, ... from a byte stream BYTES created
     with vpack.  The vunpack call modifies BYTES to save the number of
     variables which have already been unpacked, so you can perform the
     unpack operation with multiple calls.  Calling vunpack as a
     subroutine with no VARi arguments resets this information, restoring
     BYTES to its original value (that is, as vpack returned it).
     Called as a function, vunpack returns 1 if more variables remain
     to be unpacked, or 0 if no more variables remain.
     For example, if BYTES contains 5 variables:
       bytes = vpack(var1, var2, var3, var4, var5);
     You can retrieve the variables by a single call to vunpack:
       vunpack, bytes, var1, var2, var3, var4, var5;
     Or by a sequence of calls to vunpack:
       vunpack, bytes, var1, var2;
       vunpack, bytes, var3, var4, var5;
  SEE ALSO: vunpack, vsave
 */
{
  if (numberof(args(-)))
    error, "unexpected keyword argument";
  bytes = args(1);
  if (structof(bytes)!=char || numberof(bytes)<136)
    error, "first argument must be byte stream from vpack to unpack";
  n = args(0);
  p = array(char, 4, 32);
  p(*) = bytes(1:128);
  prims = array(0, 32);
  for (i=3 ; i>=0 ; --i) prims = (prims<<8) | (p(i+1,) & 0xff);
  if (sizeof(long) > 4) {
    list = where(prims(3:18:3) & 0x80000000);
    if (numberof(list)) prims(3*list) |= ~0xffffffff;
  }
  sz = prims(1:22:3);
  sz(7:8) = [2*sz(6), 1];
  p = bytes(129:136);
  addr = 0;
  for (i=7 ; i>=0 ; --i) addr = (addr<<8) | (p(i+1) & 0xff);
  if (addr == 0) addr = 136;
  if (n > 1) {
    f = vopen(bytes, 1);
    set_primitives, f, prims;
    save, f, complex;
    flag = (n==2 && args(0,2)==1);
  }
  for (i=2 ; i<=n ; ++i) {
    var = [];
    if (addr < sizeof(f)) {
      s = 0;
      _read, f, addr, s;
      addr += sz(4);
      if (s > 0) {
        d = array(0, s/16);
        s = s%16;
        _read, f, addr, d;
        addr += sz(4)*numberof(d);
        if (s == 1) var = array(char, d);
        else if (s == 2) var = array(short, d);
        else if (s == 3) var = array(int, d);
        else if (s == 4) var = array(long, d);
        else if (s == 5) var = array(float, d);
        else if (s == 6) var = array(double, d);
        else if (s == 7) var = array(complex, d);
        else if (s == 8) {
          nn = 0;
          _read, f, addr, nn;
          addr += sz(4);
          var = array(char, nn);
        }
        _read, f, addr, var;
        addr += sz(s)*numberof(var);
        if (s == 8) {
          eq_nocopy, nn, var;
          var = array(string, d);
          nn = strchar(nn);
          if (numberof(nn) == 2*numberof(var)) {
            list = where(nn(1:numberof(var)));
            nn = nn(numberof(var)+1:0);
            nn(list) = "";
          }
          var(*) = nn;
        }
      }
    }
    if (!flag) args, i, var;
  }
  close, f;  /* harmless if f never opened */
  p = array(char, 8);
  if (n>1 || !am_subroutine())
    for (i=0 ; i<sizeof(addr) ; ++i) p(i+1) = (addr >> (8*i)) & 0xff;
  bytes(129:136) = p;
  args, 1, bytes;
  if (flag) return var;
  else if (!am_subroutine()) return addr >= numberof(bytes);
}
wrap_args, vunpack;

extern popen;
/* DOCUMENT f= popen(command, mode)
     opens a pipe to COMMAND, which is executed as with the system
     function.  If MODE is 0, the returned file handle is open for
     reading, and you are reading the stdout produced by COMMAND.
     If MODE is 1, f is opened for writing and you are writing to
     the stdin read by COMMAND.
  SEE ALSO: open, system
 */

extern fflush;
/* DOCUMENT fflush, file
     flush the I/O buffers for the text file FILE.  (Binary files are
     flushed at the proper times automatically.)  You should only need
     this after a write, especially to a pipe.
   SEE ALSO: write, popen
 */

func create(filename)
/* DOCUMENT f= create(filename)
     is a synonym for       f= open(filename, "w")
     Creates a new text file FILENAME, destroying any existing file of
     that name.  Use the write function to write into the file F.
   SEE ALSO: write, close, open
 */
{ return open(filename, "w"); }
errs2caller, create;

extern close;
/* DOCUMENT close, f
     closes the I/O stream F (returned earlier by the open function).
     If F is a simple variable reference (as opposed to an expression),
     the close function will set F to nil.  If F is the only reference
     to the I/O stream, then "close, f" is equivalent to "f= []".
     Otherwise, "close, f" will close the file (so that subsequent
     I/O operations will fail) and print a warning message about the
     outstanding ("stale") references.
  SEE ALSO: open, read, write, rdline, bookmark, backup, save, restore,
            rename, remove
 */

extern filepath;
/* DOCUMENT filepath(file);
     Return full path name of file(s).  Argument FILE can be either an open
     binary/text file or an array of file names (in the latter case tilde
     expansion is performed and the result will have the same shape as the
     input).

   SEE ALSO: cd, lsdir, mkdir, open.
*/

extern rename;
extern remove;
/* DOCUMENT rename, old_filename, new_filename
            remove filename
     rename or remove a file.
   SEE ALSO: open, close, openb
 */

/*= SECTION(filetxt) text i/o to terminal, file, or string =================*/

extern print;
/* DOCUMENT print, object1, object2, object3, ...
         or print(object1, object2, object3, ...)
     prints an ASCII representation of the OBJECTs, in roughly the format
     they could appear in Yorick source code.  When invoked as a subroutine
     (in the first form), output is to the terminal.  When invoked as a
     function (int the second form), the output is stored as a vector of
     strings, one string per line that would have been output.
     Printing a structure definition prints the structure definition;
     printing a function prints its "func" definition; printing files,
     bookmarks, and other objects generally provides some sort of
     useful description of the object.
  SEE ALSO: totxt, pr1, print_format, write, exit, error, nameof, typeof
 */

func pr1(x)
/* DOCUMENT pr1(x)
     returns text representing expression X, equivalent to print(X)(1).
   SEE ALSO: print, swrite, totxt
 */
{ return print(x)(1); }

extern print_format;
/* DOCUMENT print_format, line_length, max_lines, char=, short=, int=,
                          float=, double=, complex=, pointer=
     sets the format string the print function will use for each of
     the basic data types.  Yorick format strings are the same as the
     format strings for the printf function defined in the ANSI C standard.
     The default strings may be restored individually by setting the
     associated format string to ""; all defaults are restored if
     print_format is invoked with no arguments.  The default format strings
     are:  "0x%02x", "%d", "%d", "%ld", "%g", "%g", and "%g+%gi".
     Note that char and short values are converted to int before being
     passed to printf, and that float is converted to double.
     If present, an integer positional argument is taken as the line
     length; <=0 restores the default line length of 80 characters,
     while nil [] leaves the line length unchanged.
     A second positional argument, if present, becomes the maximum number
     of lines to output; <=0 restores the default of 5000 lines.  A single
     print command will not produce more than this many lines of output;
     output simply stops without any additional messages.
  SEE ALSO: print, write, totxt, nameof, typeof
 */

func totxt(x, fmt)
/* DOCUMENT totxt(x)
         or totxt(x, fmt)
     returns text representing expression X.  If X is not numeric,
     then totxt(x) is the same as print(x).  If X is numeric, then
     totxt returns an array of strings with the same dimensions as X.
     Integers get %d format, while reals get %g format, unless you
     specify FMT.  FMT can be a single numeric format, or just a
     number with the following interpretation:
       FMT = integer w  means %wd for integers or %wf for reals
       FMT = real w.p  means %wd for integers or %w.pf for reals
     In either case, a negative value -w or -w.p switches to hex
     format for integers %wx or exponential format %w.pe for reals.
   SEE ALSO: print, swrite, tonum
 */
{
  i = identof(x);
  if (i > 6) return print(x);
  if (is_void(fmt)) {
    fmt = (i>3)? "%g" : ((i==3)? "%ld" : "%d");
  } else if (structof(fmt) != string) {
    a = fmt<0;
    fmt = abs(fmt);
    w = long(fmt);
    p = long(100.*(fmt-w)+1.e-6);
    if (!(p%10)) p /= 10;
    w = w? print(w)(1) : "";
    if ((identof(fmt)>3) && (i>3)) p = "."+print(p)(1);
    else p = "";
    if (i > 3) a = a? "e" : "f";
    else if (i == 3) a = a? "lx" : "ld";
    else a = a? "x" : "d";
    fmt = "%"+w+p+a;
  }
  if (i < 6) return swrite(format=fmt, x);
  return swrite(format=fmt, x.re)+"+"+swrite(format=fmt, x.im)+"i";
}

extern read;
extern sread;
/* DOCUMENT n= read(f, format=fstring, obj1, obj2, ...)
         or n= read(prompt= pstring, format=fstring, obj1, obj2, ...)
         or n= sread(source, format=fstring, obj1, obj2, ...)
     reads text from I/O stream F (1st form), or from the keyboard (2nd
     form), or from the string or string array SOURCE (3rd form),
     interprets it according to the optional FSTRING, and uses that
     interpretation to assign values to OBJ1, OBJ2, ...  If the input
     is taken from the keyboard, the optional prompt PSTRING (default
     "read> ") is printed before each line is read.  The Yorick write
     function does not interact with the read function -- writes are
     always to end-of-file, and do not affect the sequence of lines
     returned by read.  The backup (and bookmark) function is the
     only way to change the sequence of lines returned by read.

     There must be one non-supressed conversion specifier (see below)
     in FSTRING for each OBJ to be read; the type of the conversion
     specifier must generally match the type of the OBJ.  That is,
     an integer OBJ requires an integer specifier (d, i, o, u, or x)
     in FSTRING, a real OBJ requires a real specifier (e, f, or g),
     and a string OBJ requires a string specifier (s or []).  An OBJ
     may not be complex, a pointer, a structure instance, or any non-
     array Yorick object.  If FSTRING is not supplied, or if it has
     fewer conversion specifiers than the number of OBJ arguments,
     then Yorick supplies default specifiers ("%ld" for integers,
     "%lg" for reals, and "%s" for strings).  If FSTRING contains more
     specifiers than there are OBJ arguments, the part of FSTRING
     beginning with the first specifier with no OBJ is ignored.

     The OBJ may be scalar or arrays, but the dimensions of every OBJ
     must be identical.  If the OBJ are arrays, Yorick behaves as
     if the read were called in a loop numberof(OBJ1) times, filling
     one array element of each of the OBJ according to FSTRING on
     each pass through the loop.  (Note that this behavior includes
     the case of reading columns of numbers by a single call to read.)

     The return value N is the total number of scalar assignments
     which were made as a result of this call.  (If there were 4
     OBJ arguments, and each was an array with 17 elements, a return
     value of N==35 would mean the following:  The first 8 elements
     of OBJ1, OBJ2, OBJ3, and OBJ4 were read, and the 9th element of
     OBJ1, OBJ2, and OBJ3 was read.)  The read function sets any
     elements of the OBJ which were not read to zero -- hence,
     independent of the returned N, the all of the old data in the
     OBJ arguments is overwritten.

     The read or sread functions continue reading until either:
     (1) all elements of all OBJ have been filled, or (2) end-of-file
     (or end of SOURCE for sread) is reached ("input failure"), or
     (3) part of FSTRING or a conversion specifier supplied by
     default fails to match the source text ("matching failure").

     The FSTRING is composed of a series of "directives" which are
     (1) whitespace -- means to skip any amount of whitespace in the
         source text
     (2) characters other than whitespace and % -- must match the
         characters in the source text exactly, or matching failure
         occurs and the read operation stops
     (3) conversion specifiers beginning with % and ending with a
         character specifying the type of conversion -- optionally
         skip whitespace, then convert as many characters as
         continue to "look like" the conversion type, possibly
         producing a matching failure
     The conversion specifier is of the form %*WSC, where:
     * is either the character '*' or not present
       A specifier beginning with %* does not correspond to any of
       the OBJ; the converted value will be discarded.
     W is either a positive decimal integer specifying the maximum
       field width (not including any skipped leading whitespace),
       or not present if any number of characters up to end-of-line
       is acceptable.
     S is either one of the characters 'h', 'l', or 'L', or not
       present.  Yorick allows this for compatibility with the C
       library functions, but ignores it.
     C is a character specifying the type of conversion:
       d   - decimal integer
       i   - decimal, octal (leading 0), or hex (leading 0x) integer
       o   - octal integer
       u   - unsigned decimal integer (same as d for Yorick)
       x, X            - hex integer
       e, f, g, E, G   - floating point real
       s   - string of non-whitespace characters
       [xxx]   - (xxx is any sequence of characters) longest string
                 of characters matching those in the list
       [^xxx]  - longest string of characters NOT matching those in
                 the list (this is how you can extend %s to be
                 delimited by something other than whitespace)
       %   - the ordinary % character; complete conversion
             specification must be "%%"

     The read function is modeled on the ANSI standard C library
     fscanf and sscanf functions, but differs in several respects:
       (1) Yorick's read cannot handle the %c, %p, or %n conversion
           specifiers in FSTRING.
       (2) Yorick's read never results in a portion of a line
           being read -- any unused part of a line is simply discarded
           (end FSTRING with "%[^\n]" if you want to save the trailing
           part of an input line).
       (3) As a side effect of (2), there are some differences between
           fscanf and Yorick's read in how whitespace extending across
           newlines is handled.
   SEE ALSO: rdline, write, open, close, bookmark, backup, save, restore,
             read_n, tonum
 */

extern rdline;
/* DOCUMENT rdline(f)
         or rdline(f, n, prompt= pstring)
     returns next line from stream F (stdin if F nil).  If N is non-nil,
     returns a string array containing the next N lines of F.  If
     end-of-file occurs, rdline returns nil strings.  If F is nil,
     uses the PSTRING to prompt for input (default "read> ").
   SEE ALSO: read, open, close, bookmark, backup, read_n, rdfile
 */

func rdfile(f, nmax)
/* DOCUMENT rdfile(f)
         or rdfile(f, nmax)
     reads all remaining lines (or at most NMAX lines) from file F.
     If NMAX is omitted, it defaults to 2^20 lines (about a million).
     The result is an array of strings, one per line of F.
   SEE ALSO: rdline
 */
{
  if (structof(f)==string) f = open(f);
  if (is_void(f)) error, "use rdline to read from stdin";
  if (is_void(nmax) || nmax<=0) nmax = 1048576;
  n = min(4096, nmax);
  s = rdline(f, n);
  while (s(0) && n<nmax) {
    grow, s, rdline(f, min(nmax-n, n));
    n = numberof(s);
  }
  if (!s(0)) s = s(where(s));
  return s;
}

autoload, "readn.i", read_n;
local read_n;
/* DOCUMENT read_n, f, n0, n1, n2, ...
     grabs the next numbers N0, N1, N2, ... from file F, skipping over
     any whitespace, comma, semicolon, or colon delimited tokens which
     are not numbers.  (Actually, only the first and last characters of
     the token have to look like a number -- 4xxx3 would be read as 4.)
     ***WARNING*** at most ten Ns are allowed
     The Ns can be arrays, provided all have the same dimensions.
   SEE ALSO: read, rdline
 */

func tonum(s, &mask, nan=)
/* DOCUMENT tonum(s)
         or tonum(s, mask)
     returns array of numbers corresponding to given array of strings S.
     For each element of S which consists of a single number (either a
     decimal integer or floating point number), tonum returns the numeric
     value.  The return value is type double and the same dimensions as S.
     Any elements of S which cannot be interpreted as single numeric
     values will have the value -1.0e99 in the result array.  You can
     specify a different value for "not a number" with the nan= keyword.
     The optional MASK is an output of type int of the same dimensions
     as S, which is 1 where S is a floating point number (with decimal
     point and/or exponent), 3 where S is an integer, and 0 where S is
     not a number.
   SEE ALSO: sread, totxt, strgrep
 */
{
  if (is_void(nan)) nan = -1.e99;
  x = array(nan, dimsof(s));
  i = strgrep("^[ \t]*[+-]?[0-9]+[ \t]*$", s);
  r = "[+-]?(\\.?[0-9]+|[0-9]+\\.[0-9]*)([dDeE][+-]?[0-9]+)?";
  r = strgrep("^[ \t]*"+r+"[ \t]*$", s);
  mask = (r(2,..)>0) | ((i(2,..)>0)<<1);
  list = where(mask);
  n = numberof(list);
  if (n) {
    y = array(0., n);
    if (sread(s(list)+" ",y) != n) error, "(BUG) number not a number??";
    x(list) = y;
  }
  return x;
}

extern write;
extern swrite;
/* DOCUMENT n= write(f, format=fstring, linesize=l, obj1, obj2, ...)
            n= write(format=fstring, linesize=l, obj1, obj2, ...)
         or strings= swrite(format=fstring, linesize=l, obj1, obj2, ...)
     writes text to I/O stream F (1st form), or to the terminal (2nd
     form), or to the STRINGS string array (3rd form), representing
     arrays OBJ1, OBJ2, ..., according to the optional FSTRING.  The
     optional linesize L defaults to 80 characters, and helps restrict
     line lengths when FSTRING is not given, or does not contain
     newline directives.  The write function always appends to the
     end of a text file; the position for a sequence of reads is
     not affected by intervening writes.

     There must be one conversion specifier (see below) in FSTRING for
     each OBJ to be written; the type of the conversion specifier must
     generally match the type of the OBJ.  That is, an integer OBJ
     requires an integer specifier (d, i, o, u, x, or c) in FSTRING,
     a real OBJ requires a real specifier (e, f, or g), a string OBJ
     requires the string specifier (s), and a pointer OBJ requires a
     the pointer specifier (p).  An OBJ may not be complex, a structure
     instance, or any non-array Yorick object.  If FSTRING is not
     supplied, or if it has fewer conversion specifiers than the
     number of OBJ arguments, then Yorick supplies default specifiers
     (" %8ld" for integers, " %14.6lg" for reals, " %s" for strings, and
     " %8p" for pointers).  If FSTRING contains more specifiers than
     there are OBJ arguments, the part of FSTRING beginning with the
     first specifier with no OBJ is ignored.

     The OBJ may be scalar or arrays, but the dimensions of the OBJ
     must be conformable.  If the OBJ are arrays, Yorick behaves as
     if he write were called in a loop dimsof(OBJ1, OBJ2, ...) times,
     writing one array element of each of the OBJ according to FSTRING
     on each pass through the loop.  The swrite function returns a
     string array with dimensions dimsof(OBJ1, OBJ2, ...).  The write
     function inserts a newline between passes through the array if
     the line produced by the previous pass did not end with a
     newline, and if the total number of characters output since the
     previous inserted newline, plus the number of characters about
     to be written on the current pass, would exceed L characters
     (L defaults to 80).  The write function returns the total
     number of characters output.

     The FSTRING is composed of a series of "directives" which are
     (1) characters other than % -- copied directly to output
     (2) conversion specifiers beginning with % and ending with a
         character specifying the type of conversion -- specify
         how to convert an OBJ into characters for output
     The conversion specifier is of the form %FW.PSC, where:
     F is zero or more optional flags:
       - left justify in field width
       + signed conversion will begin with either + or -
         (space) signed conversion  will begin with either space or -
       # alternate form (see description of each type below)
       0 pad field width with leading 0s instead of leading spaces
     W is either a decimal integer specifying the minimum field width
       (padded as specified by flags), or not present to use the
       minimum number of characters required.
     .P is either a decimal integer specifying the precision of the
       result, or not present to get the default.  For integers, this
       is the number of digits to be printed (possibly forcing leading
       zeroes), and defaults to 1.  For reals, this is the number of
       digits after the decimal point, and defaults to 6.  For strings,
       this is the maximum number of characters to print, and defaults
       to infinity.
     S is either one of the characters 'h', 'l', or 'L', or not
       present.  Yorick allows this for compatibility with the C
       library functions, but ignores it.
     C is a character specifying the type of conversion:
       d, i  - decimal integer
       o     - octal integer (# forces leading 0)
       u     - unsigned decimal integer (same as d for Yorick)
       x, X            - hex integer (# forces leading 0x)
       f     - floating point real in fixed point notation
               (# forces decimal)
       e, E  - floating point real in scientific notation
       g, G  - floating point real in fixed or scientific notation
               depending on the value converted (# forces decimal)
       s   - string of ASCII characters
       c   - integer printed as corresponding ASCII character
       p   - pointer
       %   - the ordinary % character; complete conversion
             specification must be "%%"

     The write function is modeled on the ANSI standard C library
     fprintf and sprintf functions, but differs in several respects:
       (1) Yorick's write cannot handle the %n conversion specifier
           in FSTRING.
       (2) Yorick's write may insert additional newlines if the OBJ
           are arrays, to avoid extremely long output lines.
   SEE ALSO: print, exit, error, read, rdline, open, close, save, restore
 */

extern bookmark;
extern backup;
/* DOCUMENT backup, f
         or bmark= bookmark(f)
            ...
            backup, f, bmark
     back up the text stream F, so that the next call to the read
     function returns the same line as the previous call to read
     (note that you can only back up one line).  If the optional
     second argument BMARK is supplied, restores the state of the
     file F to its state at the time the bookmark function was
     called.
     After a matching failure in read, use the single argument form
     of backup to reread the line containing the matching failure.
   SEE ALSO: read, rdline, open, close
 */

func select_name(list, index=,  prompt=, forever=,
                 label=, width=, sep=, eol=, bol=, maxcols=)
/* DOCUMENT select_name(list)
     Print out array of strings LIST (using print_columns) and
     interactively ask the user a number/item in the list and return the
     selected item.  If keyword INDEX is true, the item number is returned
     rather than its value.  The prompt string can be set with keyword PROMPT
     (default is " Select one item: ").  If keyword FOREVER is true the user
     is prompted until a valid choice is made.

     Other keywords are passed to print_columns: LABEL (as
     LABEL), WIDTH, SEP, EOL, BOL and MAXCOLS.

   SEE ALSO: print_columns. */
{
  number = numberof(list);
  print_columns, list, label=(is_void(label) ? " - " : label),
    width=width, sep=sep, eol=eol, bol=bol, maxcols=maxcols;
  if (is_void(prompt)) prompt=" Select one item: ";
  for (;;) {
    t = string(0);
    k = 0;
    s = rdline(prompt=prompt);
    if (sread(s, format="%d %s", k, t) == 1 && k >= 1 && k <= number) break;
    if (numberof((k = where(list == s))) == 1) {
      k = k(1);
      break;
    }
    if (! forever) return;
  }
  return (index ? k : list(k));
}

func select_file(dir, prompt=, width=, forever=, all=, pattern=)
/* DOCUMENT select_file()
         or select_file(dir)
     Interactively select name of an existing file starting at current working
     directory or at last selected directory or at DIR if this argument is
     specified.  The function returns full path of selected file or nil [] if
     no valid selection is made.  If keyword FOREVER is true, a file must be
     selected for the function to return.

     If keyword ALL is true, then all files and directories get displayed --
     even the "hidden" ones which name start with a dot.  In any cases, the
     current and parent directories ("." and "..") get displayed to allow the
     user to re-scan the current directory or to go into the parent directory.

     Keyword PATTERN can be set to a regular expression to select only files
     that match PATTERN.  For instance, PATTERN="\\.(tgz|tar\\.gz)$" would
     match any files with suffix ".tgz" or ".tar.gz".

     Keyword WIDTH can be used to specify a different text width than the
     default of 79 characters.

     Keyword PROMPT can be set to change the default prompt:
       " Select file/directory: "

   SEE ALSO: lsdir, regmatch, print_columns. */
{
  /* fool codger */ extern __select_file_dir;  local dir_list;
  if (is_void(width)) width = 79;
  if (is_void(prompt)) prompt=" Select file/directory: ";
  if (! is_void(pattern) && ! (is_string(pattern) && is_scalar(pattern))) {
    error, "value of keyword PATTERN must be nil or a scalar string";
  }
  cwd = get_cwd();
  if (! is_void(dir)) __select_file_dir = dir;
  if (structof(__select_file_dir) != string) {
    __select_file_dir = cwd;
  } else {
    __select_file_dir = cd(__select_file_dir);
  }

  hline = "-------------------------------------";
  for (;;) {
    file_list = lsdir(__select_file_dir, dir_list);
    if (! all) {
      if ((n = numberof(file_list)) > 0) {
        i = where(strpart(file_list, 1:1) != ".");
        if (numberof(i) != n) file_list = file_list(i);
      }
      if ((n = numberof(dir_list)) > 0) {
        i = where(strpart(dir_list, 1:1) != ".");
        if (numberof(i) != n) dir_list = dir_list(i);
      }
    }
    if (pattern && (n = numberof(file_list)) > 0) {
      i = where(strgrep(pattern, file_list)(2,..) >= 0);
      if (numberof(i) != n) file_list = file_list(i);
    }
    grow, dir_list, ".", ".."; /* use . to allow reading directory again */
    dir_list = dir_list(sort(dir_list));
    list = dir_list + "/";
    if (is_array(file_list)) {
      grow, list, file_list(sort(file_list));
      file_list = [];
    }
    ndirs = numberof(dir_list);
    number = numberof(list);

    /* Print out directory list. */
    text = print_columns(list, label=": ", width=width,
                         sep=, eol=" ", bol="| ", maxcols=);
    text_len = strlen(text(1))+1;
    len = max(width, text_len);
    while (strlen(hline) < len) hline += hline;
    head_line = "[" + __select_file_dir + "]";
    n = (len - strlen(head_line) - 2)/2;
    if (n > 0) head_line = strpart(hline, 1:n)+head_line;
    n = len - strlen(head_line) - 2;
    if (n > 0) head_line += strpart(hline, 1:n);
    write, format=",%s.\n", head_line;
    write, format=swrite(format="%%-%ds|\n", len-1), text;
    write, format="`%s'\n", strpart(hline, 1:len-2);
    for (;;) {
      t = string(0);
      k = 0;
      s = rdline(prompt=prompt);
      if (sread(s, format="%d %s", k, t) == 1 && k >= 1 && k <= number) break;
      if (numberof((k = where(list == s))) == 1) {
        k = k(1);
        break;
      }
      if (numberof((k = where(dir_list == s))) == 1) {
        k = k(1);
        break;
      }
      if (! forever) {
        cd, cwd;
        return;
      }
    }
    if (k > ndirs) {
      cd, cwd;
      return __select_file_dir + list(k);
    }
    __select_file_dir = cd(__select_file_dir + dir_list(k));
  }
}

func print_columns(list, label=, width=, start=, sep=, eol=, bol=, maxcols=)
/* DOCUMENT print_columns, list;
         or print_columns(list);
     Write array of strings LIST in columns.  In subroutine form, the result
     is printed to standard output; otherwise, the function returns an array
     of formatted strings (one per row).

     The maximum width (in number of characters) of each row can be specified
     with keyword WIDTH (default 79).  But actual width may be larger, since
     at least one column is produced.

     The maximum number of columns may be limited by using keyword MAXCOLS (by
     default, there is no limit).

     Keywords BOL, SEP and EOL, can be set to scalar strings to use at begin
     of line, between each column, at end of line respectively.  SEP can also
     be the number of spaces to insert between columns.  The default are:
     BOL="", SEP=5 (five spaces) and EOL=string(0).

     Keyword LABEL can be used to number items. LABEL must be a scalar string.
     If LABEL contains a "%d", it is used to format the index; otherwise,
     LABEL is the string to use as separator between indices and items.  For
     instance:
       label="[%d] "  yields: "[1] first_item    [2] second_item  ..."
       label=" - "    yields: "1 - first_item    2 - second_item  ..."

     Keyword START can be used to specify the starting index for numbering
     items (default START=1).


   SEE ALSO: swrite, select_name, select_file. */
{
  number = numberof(list);
  if (is_scalar(label) && is_string(label)) {
    if (is_void(start)) start = 1;
    index = indgen(start:start+number-1);
    if (strmatch(label, "%d")) {
      index = swrite(format=label, index);
    } else {
      index = swrite(format="%d", index) + label;
    }
    len = strlen(index);
    if (max(len) != min(len)) {
      /* Justify index list. */
      index = swrite(format=swrite(format="%%%ds", max(len)), index);
    }
    list = index + list(*);
  } else if (is_void(label)) {
    list = list(*);
  } else {
    error, "value of keyword LABEL must be nil or a scalar string";
  }

  if (is_void(bol)) bol = "";
  if (is_void(eol)) eol = string(0);
  if (is_void(width)) width = 79;
  if (structof(sep) != string) {
    /* Convert margin separator into spaces. */
    if (is_void(sep)) sep = 5;
    sep = swrite(format=swrite(format="%%%ds", sep), "");
  }
  len = max(strlen(list));
  slen = strlen(sep);
  ncols = (width + slen - strlen(bol) - strlen(eol))/(len + slen);
  if (! is_void(maxcols) && ncols > maxcols) ncols = maxcols;
  if (ncols < 1) ncols = 1;
  nrows = (number+ncols-1)/ncols;

  (tmp = array(string, nrows, ncols))(1:number) = unref(list);
  if (ncols > 1) {
    fmt = swrite(format="%%-%ds%s", len, sep);
    for (j=1 ; j<ncols ; ++j) bol += swrite(format=fmt, tmp(,j));
  }
  fmt = (eol ? swrite(format="%%-%ds%s", len, eol) : "%s");
  bol += swrite(format=fmt, tmp(,ncols));
  if (! am_subroutine()) return bol;
  write, format="%s\n", bol;
}

/*= SECTION(include) including files and parsing strings ===================*/

extern include;
extern require;
/* DOCUMENT #include "yorick_source.i"
            require, source
            include, source
         or include, source, now

     The SOURCE argument can be a scalar string, interpreted as a
     filename like "yorick_source.i", the text in a char array, or
     the text in a 1D array of strings (as returned by a two argument
     call to rdline).

     #include is a parser directive, not a Yorick statement.  Use it
     to read Yorick source code which you have saved in a file; the
     file yorick_source.i will be read one line at a time, exactly as
     if you had typed those lines at the keyboard.  The following
     directories are searched (in this order) to find yorick_source.i:

        .               (current working directory)
        ~/yorick        (your personal directory of Yorick functions)
        ~/Yorick        (your personal directory of Yorick functions)
        Y_SITE/i        (Yorick distribution library)
        Y_SITE/contrib  (contributed source at your site)
        Y_SITE/i0       (Yorick startup and package include files)
        Y_HOME/lib      (Yorick architecture dependent include files)

     To find out what is available in the Y_SITE/i directory,
     type:
         library
     You can also type
         Y_SITE
     to find the name of the site directory at your site, go to the
     include or contrib subdirectory, and browse through the *.i files.
     This is a good way to learn how to write a Yorick program.  Be
     alert for files like README as well.

     The require function checks to see whether FILENAME has already
     been included (actually whether any file with the same final
     path component has been included).  If so, require is a no-op,
     otherwise, the action is the same as the include function with
     NOW == 1.

     The include function causes Yorick to parse and execute FILENAME
     immediately.  The effect is similar to the #include parser
     directive, except the finding, parsing, and execution of FILENAME
     occurs at runtime.  The NOW argument has the following meanings:
       NOW == -1   filename pushed onto stack, popped and parsed
                   when all pending input is exhausted
       NOW == 0    (or nil, default) parsed just before next input
                   line would be parsed
       NOW == 1    parsed immediately, resuming current interpreted
                   program when finished (like require)
       NOW == 2    like 0, except no error if filename does not exist
       NOW == 3    like 1, except no error if filename does not exist

   SEE ALSO: set_path, Y_SITE, plug_in, autoload, include_all, funcdef
             include1
 */

extern include1;
/* DOCUMENT function = include1(source)
     is similar to include, but parses SOURCE only until the first
     executable *main* program, which it returns as an interpreted
     function rather than executing immediately.

     The SOURCE argument can be a scalar string, interpreted as a
     filename like "yorick_source.i", the text in a char array, or
     the text in a 1D array of strings (as returned by a two argument
     call to rdline).

   SEE ALSO: include, funcdef
 */

extern current_include;
/* DOCUMENT current_include()
     If Yorick is parsing a file, this function returns the absolute path
     of this file; otherwise, this function returns nil.

   SEE ALSO: include, require.
 */

extern get_includes;
/* DOCUMENT get_includes()
     Returns an array of strings with the names of all included files so
     far.

   SEE ALSO: set_path, current_include, include, require.
 */

func include_all(dir, ..)
/* DOCUMENT include_all, dir1, dir2, ...
     include all files in directories DIR1, DIR2, ..., with names
     ending in the ".i" extension.  (This is mostly for use to load
     the i-start directories when yorick starts; see i0/stdx.i.)
     If any of the DIRi do not exist, or are empty, they are
     silently skipped.  Filenames beginning with "." are also skipped,
     even if they end in ".i".  The files are included in alphabetical
     order, DIR1 first, then DIR2, and so on.

   SEE ALSO: include, include1, autoload
 */
{
  if (is_void(dir)) return;
  dir = array(dir, more_args()+1);
  for (i=2 ; i<=numberof(dir) ; ++i) dir(i) = next_arg();
  /* provide hook for mpy to prevent all ranks from calling lsdir */
  files = _include_all_hook(dir);
  for (i=1 ; i<=numberof(files) ; ++i) include, files(i), 3;
}
func include_all_ls(dir)
{
  /**/ local files;
  for (i=1 ; i<=numberof(dir) ; ++i) {
    list = lsdir(dir(i));
    if (structof(list) != string) continue;
    list = list(where((strpart(list,1:1)!=".")&(strpart(list,-1:0)==".i")));
    if (!numberof(list)) continue;
    list = list(sort(list));
    if (strpart(dir(i),0:0) != "/") dir(i) += "/";
    grow, files, dir(i)+list;
  }
  return files;
}
_include_all_hook = include_all_ls;

func istart_include
{
  dir = istart_places(Y_SITE);
  if (Y_HOME!=Y_SITE) grow, dir, istart_places(Y_HOME);
  if (Y_HOME_PKG && (Y_HOME_PKG!=Y_SITE)) grow, dir, istart_places(Y_HOME_PKG);
  if (!_istart_hook()) grow, dir, [Y_USER+"i-start"];
  files = _include_all_hook(dir);
  for (i=1 ; i<=numberof(files) ; ++i) include, files(i), 3;
}
_istart_hook = batch;

func istart_places(dir)
{
  i = strfind("/",dir, n=1024);
  i = i(1 : min(where(i<0))-1);
  parts = strpart(dir, grow([0],i));
  if (numberof(parts) < 2) return dir+"i-start";
  if (parts(0) == "") parts = parts(1:-1);
  list = where(strmatch(parts,"yorick",1));
  if (numberof(list)) {
    i = list(1);
  } else {
    list = where((parts=="lib") | (parts=="share"));
    if (numberof(list)) i = max(list(0)+1,i);
    else i = numberof(parts);
  }
  parts += "/";
  dir = array(sum(parts(1:i)), numberof(parts)-i+1);
  parts = (i<numberof(parts))? parts(i+1:0) : [];
  for (i=1 ; i<=numberof(parts) ; ++i) dir(i+1:0) += parts(i);
  parts = array("site-i", numberof(dir));
  parts(1) = "i-start";
  return dir(::-1) + parts;
}

func customize { if (!_istart_hook()) include, "custom.i"; }

extern autoload;
/* DOCUMENT autoload, ifile, var1, var2, ...
         or autoload, ifile
     causes IFILE to be included when any of the variables VAR1, VAR2, ...
     is referenced as a function or subroutine.  Multiple autoload
     calls may refer to a single IFILE; the effect is cumulative.  Note
     that any reference to a single one of the VARi causes all of them
     to be replaced (when IFILE is included).
     The semantics of this process are complicated, but should work
     as expected in most cases: After the call to autoload, the VARi
     may not be redefined (e.g.- VARi=something or func VARi) without
     generating a warning message, and causing all the VARi for the
     same IFILE to become undefined.  The semantic subtlety arises
     from the yorick variable scoping rules; if any of the VARi has local
     scope for any function in the calling chain when the inclusion of
     IFILE is actually triggered, only those local values will be
     replaced.  (The autoload function is no different than the require
     or include functions in this regard.)
     The second form, with no VARi, cancels the autoload, without giving
     any warning; all the VARi become undefined.

     Before IFILE is included, the VARi behave like [] (nil) variables
     as far as their response to the is_void function, and the ! and ?
     operators.  (You can use is_func to discover whether a variable is
     an autoload.)  Only their actual use in a function or subroutine call
     will trigger the autoload.  While the IFILE may define the VARi
     as any type of object, the autoload feature only works as intended
     if the VARi are defined as interpreted or built-in functions.  The
     only way it makes sense for a VARi to be a built-in function, is
     if the IFILE executes a plug_in command to dynamically load an
     associated compiled library.

     If IFILE (or a file with the same name) has already been included,
     autoload is a silent no-op.  This is exactly analogous to the
     behavior of the require function; it does not harm to call either
     require or autoload if the IFILE has already been included.  Note
     that you may want to place a require at the beginning of a file
     you expect to be autoloaded, in preference to providing separate
     autoloads for the second file.

   SEE ALSO: include, require, plug_in, is_func
 */

extern funcdef;
/* DOCUMENT function = funcdef(command_line)
     creates an anonymous interpreted function from the input
     COMMAND_LINE, equivalent to
     func function {
       command_line;
     }
     The COMMAND_LINE string is restricted to the following
     format:
       "funcname arg1 arg2 ..."
     where each of the arguments is one of
     (a) a symbol (that is, a yorick variable name)
     (b) a decimal integer
     (c) a real number
     (d) a quoted string
     The quoted string is enclosed in double quotes, and a
     backslash can be used to escape a double quote or a
     backslash (but the other backslash escape sequences are
     not recognized and unnecessary - just insert the ascii code).

     Note that funcdef merely creates the function; if you want
     to execute it and discard it, use the following statement:
       funcdef(command_line);

     The huge advantage of funcdef over the full yorick parser
     is that it is stateless, which means you can invoke it to
     generate actions for event callbacks.  The extreme simplicity
     of the permitted COMMAND_LINE is not a limitation for this
     application, because you are free to invoke an arbitrarily
     complex "funcname", and to provide it with arbitrary inputs.

     The intent with funcdef is not to permit you to create an
     arbitrary toolkit of interpreted functions, but merely to
     allow you to invoke such a toolkit; the toolkit itself is
     supposed to be parsed by the ordinary include, require, or
     autoload mechanisms.  Generally, you will have to design
     an interpreted toolkit somewhat differently if it is to be
     invoked by funcdef.  For example, funcdef does not allow
     you to set variables as in x=value, but you can use the
     funcset (or similarly designed) function to set variables
     like this:
       funcdef("funcset x value")

     Do not attempt to use funcdef to input vast amounts of data.
     As a rule of thumb, if your funcdef strings have more than a
     couple of dozen tokens, you probably haven't thought hard
     enough about what you are doing.

   SEE ALSO: include, spawn, funcset
*/

func funcset(&v1,x1,&v2,x2,&v3,x3,&v4,x4,&v5,x5,&v6,x6,&v7,x7,&v8,x8)
/* DOCUMENT funcset var1 val1 var2 val2 ...

     Equivalent to
       var1=val1; var2=val2; ...

     This function it is not useful for yorick programs.  It is intended
     to be used to create functions with funcdef that set variable values.

     Handles at most 8 var/val pairs.
     As a special case, if given an odd number of arguments, funcset
     sets the final var to [], e.g.-
       funcset var1 12.34 var2
     is equivalent to
       var1=12.34; var2=[];

   SEE ALSO: funcdef
 */
{
  v1 = x1;
  if (is_void(x1)) return; else v2 = x2;
  if (is_void(x2)) return; else v3 = x3;
  if (is_void(x3)) return; else v4 = x4;
  if (is_void(x4)) return; else v5 = x5;
  if (is_void(x5)) return; else v6 = x6;
  if (is_void(x6)) return; else v7 = x7;
  if (is_void(x7)) return; else v8 = x8;
}

/*= SECTION(plugin) loading plugins ========================================*/

extern plug_in;
/* DOCUMENT plug_in, "pkgname"

     Dynamically link to yorick package "pkgname".  The compiled
     functions of the package are in a shared object file; these
     files have a naming convention which differs slightly on different
     platforms.  On most UNIX systems (including Mac OS X), the
     binary file is named pkgname.so.  On MS Windows systems, the
     binary file is named pkgname.dll.  On HPUX systems, the name is
     pkgname.sl.  The "pkgname" argument to plug_in does not include
     this platform-dependent file extension, so that the yorick code
     containing the plug_in command will be portable.

     After dynamically linking the compiled routines in the pkgname
     shared object binary, yorick runs the function (which must be
     present) yk_pkgname in order to initialize the package.  At
     minimum yk_pkgname returns lists of the new compiled (builtin)
     functions defined by the package and the names by which they
     may be invoked by interpreted code.

     Additionally, yk_pkgname returns a list of files to be included
     containing interpreted wrapper functions for the compiled routines
     and DOCUMENT comments for the help system.  Conventionally, these
     include files are located in the Y_SITE/i0 or Y_HOME/lib directories,
     and the name (of one) of the file(s) is pkgname.i.  If the package
     has been statically linked (i.e.- not by plug_in), these .i files
     are automatically included when yorick starts.  However, if the
     package is loaded dynamically by plug_in, you must arrange to
     include one or all of these .i files as you would any interpreted
     package (e.g.- by the autoload or require functions, or manually).

     The upshot of all this is that the plug_in function is designed
     to be placed at the top of the .i files associated with the
     package.  You are not supposed to call plug_in manually, rather
     when you #include (or autoload) a .i file which needs compiled
     functions, that .i file invokes plug_in to perform any required
     dynamic linking to compiled code.  Thus, the end user does not
     do anything differently for a package that uses dynamically loaded
     compiled code, than for a purely interpreted package.

     Yorick dynamic library support solves a distribution problem.  For
     debugging and creating compiled packages for your own use, you want
     to build special versions of yorick with your compiled routines
     statically linked.  In order to support platforms on which there
     is no dynamic linking, if you call the plug_in function for a
     package that is statically linked (e.g.- plug_in,"yor"), the
     function will silently become a no-op when it notices that the
     "pkgname" package was already loaded at startup.

   SEE ALSO: plug_dir, include, require, autoload
 */

extern plug_dir;
/* DOCUMENT old_dirs = plug_dir(dirname)
         or plug_dir
         or current_dirs = plug_dir()
     causes plug_in to look in DIRNAME for dynamic library files, in
     addition to Y_HOME/lib.  DIRNAME may be an array of strings to
     search multiple directories.  The return value is the previous
     list of directories searched by plug_in.  No checks are made
     for repeats, so be careful not to grow the list indiscriminately.
     In the second form (or called as a subroutine with DIRNAME []),
     empties the plug_in search path; in the third form does not
     alter the current search path.  Note that Y_HOME/lib is omitted
     from the end of the return value, even though it is searched.
   SEE ALSO: plug_in
 */

/*= SECTION(filebin) save and restore binary data ==========================*/

func openb(filename, clogfile, update, open102=, one=)
/* DOCUMENT file = openb(filename)
         or file = openb(filename, clogfile)
     open the existing file FILENAME for read-only binary I/O.
     (Use updateb or createb, respectively, to open an existing file
      with read-write access or to create a new file.)
     If the CLOGFILE argument is supplied, it represents the structure
     of FILENAME in the Clog binary data description language.
     After an openb, the file variable may be used to extract variables
     from the file as if it were a structure instance.  That is, the
     expression "file.var" refers to the variable "var" in file "file".
     A complete list of the variable names present in the file may
     be obtained using the get_vars function.  If the file contains
     history records, the jt and jc functions may be used to set the
     current record -- initially, the first record is current.
     The restore function may be used to make memory copies of data
     in the file; this will be faster than a large number of
     references to "file.var".
     The openb function will recognize families of PDB or netCDF files
     by their sequential names and open all files subsequent to FILENAME
     in such a family as well as FILENAME itself.  You can use the one=1
     keyword to suppress this behavior and open only FILENAME.
     FILENAME may be a file handle to skip the initial open operation.
     This feature is intended to enable in-memory files created with
     vopen to be opened:
       file = openb(vopen(char_array,1));
     FILENAME may also be char_array directly, as returned by vsave.
   SEE ALSO: updateb, createb, open, vopen, cd
             show, jt, jc, restore
             get_vars, get_times, get_ncycs, get_member, has_records
             set_blocksize, dump_clog, read_clog, recover_file
             openb_hooks, open102, close102, get_addrs
 */
{
  if (is_stream(filename)) f = filename;
  else if (structof(filename)==char) f = vopen(filename, 1);
  else f = open(filename, (update? update : "rb"));
  if (!is_void(clogfile)) return read_clog(f, clogfile);
  if (!is_void(open102)) yPDBopen= ((open102&3)|(at_pdb_open&~3));
  else yPDBopen= at_pdb_open;
  _openb_one = (one || (structof(filename)!=string));
  for (hooks=openb_hooks ; hooks ; hooks=_cdr(hooks)) {
    if (_car(hooks)(f)) continue;
    if (has_records(f)) edit_times, f;  /* force increasing times */
    return f;
  }
  return [];
}
errs2caller, openb;

autoload, "show.i", show;
local show;
/* DOCUMENT show, f
         or show, f, pat
         or show, f, 1
     prints a summary of the variables contained in binary file F.
     If there are too many variables, use the second form to select
     only those variables whose first few characters match PAT.
     In the third form, continues the previous show command where it
     left off -- this may be necessary for files with large numbers of
     variables.
     The variables are printed in alphabetical order down the columns.
     The print function can be used to obtain other information about F.
   SEE ALSO: openb, jt, jc
 */

autoload, "collec.i", collect;
local collect;
/* DOCUMENT result= collect(f, name_string)
     scans through all records of the history file F accumulating the
     variable NAME_STRING into a single array with one additional
     index varying from 1 to the number of records.

     NAME_STRING can be either a simple variable name, or a name
     followed by up to four simple indices which are either nil, an
     integer, or an index range with constant limits.  (Note that
     0 or negative indices count from the end of a dimension.)

     Examples:
        collect(f, "xle")        -- collects the variable f.xle
        collect(f, "tr(2,2:)")   -- collects f.tr(2,2:)
        collect(f, "akap(2,-1:0,)") -- collects f.akap(2,-1:0,)
                     (i.e.- akap in the last two values of its
                            second index)

   SEE ALSO: get_times
 */

extern get_member;
/* DOCUMENT get_member(f_or_s, member_name)
     returns F_OR_S member MEMBER_NAME, like F_OR_S.MEMBER_NAME syntax,
     but MEMBER_NAME can be a computed string.  The F_OR_S may be a
     binary file or a structure instance.
   SEE ALSO: openb
 */

extern read_clog;
/* DOCUMENT file= read_clog(file, clog_name)
     raw routine to set the binary data structure of FILE according
     to the text description in the Contents Log file CLOG_NAME.
 */

func recover_file(filename, clogfile)
/* DOCUMENT recover_file, filename
         or recover_file, filename, clogfile
     writes the descriptive information at the end of a corrupted
     binary file FILENAME from its Contents Log file CLOGFILE, which
     is FILENAME+"L" by default.
 */
{
  if (is_void(clogfile)) clogfile= filename+"L";
  if (clogfile==filename+"L") {  /* open clobbers this one */
    changed= 1;
    rename, clogfile, filename+"M";
    clogfile= filename+"M";
  } else {
    changed= 0;
  }
  f= open(filename, "r+b");
  i= array(char, 12);
  _read, f, 0, i;
  read_clog, f, clogfile;
  if (string(&i)=="!<<PDB:II>>!") _set_pdb, f, at_pdb_close;
  else _init_clog, f;
  close, f;
  if (changed) remove, clogfile;
}

extern _not_pdb;
/* DOCUMENT _not_pdb(file, familyOK)
     returns 1 if FILE is not a PDB file, otherwise returns 0 after
     setting the structure and data tables, and cataloguing any
     history records.  Used to open an existing file.  Also detects
     a file with an appended Clog description.
     Before calling _not_pdb, set the variable yPDBopen to the value
     of at_pdb_open you want to be in force.  (For historical reasons
     -- in order to allow for the open102 keyword to openb -- _not_pdb
     looks at the value of the variable yPDBopen, rather than at_pdb_open
     directly.)
 */

local close102, open102, close102_default;
/* DOCUMENT close102  is a keyword for createb or updateb,
            open102   is a keyword for openb or updateb
            close102_default   is a global variable (initially 0)
              ***Do not use close102_default -- use at_pdb_close
                 -- this is for backward compatibility only***

            close102=1  means to close the PDB file "Major-Order:102"
            close102=0  means close it "Major-Order:101"
               if not specified, uses 1 if close102_default non-zero,
               otherwise the value specified in at_pdb_close

            open102=1   means to ignore what the PDB file says internally,
                        and open it as if it were "Major-Order:102"
            open102=0   (the default) means to assume the PDB file is
                        correctly writen
            open102=2   means to assume that the file is incorrectly
                        written, whichever way it is marked
            open102=3   means to ignore what the PDB file says internally,
                        and open it as if it were "Major-Order:101"

     The PDB file format comes in two styles, "Major-Order:101", and
     "Major-Order:102".  Yorick interprets these correctly by default,
     but other codes may ignore them, or write them incorrectly.

     Unlike Yorick, not all codes are able to correctly read both
     styles.  If you are writing a file which needs to be read by
     a "102 style" code, create it with the close102=1 keyword.

     If you notice that a file you though was a history file isn't, or
     that the dimensions of multi-dimensional variables are transposed
     from the order you expected, the code which wrote the file probably
     blew it.  Try openb("filename", open102=2).  The choices 1 and 3
     are for cases in which you know the writing code was supposed to
     write the file one way or the other, and you don't want to be
     bothered.

     The open102 and close102 keywords, if present, override the
     defaults in the variables at_pdb_open and at_pdb_close.

   SEE ALSO: at_pdb_open, at_pdb_close
 */
close102_default= [];

local at_pdb_open, at_pdb_close;
/* DOCUMENT at_pdb_open
            at_pdb_close
     bits for optional behavior when a PDB file is opened or closed:

     at_pdb_open:
     000  Major-Order:  value specified in file is correct
     001  Major-Order:102 always
     002  Major-Order:  opposite from what file says
     003  Major-Order:101 always

     004  Strip Basis @... suffices from variable names (when possible)
          Danger!  If you do this and open a file for update, the variable
          names will be stripped when you close the file!
     010  Use Basis @history convention on input

     The 001 and 002 bits may be overridden by the open102 keyword.
     The default value of at_pdb_open is 010.

     at_pdb_close (the value at the time the file is opened or created
                   is remembered):
     001  Write Major-Order 102 PDB file
     002  Write PDB style history data
        The following are no-ops unless bit 002 is set:
     004  Use Basis @history convention on output
     010  Do NOT pack all history record variables into
          a single structure instance.

     The 001 bit may be overridden by the close102 keyword or if
     close102_default is non-zero.
     The default value of at_pdb_close is 007.

   SEE ALSO: close102_default
 */
at_pdb_open= 010;
at_pdb_close= 007;

func _not_pdbf(f) { return _not_pdb(f, !_openb_one); }

extern _init_pdb;
extern _set_pdb;
/* DOCUMENT _init_pdb, file, at_pdb_close
            _set_pdb, file, at_pdb_close
     initializes a PDB binary file.  Used after creating a new file --
     must be called AFTER the primitive data formats have been set.
     The _set_pdb call only sets the CloseHook, on the assumption that
     the file header has already been written (as in recover_file).
   SEE ALSO: createb, recover_file, at_pdb_close
 */

extern _init_clog;
/* DOCUMENT _init_clog, file
     initializes a Clog binary file.  Used after creating a new file --
     must be called AFTER the primitive data formats have been set.
 */

extern dump_clog;
/* DOCUMENT dump_clog, file, clog_name
     dumps a Contents Log of the binary file FILE into the text file
     CLOG_NAME.  Any previous file named CLOG_NAME is overwritten.
  SEE ALSO: openb
 */

func _not_cdf(file)
/* DOCUMENT _not_cdf(file)
     is like _not_pdb, but for netCDF files.
 */
{
  i= array(char, 4);
  _read, file, 0, i;
  if (string(&i)!="CDF\001") return 1;  /* test magic number */
  require, "netcdf.i";
  if (_openb_one) _nc_open_filename = [filename];
  return raw_not_cdf(file);
}

local openb_hooks;
/* DOCUMENT openb_hooks
     list of functions to be tried by openb if the file to be
     opened is not a PDB file.  By default,
       openb_hooks= _lst(_not_pdbf, _not_cdf).
     The hook functions will be called with the file as argument
     (e.g.- _not_cdf(file)), beginning with _car(openb_hooks), until
     one of them returns 0.  Note that a hook should return 0 if it
     "recognizes" the file as one that it should be able to open, but
     finds that the file is misformatted (alternatively, it could call
     error to abort the whole process).
 */
openb_hooks= _lst(_not_pdbf, _not_cdf);

func createb(filename, primitives, close102=, clog=)
/* DOCUMENT file= createb(filename)
         or file= createb(filename, primitives)
     creates FILENAME as a PDB file in "w+b" mode, destroying any
     existing file by that name.  If the PRIMITIVES argument is
     supplied, it must be the name of a procedure that sets the
     primitive data types for the file.  The default is to create
     a file with the native primitive types of the machine on which
     Yorick is running.  The following PRIMITIVES functions are
     predefined:
        sun_primitives    -- appropriate for Sun, HP, IBM, and
                             most other workstations
        sun3_primitives   -- appropriate for old Sun-2 or Sun-3
        dec_primitives    -- appropriate for DEC (MIPS) workstations, Windows
        alpha_primitives  -- appropriate for DEC alpha workstations
        sgi64_primitives  -- appropriate for 64 bit SGI workstations
        cray_primitives   -- appropriate for Cray 1, XMP, and YMP
        mac_primitives    -- appropriate for MacIntosh
        macl_primitives   -- appropriate for MacIntosh, 12-byte double
        i86_primitives    -- appropriate for Linux i86 machines
        pc_primitives     -- appropriate for IBM PC
        vax_primitives    -- appropriate for VAXen only (H doubles)
        vaxg_primitives   -- appropriate for VAXen only (G doubles)
        xdr_primitives    -- appropriate for XDR files

     FILENAME may also be char (that is, the char datatype) in order to
     create an in-memory binary file using vopen.  Such a file must be
     closed with vclose or everything written to it will be lost.

  SEE ALSO: openb, updateb, vopen, vsave, cd
            save, add_record, set_filesize, set_blocksize
            close102, close102_default, at_pdb_open, at_pdb_close
 */
{
  file = (filename==char)? vopen(,1) : open(filename, "w+b"+(clog?"c":""));
  if (!is_void(primitives)) primitives, file;
  if (!is_void(close102)) yPDBclose= ((close102&1)|(at_pdb_close&~1));
  else if (is_void(close102_default)) yPDBclose= at_pdb_close;
  else yPDBclose= ((close102_default&1)|(at_pdb_close&~1));
  _init_pdb, file, yPDBclose;
  return file;
}
errs2caller, createb;

func sun_primitives(file)
/* DOCUMENT sun_primitives, file
     sets FILE primitive data types to be native to Sun, HP, IBM, etc.
 */
{
  set_primitives, file, __sun;
}

func sun3_primitives(file)
/* DOCUMENT sun3_primitives, file
     sets FILE primitive data types to be native to Sun-2 or Sun-3.
 */
{
  set_primitives, file, __sun3;
}

func dec_primitives(file)
/* DOCUMENT dec_primitives, file
     sets FILE primitive data types to be native to DEC (MIPS) workstations.
 */
{
  set_primitives, file, __dec;
}

func alpha_primitives(file)
/* DOCUMENT alpha_primitives, file
     sets FILE primitive data types to be native to DEC alpha workstations.
 */
{
  set_primitives, file, __alpha;
}

func sgi64_primitives(file)
/* DOCUMENT sgi64_primitives, file
     sets FILE primitive data types to be native to 64-bit SGI workstations.
 */
{
  set_primitives, file, __sgi64;
}

func cray_primitives(file)
/* DOCUMENT cray_primitives, file
     sets FILE primitive data types to be native to Cray 1, XMP, and YMP.
 */
{
  set_primitives, file, __cray;
}

func mac_primitives(file)
/* DOCUMENT mac_primitives, file
     sets FILE primitive data types to be native to MacIntosh, 8 byte double.
 */
{
  set_primitives, file, __mac;
}

func macl_primitives(file)
/* DOCUMENT macl_primitives, file
     sets FILE primitive data types to be native to MacIntosh, long double.
 */
{
  set_primitives, file, __macl;
}

func i86_primitives(file)
/* DOCUMENT i86_primitives, file
     sets FILE primitive data types to be native to Linux i86 machines.
 */
{
  set_primitives, file, __i86;
}

func pc_primitives(file)
/* DOCUMENT pc_primitives, file
     sets FILE primitive data types to be native to IBM PC.
 */
{
  set_primitives, file, __ibmpc;
}

func vax_primitives(file)
/* DOCUMENT vax_primitives, file
     sets FILE primitive data types to be native to VAXen, H-double, only.
 */
{
  set_primitives, file, __vax;
}

func vaxg_primitives(file)
/* DOCUMENT vaxg_primitives, file
     sets FILE primitive data types to be native to VAXen, G-double, only.
 */
{
  set_primitives, file, __vaxg;
}

func xdr_primitives(file)
/* DOCUMENT xdr_primitives, file
     sets FILE primitive data types to be XDR (external data representation).
 */
{
  set_primitives, file, __xdr;
}

extern get_primitives;
/* DOCUMENT prims = get_primitives(file)
     Return the primitive data types for FILE as an array of 32
     integers.  The format is described under set_primitives.
   SEE ALSO: set_primitives, __xdr, __i86
 */

func set_primitives(file, p)
/* DOCUMENT set_primitives, file, prims
     Return the primitive data types for FILE as an array of 32
     integers.  Versions for particular machines are defined in
     prmtyp.i, and can be accessed using functions like
     sun_primitives or i86_primitives.  See __xdr for a complete
     list.  The format is:
     [size, align, order] repeated 6 times for char, short, int,
       long, float, and double, except that char align is always 1,
       so result(2) is the structure alignment (see struct_align).
     [sign_address,  exponent_address, exponent_bits,
      mantissa_address, mantissa_bits,
      mantissa_normalization, exponent_bias] repeated twice for
       float and double.  See the comment at the top of prmtyp.i
       for an explanation of these fields.
     the total number of items is thus 3*6+7*2=32.
   SEE ALSO: get_primitives, createb, __xdr, __i86
*/
{
  install_struct, file, "char",    1, 1, p(3);
  install_struct, file, "short",   p(4),p(5),p(6);
  install_struct, file, "int",     p(7),p(8),p(9);
  install_struct, file, "long",    p(10),p(11),p(12);
  install_struct, file, "float",   p(13),p(14),p(15), p(19:25);
  install_struct, file, "double",  p(16),p(17),p(18), p(26:32);
  struct_align, file, p(2);
}

local __xdr;
local __vaxg;
local __vax;
local __ibmpc;
local __i86;
local __macl;
local __mac;
local __cray;
local __sgi64;
local __alpha;
local __dec;
local __sun;
local __sun3;
/* DOCUMENT primitive data types for various machines:
       little-endians
   __i86      Intel x86 Linux
   __ibmpc    IBM PC (2 byte int)
   __alpha    Compaq alpha
   __dec      DEC workstation (MIPS), Intel x86 Windows
   __vax      DEC VAX (H-double)
   __vaxg     DEC VAX (G-double)
       big-endians
   __xdr      External Data Representation
   __sun      Sun, HP, SGI, IBM-RS6000, MIPS 32 bit
   __sun3     Sun-2 or Sun-3 (old)
   __sgi64    SGI, Sun, HP, IBM-RS6000 64 bit
   __mac      MacIntosh 68000 (power Mac, Gx are __sun)
   __macl     MacIntosh 68000 (12 byte double)
   __cray     Cray XMP, YMP
   SEE ALSO: set_primitives
 */
__xdr = __i86 =
/*  sizeof, alignment, order
 *   char       short      int        long       float      double */
  [ 1, 1, 1,   2, 2, 1,   4, 4, 1,   4, 4, 1,   4, 4, 1,   8, 4, 1,
/* sign addr,  exp addr, exp len,  man addr, man len, man norm,  exp bias
 *           float                      double */
    0, 1,8,  9,23, 0,  0x7f,   0, 1,11, 12,52, 0, 0x3ff];
__i86(3:18:3) = -1;
__ibmpc = __alpha = __dec = __i86;
__ibmpc([7,8,11,14,17]) = 2;
__alpha([10,11,17]) = 8;
__dec(17) = 8;
__sun = __sun3 = __sgi64 = __mac = __xdr;
__sun(17) = 8;
__sun3(5:17:3) = 2;
__sgi64([10,11,17]) = 8;
__mac([7,8,11,14,17]) = 2;
__macl = __mac;
__macl(16) = 12;
__macl(26:32) = [0, 1,15, 32,64, 1, 0x3ffe];
__cray =
  [ 1, 1, 1,   8, 8, 1,   8, 8, 1,   8, 8, 1,   8, 8, 1,   8, 8, 1,
    0, 1,15, 16,48, 1, 0x4000,   0, 1,15, 16,48, 1, 0x4000];
__vax = __vaxg =
  [ 1, 1, -1,   2, 1, -1,   4, 1, -1,   4, 1, -1,   4, 1, 2,   8, 1, 2,
    0, 1,8,  9,23, 0,  0x81,   0, 1,8,  9,55, 0,  0x81];
__vaxg(26:32) = [0, 1,11, 12,52, 0, 0x401];

func updateb(filename, primitives, close102=, open102=, clog=)
/* DOCUMENT file= updateb(filename)
         or file= updateb(filename, primitives)
     open a binary data file FILENAME for update (mode "r+b").
     The optional PRIMITIVES argument is as for the createb function.
     If the file exists, it is opened as if by openb(filename),
     otherwise a new PDB file is created as if by createb(filename).
   SEE ALSO: openb, createb, cd, save, restore, get_vars, get_addrs
             close102, close102_default, open102, at_pdb_open, at_pdb_close
 */
{
  if (is_void(open(filename, "r", 1)))   /* "rb" does much more work */
    return createb(filename, primitives, close102=close102, clog=clog);
  else
    return openb(filename,,(clog?"r+bc":"r+b"), open102=open102);
}
errs2caller, updateb;

extern save;
extern restore;
/* DOCUMENT save, obj, var1, var2, ...
            restore, obj, var1, var2, ...
            grp = save(var1, var2, ...)
            grp = restore(var1, var2, ...)
     saves the variables VAR1, VAR2, etc. in the object OBJ, or restores
     them from that object.  An object can be a binary file handle, in which
     case there may be restrictions on the type of the VARi; in particular,
     the VARi will need to be arrays or structure definitions.  In general,
     the kind of object OBJ determines what kinds of variables can be
     saved in it.

     Called as functions, save and restore return a grp object, a very
     light weight in-memory container that can hold any kind of yorick
     variable.  In the case of save, the grp contains the the specified
     variables VARi.  For group objects (not necessarily other objects),
     the saved items are not copies, but references.  However, if you
     redefine a VARi after a save to a group object, the group member
     corresponding to that VARi does not change.  Hence, groups are a
     way to maintain "namespaces" in yorick.  The return value from
     save is simply a group object containing the VARi.  The return
     value from restore is more interesting: it is a group object containing
     the values of the VARi before they were restored.  This enables you to
     put things back the way they were before a restore, after you are
     finished using the restored variables.

     Special cases of save:
       grp = save();   // return an empty group object
       obj = save(*);  // return the entire global symbol table as an object
       save, obj;      // saves entire global symbol table in OBJ, silently
         skipping any variables whose data type OBJ does not support
     Other special cases:
       restore, obj;   // restores all named variables in OBJ
       save, use, var1, var2, ...;
       restore, use, var1, var2, ...;
         save and restore to the current context object (see help,use).
         Normally you should use the use function, not these special forms.

     Each VARi may be a simple variable reference, in which case the name
     of the VARi specifies which member of the object.  (In the case of
     save, a VARi whose name matches no current object member will create
     a new object member of that name.)  However, any of the VARi may
     instead be be a pair of arguments instead of a single argument:
       VARi -->  MEMBSPECi, VALi
     where MEMBSPECi is an expression (but NOT a simple variable reference)
     whose value specifies which object member, and the VALi argument is
     the external value.  In the case of save, VALi may also be an
     expression; in the case of restore, VALi must be the simple variable
     reference for the variable which restore will set to the specified
     object member.  For example:
       var2 = 3*x+7;
       save, obj, var1, var2, var3;
       save, obj, var1, "var2", 3*x+7, var3;
       save, obj, var1, swrite(format="var%ld",8/4), 3*x+7, var3;
     All three save calls do the same thing.  The corresponding restore
     works by name; the order need not be the same as the save:
       restore, obj, var2, var3, var1;
     puts the saved values back where they started, while:
       restore, obj, var2, swrite(format="var%ld",1), x;
     puts var2 back to its saved value, but sets x to the value saved
     as var1.  You can use the noop() function to make an expression out
     of a variable holding a MEMBSPEC.  For example, if varname="var1", then
       restore, obj, noop(varname), x;  // or
       restore, obj, varname+"", x;
     will set x to the value saved as var1, while
       restore, obj, varname, x;   // error!
     attempts to restore two variables named "varname" and "x" from obj.

     For the save function, each VARi may also be a keyword argument:
       VARi -->  member=VALi
     which behaves exactly the same as:
       VARi -->  "member",VALi
     but is slightly more efficient, since it avoids the string argument.
     You can also omit the "save" in a subroutine call if all arguments
     are keywords:
       save, obj, m1=val1, m2=val2, ...;
     is the same thing as:
       obj, m1=val1, m2=val2, ...;

     Some kinds of objects (including the group objects, but usually not
     binary file handles) support anonymous members.  For such objects,
     the order in which the members were saved is significant, and member
     names are optional.  You can create anonymous members by passing
     string(0) to save as the MEMBSPEC.  Unlike ordinary names, each save
     with string(0) as the name creates a new member (rather than overwriting
     the existing member with that name).  All members (named as well as
     anonymous) are numbered starting from 1 for the first member, in the
     order in which they are created.  For objects supporting anonymous
     members, MEMBSPEC may also be an integer, which is the member index.

     In fact, MEMBSPECi can be any of the following:
     scalar string   - member name, string(0) on save creates anonymous member
     scalar index    - member index
     string array    - VALi a group with those members (string(0) on save OK)
     index array     - VALi a group with those members
     min:max:step    - VALi a group with those members
     nil []          - save only: if VALi is not an object, same as string(0),
       if VALi is an object, merge with OBJ, that is members of VALi become
       members of OBJ, creating or overwriting named members and always
       appending anonymous members.
     MEMBSPEC indices and index ranges accept zero or negative values with
     the same meaning as for array indices, namely 0 represents the last
     member, -1 the second to the last, and so on.  Unlike array indices,
     the non-positive index values also work in index array MEMBSPECs.

     See help,oxy (object extension to yorick) for more on objects.

     As a final remark, notice that you can use save and restore to
     construct group objects without having any side effects -- that is,
     without "damaging" the state of any other variables.  For example,
     suppose we want to create an object bump consisting of three
     variables x, y, and z, that need to be computed.  In order to do
     that without clobbering existing values of x, y, and z, or anything
     else, we can do this:
       bump = save(x, y, z);        // save current values of x, y, z
       scratch = save(scratch, xy); // save scratch variables (xy and scratch)
       xy = span(-4, 4, 250);
       x = xy(,-:1:250);
       y = xy(-:1:250,);
       z = sqrt(0.5/pi)*exp(-0.5*abs(x,y)^2);
       bump = restore(bump);        // put back old x,y,z, set bump to new
       restore, scratch;            // restore xy and scratch itself

   SEE ALSO: oxy, is_obj, openb, createb, use, noop, gaccess
 */

func jt(file, time)
/* DOCUMENT jt, time
         or jt, file, time
         or jt, file
         or jt, file, -
     jump to the record nearest the specified TIME.  If no FILE is
     specified, the current record of all open binary files containing
     records is shifted.
     If both FILE and TIME are specified and jt is called as a function,
     it returns the actual time of the new current record.

   N.B.: "jt, file" and "jt, file, -" are obsolete.  Use the jr function to
     step through a file one record at a time.

     If only the FILE is specified, increment the current record of that
     FILE by one.  If the TIME argument is - (the pseudo-index range
     function), decrement the current record of FILE by one.
     If the current record is the last, "jt, file" unsets the current record
     so that record variables will be inaccessible until another jt or jc.
     The same thing happens with "jt, file, -" if the current record was the
     first.
     If only FILE is specified, jt returns 1 if there is a new current
     record, 0 if the call resulted in no current record.  Thus "jt(file)"
     and "jt(file,-)" may be used as the condition in a while loop to step
     through every record in a file:
        file= openb("example.pdb");
        do {
          restore, file, interesting_record_variables;
          ...calculations...
        } while (jt(file));

   SEE ALSO: jc, _jt, edit_times, show, jr
 */
{
  return is_void(time)? _jt(file) : _jt(file, time);
}

func jc(file, ncyc)
/* DOCUMENT jc, file, ncyc
     jump to the record of FILE nearest the specified NCYC.
   SEE ALSO: jt, _jc, edit_times, show, jr
 */
{
  return _jc(file, ncyc);
}

extern _jr;
extern _jt;
extern _jc;
/* DOCUMENT _jt, file, time
            _jc, file, ncyc
            _jr, file
     are raw versions of jt and jc provided to simplify redefining
     the default jt and jc functions to add additional features.
     For example, you could redefine jt to jump to a time, then
     plot something.  The new jt can pass its arguments along to
     _jt, then call the appropriate plotting functions.
     There is a raw version of jr as well.
 */

func jr(file, i)
/* DOCUMENT jr, file, i
         or _jr(file, i)
     Jump to a particular record number I (from 1 to n_records) in a
     binary file FILE.  The function returns 1 if such a record exists,
     0 if there is no such record.  In the latter case, no action is
     taken; the program halts with an error only if jr was invoked
     as a subroutine.  Record numbering wraps like array indices; use
     jr, file, 0  to jump to the last record, -1 to next to last, etc.
   SEE ALSO: jt, jc, edit_times, show
 */
{
  return _jr(file, i);
}

extern add_record;
/* DOCUMENT add_record, file, time, ncyc
         or add_record, file, time, ncyc, address
         or add_record, file
     adds a new record to FILE corresponding to the specified TIME and
     NCYC (respectively a double and a long).  Either or both TIME
     and NCYC may be nil or omitted, but the existence of TIME and
     NCYC must be the same for every record added to one FILE.
     If present, ADDRESS specifies the disk address of the new record,
     which is assumed to be in the current file.  Without ADDRESS, or
     if ADDRESS<0, the next available address is used; this may create
     a new file in the family (see the set_filesize function).
     The add_record function leaves the new record current
     for subsequent save commands to actually write the data.

     The TIME, NCYC, and ADDRESS arguments may be equal length vectors
     to add several records at once; in this case, the first of the
     newly added records is the current one.  If all three of TIME,
     NCYC, and ADDRESS are nil or omitted, no new records are added,
     but the file becomes a record file if it was not already, and in
     any case, no record will be the current record after such an
     add_record call.

     After the first add_record call (even if no records were added),
     subsequent add_variable commands will create record variables.
     After the first record has been added, subsequent save commands
     will create any new variables as record variables.
     After a second record has been added using add_record, neither
     save commands nor add_variable commands may be used to introduce
     any new record variables.
   SEE ALSO: save, createb, updateb, openb, set_filesize, set_blocksize
             add_variable
 */

extern add_variable;
/* DOCUMENT add_variable, file, address, name, type, dimlist
     adds a variable NAME to FILE at the specified ADDRESS, with the
     specified TYPE and dimensions given by DIMLIST.  The DIMLIST may
     be zero or more arguments, as for the "array" function.  If the
     ADDRESS is <0, the next available address is used. Note that,
     unlike the save command, add_variable does not actually write any
     data -- it merely changes Yorick's description of the contents of
     FILE.
     After the first add_record call, add_variable adds a variable to
     the record instead of a non-record variable.  See add_record.
   SEE ALSO: save, openb, createb, updateb, add_record,
             add_member, install_struct, data_align
 */

extern set_blocksize;
/* DOCUMENT set_blocksize, file, blocksize
         or set_blocksize, blocksize
     sets smallest cache block size for FILE to BLOCKSIZE.  BLOCKSIZE
     is rounded to the next larger number of the form 4096*2^n if
     necessary; cache blocks for this file will be multiples of
     BLOCKSIZE bytes long.  The default BLOCKSIZE is 0x4000 (16 KB)
     initially.  The second form, with no FILE argument, sets the
     default BLOCKSIZE.
   SEE ALSO: openb, updateb, createb, save, restore, _read, _write,
             set_cachesize
 */

extern set_cachesize;
/* DOCUMENT set_cachesize, maxBlockSize, totalCacheSize
     Sets largest cache block size to  MAXBLOCKSIZE.  MAXBLOCKSIZE
     is rounded to the next larger number of the form 4096*2^n if
     necessary.
     Sets the total cache size to TOTALCACHESIZE.  TOTALCACHESIZE
     will be set to 4*MAXBLOCKSIZE if it is smaller than that.
     The default MAXBLOCKSIZE is 0x080000 (512k) and the default
     TOTALCACHESIZE is  0x140000 (1.25 Mbytes).
   SEE ALSO: set_blocksize, openb, updateb, createb
 */

extern set_filesize;
/* DOCUMENT set_filesize, file, filesize
     sets the new family member threshhold for FILE to FILESIZE.
     Whenever a new record is added (see add_record), if the current file
     in the FILE family has at least one record and the new record would
     cause the current file to exceed FILESIZE bytes, a new family
     member will be created to hold the new record.
     Note that set_filesize must be called after the first call to
     add_record.
     The default FILESIZE is 0x800000 (8 MB).
   SEE ALSO: openb, updateb, createb, add_record
 */

extern get_vars;
/* DOCUMENT name_lists= get_vars(file)
     returns the lists of non-record and record variable names in the
     binary FILE.  The return value is an array of two pointers to
     arrays of type string; *name_lists(1) is the array of non-record
     variable names (or nil if there are none), *name_lists(2) is the
     array of record variable names.
     The get_addrs function returns corresponding lists of disk
     addresses; the get_member function can be used in conjunction
     with the dimsof, structof, and typeof functions to determine
     the other properties of a variable.
   SEE ALSO: openb, updateb, restore, jt, jc, has_records, get_addrs,
             set_vars
 */

extern set_vars;
/* DOCUMENT set_vars, file, names
         or set_vars, file, nonrec_names, rec_names
     Change the names of the variables in FILE to NAMES.  If the
     file has record variables, you can use the second form to change
     the record variable names.  Either of the two lists may be nil
     to leave those names unchanged, but if either is not nil, it must
     be a 1D array of strings whose length exactly matches the number
     of that type of variable actually present in the file.
   SEE ALSO: openb, updateb, has_records, get_vars
 */

extern get_addrs;
/* DOCUMENT addr_lists= get_addrs(file)
     returns the byte addresses of the non-record and record variables
     in the binary file FILE, and lists of the record addresses, file
     indices, and filenames for file families with history records.
          *addr_lists(1)   absolute addresses of non-record variables
          *addr_lists(2)   relative addresses of record variables
                           (add record address to get absolute address)
             The order of these two address lists matches the
             corresponding lists of names returned by get_vars.
          *addr_lists(3)   absolute addresses of records
          *addr_lists(4)   list of file indices corresponding to
                           addr_lists(3); indices are into addr_lists(5)
          *addr_lists(5)   list of filenames in the family
   SEE ALSO: openb, updateb, restore, jt, jc, has_records, get_vars
 */

func has_records(file)
/* DOCUMENT has_records(file)
     returns 1 if FILE has history records, 0 if it does not.
 */
{
  return get_vars(file)(2)? 1n : 0n;
}

extern get_times;
extern get_ncycs;
/* DOCUMENT times= get_times(file)
            ncycs= get_ncycs(file)
     returns the list of time or ncyc values associated with the records
     if FILE, or nil if there are none.  The time values are not guaranteed
     to be precise (but they should be good to at least 6 digits or so);
     the precise time associated with each record may be stored as a record
     variable.
   SEE ALSO: collect, openb, updateb, restore, jt, jc, edit_times
 */

extern edit_times;
/* DOCUMENT edit_times, file
         or edit_times, file, keep_list
         or edit_times, file, keep_list, new_times, new_ncycs
     edits the records for FILE.  The KEEP_LIST is a 0-origin index list
     of records to be kept, or nil to keep all records.  The NEW_TIMES
     array is the list of new time values for the (kept) records, and
     the NEW_NCYCS array is the list of new cycle number values for the
     (kept) records.  Either NEW_TIMES, or NEW_NCYCS, or both, may be
     nil to leave the corresponding values unchanged.  If non-nil,
     NEW_TIMES and NEW_NCYCS must have the same length as KEEP_LIST,
     or, if KEEP_LIST is nil, as the original number of records in
     the file.  If KEEP_LIST, NEW_TIME, and NEW_NCYCS are all omitted
     or nil, then edit_times removes records as necessary to ensure
     that the remaining records have monotonically increasing times,
     or, if no times are present, monotonically increasing ncycs.
     (The latest record at any given time/ncyc is retained, and earlier
     records are removed.)
     In no case does edit_times change the FILE itself; only Yorick's
     in-memory model of the file is altered.
   SEE ALSO: get_times, get_ncycs, jt, jc
 */

extern _read;
extern _write;
/* DOCUMENT _write, file, address, expression
            _read, file, address, variable
         or nbytes= _read(file, address, variable);
     are low level read and write functions which do not "see" the
     symbol table for the binary FILE.  The ADDRESS is the byte address
     at which to begin the write or read operation.  The type and number
     of objects of the EXPRESSION or VARIABLE determines how much data
     to read, and what format conversion operations to apply.  In the
     case of type char, no conversion operations are ever applied, and
     _read will return the actual number of bytes read, which may be
     fewer than the number implied by VARIABLE in this one case.
     (In all other cases, _read returns numberof(VARIABLE).)
     If the FILE has records, the ADDRESS is understood to be in the
     file family member in which the current record resides.
   SEE ALSO: openb, createb, updateb, save, restore, sizeof
 */

extern add_member;
/* DOCUMENT add_member, file, struct_name, offset, name, type, dimlist
     adds a member to a data type in the file FILE.  The data type name
     (struct name) is STRUCT_NAME, which will be created if it does
     not already exist.  The new member will be at OFFSET (in bytes)
     from the beginning of an instance of this structure, and will
     have the specified NAME, TYPE, and DIMLIST.  Use OFFSET -1 to
     have add_member compute the next available offset in the structure.
     The TYPE can be either a structure definition, or a string naming
     a previously defined data type in FILE.  The optional DIMLIST is
     as for the "array" function.
     The STRUCT_NAME built from a series of add_member calls cannot be
     used until it is installed with install_struct.
     This function should be used very sparingly, mostly in code which
     is building the structure of a foreign-format binary file.
   SEE ALSO: add_variable, install_struct, struct_align
 */

extern install_struct;
/* DOCUMENT install_struct, file, struct_name
         or install_struct, file, struct_name, size, align, order
         or install_struct, file, struct_name, size, align, order, layout
     installs the data type named STRUCT_NAME in the binary FILE.  In
     the two argument form, STRUCT_NAME must have been built by one or
     more calls to the add_member function.  In the 5 and 6 argument calls,
     STRUCT_NAME is a primitive data type -- an integer type for the 5
     argument call, and a floating point type for the 6 argument call.
     The 5 argument form may also be used to declare opaque data types.
     SIZE is the size of an instance in bytes, ALIGN is its alignment
     boundary (also in bytes), and ORDER is the byte order.  ORDER is
     1 for most significant byte first, -1 for least significant byte
     first, and 0 for opaque (unconverted) data.  Other ORDER values
     represent more complex byte permutations (2 is the byte order for
     VAX floating point numbers).  If ORDER equals SIZE, then the data
     type is not only opaque, but also must be read sequentially.
     LAYOUT is an array of 7 long values parameterizing the floating
     point format, [sign_address, exponent_address, exponent_size,
     mantissa_address, mantissa_size, mantissa_normalized, exponent_bias]
     (the addresses and sizes are in bits, reduced to MSB first order).
     Use, e.g., nameof(float) for STRUCT_NAME to redefine the meaning
     of the float data type for FILE.
   SEE ALSO: add_variable, add_member
 */

extern data_align;
/* DOCUMENT data_align, file, alignment
     in binary file FILE, align new variables to begin at a byte address
     which is a multiple of ALIGNMENT.  (This affects placement of data
     declared using save and add_variable.  For add_variable, data_align
     has an effect only if the address is not specified.)  If ALIGNMENT
     is <=0, new variables will be aligned as they would be if they were
     data structure members.  The default value is 0.
   SEE ALSO: save, add_variable
 */

extern struct_align;
/* DOCUMENT struct_align, file, alignment
     in binary file FILE, align new struct members which are themselves
     struct instances to begin at a byte address which is a multiple of
     ALIGNMENT.  (This affects members declared explicitly by add_member,
     as well as implicitly by save or add_variable.)  If ALIGNMENT is <=0,
     returns to the default for this machine.  The struct alignment is in
     addition to the alignment implied by the most restrictively aligned
     member of the struct.  Most machines want ALIGNMENT of 1.
   SEE ALSO: add_member
 */

extern add_next_file;
/* DOCUMENT failure= add_next_file(file, filename, create_flag)
     adds the next file to the FILE, which must contain history records.
     If FILENAME is non-nil, the new file will be called that, otherwise
     the next sequential filename is used.  If CREATE_FLAG is present
     and non-zero, the new file will be created if it does not already
     exist.  If omitted or nil, CREATE_FLAG defaults to 1 if the file has
     write permission and 0 if it does not.
     Returns 0 on success.
   SEE ALSO: openb, updateb, createb, add_record
 */

/*= SECTION(oxy) object oriented extensions to yorick ======================*/

local oxy;
/* DOCUMENT oxy
     Object extension to yorick.  Various yorick packages may create
     "objects", which are collections of data and methods (functions)
     for operating on that data.  Yorick objects are much more free-form
     than other object-oriented languages, in keeping with the fact that
     yorick has no declarative statements.

     Yorick objects have zero or more members; members may be anonymous
     or named.  For objects supporting anonymous members (an optional
     feature), all members whether anonymous or named can be accessed
     by a 1-origin index, as if the members were a 1D array.  Object
     indexing has unusual semantics, similar to the semantics of the
     arguments to the save and restore commands, in which it is the
     name of the variable passed as an argument to the object, rather
     than the value of the variable, which determines which member is
     to be extracted.  For example,
       obj(i)
     refers to the member of obj named "i", whether the value of i is
     1 or 100 or "j" or sqrt(3)+2i or span(0,1,200).  However, by passing
     an expression, rather than a simple variable reference, you can
     make the value of the argument (now that it has no name) be significant.
     Hence, obj("i") is also member "i", while obj(7) is the 7th member,
     obj(3*n+2) is the 3*n+2nd member, and so on.  You can use the noop()
     function to make a variable into an expression, if the member specifier
     happens to be stored in a variable:
       obj(noop(membspec))

     Objects also accept some special arguments:
       obj()   returns the whole object (same as without the parens)
       obj(*)  returns the number of members
       obj(*,) returns an array of member names (string(0) for anonymous)
                 in the index order (if the object supports indexing)
       obj(*,m)  returns an array of the specified member names, or the
                 specified member indices if M is a string array
                 (if the object supports indexing)
       obj(..) returns the attribute object associated with obj, which
                 may be an empty object, or nil [] if obj does not support
                 attributes

     When called as a subroutine, objects accept keyword arguments as a
     shorthand for the save command:
       obj, m1=val1, m2=val2, ...;
     is the same as:
       save, obj, m1=val1, m2=val2, ...;

     Yorick has a generic object, called a "group", which holds an arbitrary
     collection of yorick variables.  You make group objects with the save
     function (see help,save), and you can also use save to add members to
     an existing group object.  Another way to create a group object is by
     passing a membspec argument to any group which specifies multiple
     members -- the result will be a group containing all the specified
     members.  Hence, membspec may be an array of strings, an array of
     indices, or an index range min:max:step, in order to produce a group
     object holding all the specified members.  (Note that dimensionality
     of membspec arrays is lost, and potentially not even number is
     preserved if a single named member is specified multiple times.)

     When you pass more than one argument to an object, the first one
     specifies a single member, and subsequent arguments apply to that
     member.  Thus,
       obj(m, i, j, k)    is similar to     obj(m)(i, j, k)
     In fact, these are exactly the same as long as M does not specify
     a function.  When M is a function, there is a slight difference,
     which is that the function (whether built-in or interpreted) is
     executed in the context of the object obj.  Unlike classic object
     oriented languages, in yorick the use of the context object is not
     automatic -- the function M must specify which object members it
     wishes to access.  You do that with the use function:
       func method(x, y, z) {
         extern var1, var2, var3;
         use, var1, var2, var3;   // initializes vari from context object
         // compute using var1, var2, var3, possibly redefining them
         return result;
         // just before return, any
         // changes to var1, var2, var3 stored back to context object
         // if this function was called by another method by the
         // use_method function, the store back to the context object
         // is deferred until the outermost method returns
       }
     In this form, the arguments to the use call must be external variables
     to the function (method).  If you put the use call(s) at the top of the
     function, you can dispense with the explicit extern statement, provided
     you are careful to check that none of the names matches any of the
     dummy parameter names (x, y, or z here).  [Eventually, the yorick
     parser may treat use specially and enforce this restriction.  For now,
     you need to be sure that any arguments are external to avoid incorrect
     behavior.  Specifically, if any of the vari is local, the external
     variable of that name will be set to nil (or the actual argument value
     if vari is a dummy parameter) when the method returns.  Ouch.]
     Although the vari look like they are extern to the method function,
     use saves their external values and arranges to replace them when the
     function returns, so they behave as if they were local to method.

     Calling use() as a function, or using a special form of restore
     provides a means for direct read-only access to a member.  For example,
     suppose you are going to modify var1 and var3, but merely read var2
     and var4 in the method context:
       func method(x, y, z) {
         use, var1, var3;   // initializes vari from context object
         local var2, var4;  // otherwise, restore arguments would be extern
         restore, use, var2, var4;
         // compute using vari
         var1 = something(var2);   var3 = something(var4);
         return result;
       }
     Or equivalently,
       func method(x, y, z) {
         use, var1, var3;   // initializes vari from context object
         // compute using vari
         var1 = something(use(var2));   var3 = something(use(var4));
         return result;
       }
     As a function, use(arg1, arg2, ...) is exactly the same as
     obj(arg1, arg2, ...), where obj is the context object for the function.
     You have to be careful not to mix this form of read-only access with
     the read-write access of use called as a subroutine: If you have changed
     the value of a variable, use(var) will pick up the unchanged value
     still stored in the context object.

     You can a invoke method function M as a subroutine as well:
       obj, m, i, j, k;
     Inside a method function, you can invoke other methods with the
     use_method function:
       use_method, sibling_method, arg1, arg2;
         or
       x = use_method(sibling_method, arg1, arg2);
     where sibling_method is another method in the same context as the
     caller.  Called as a function, use_method is very nearly the same
     as use; you could write the second line as:
       x = use(sibling_method, arg1, arg2);
     (See help,use for an explanation of the differences.)

     The obj(method,arglist) syntax permits the method argument to be
     an expression whose value is a function.  This allows you to
     invoke functions which are not members of obj in the context of
     obj.  This is useful in two situations: First, you can define
     "friend" functions in C++ parlance, which are not a part of the
     object, but nevertheless understand some part of its structure.
     Second, this is the only way you can call a function in a base
     class which you have shadowed by a virtual function in one of
     its derived classes.  For example, suppose the base_interface
     object contains the method functions f1, f2, f3, etc., and is
     used to derive an object with
       obj = save(,base_interface, data1, data2);
       save, obj, f2;
     where f2 is a different function than the one in base_interface.
     Inside the function f2, you can call the one in base_interface
     like this:
       func f2(a, b) {
         use, data1, data2;
         x = use_method(base_interface(f2), a, b);
         return x + data1 + data2; // or any other expression
       }

     This stripped down facility lets you do most of the things (except
     arguably type checking) other object oriented languages feature,
     although it is a little difficult to see how to do this at first.
     For example, you can think of a "class" as the constructor function
     which makes instances:
       func myclass(data1, data2, ...) {
         // build and return the class, for example:
         return save(method1, method2, ..., data1, data2, ...);
       }
       func method1(x) { return something; }
       func method2(x,y) { return something; }
     You make an instance with:
       mything = myclass(d1, d2);
     Then you can use your object: mything(method2,x,y) and so on.

     This is slightly untidy, because you have to worry about never
     colliding with the method names.  So you could bundle it up neatly
     by making myclass itself an object containing all its methods and
     constructor(s):
       myclass = save(new, method1, method2, ...);  // save old values
       func new(data1, data2, ...) {
         // build and return the class, for example:
         return save(method1, method2, ..., data1, data2, ...);
       }
       func method1(x) { return something; }
       func method2(x,y) { return something; }
       myclass = restore(myclass);   // swap new values into myclass
     (See help,save for examples of this trick.)  Now the only variable
     you have to worry about not clobbering is myclass, and you create
     your object with:
       mything = myclass(new, d1, d2);
     and use it as before.

     There are many ways to handle inheritance (multiple inheritance
     is no harder); the simplest is just to use the save function to
     concatentate new members onto the base class.  [Probably should
     illustrate a good style here...]  The complete absence of type
     checking is actually an advantage here: If an object has a method
     with the required arguments and semantics, it will be usable no
     matter what other members it has.

     You can also extract object members using the dot operator:
       obj.member    is a synonym for    obj("member")
     but note that obj.member(args) may be very different from
     obj(member,args); in general you are better off not using the
     dot operator with objects.  In particular, note that
       obj.member = value;   // ERROR!
     does NOT set the member to value.  (Use obj,member=value;)

   SEE ALSO: use, save, restore, is_obj, openb, createb, noop, closure,
             gaccess
 */

extern is_obj;
/* DOCUMENT is_obj(x)
 *       or is_obj(x,m)
 *       or is_obj(x,m,errflag)
 *   returns 1 if X is an object, else 0.  If X is an object which permits
 *   numerical indexing of its members, returns 3.  With second parameter M,
 *   query is for member M of object X.  If M specifies multiple members
 *   (an index range, index list, or list of member names), then returns
 *   an array of results.  Note that is_obj(x,) may return [] if the
 *   object X is empty.  With third parameter ERRFLAG non-nil and non-zero,
 *   is_obj will return -2 if X is not an object, and -1 wherever M
 *   specifies a non-member (either a name not present or an index out
 *   of range).  Without ERRFLAG, is_obj raises an error when M is not
 *   a member.
 * SEE ALSO: oxy, save, restore
 */

extern use;
extern use_method;
/* DOCUMENT use, var1, var2, ...
         or use(membspec, arg1, arg2, ...)
         or use_method, memberspec, arg1, arg2, ...
         or use_method(memberspec, arg1, arg2, ...)
     Access the context object in an object method function (see help,oxy).

     In the first form, the VARi must be extern to the calling function,
     and you get read-write access to the VARi.  That is, if you redefine
     any of the VARi, your changes will be saved back to the context object
     when the calling function returns.  Even though the VARi are external
     to the method function, use arranges for their external values to be
     replaced when the calling function returns, just as if they had been
     local variables.

     The second form is equivalent to obj(membspec, arg1, arg2, ...),
     where obj is the context object.  You can use this whenever you need
     only read access to membspec.

     The third form permits you to call a method member of the context
     as a subroutine (like "noop,use(memberspec,arg1,arg2,...)").
     The fourth form is similar to the second -- use_method(...) is
     nearly a synonym for use(...), with two exceptions:
     1. use_method(memberspec) invokes the memberspec function with
        zero arguments, while use(memberspec) returns the function
        rather than invoking it.
     2. use_method raises an error if memberspec is not a function,
        while use does not (for example, memberspec could be an array).

     When you invoke a method with use_method (or use as a function),
     any calls to the subroutine form of use (the first form) will
     be cumulative with calls made by the caller.  When a method invoked
     by use_method (or use as a function) returns, the VARi from any
     use subroutine calls it has made are *not* restored when it returns.
     Instead, the restore and the update of the context object itself
     are deferred until the outermost method returns -- the one that
     was not invoked by use_method.

     Therefore, you should generally use the first form for access to
     data members of an object, and the other forms for access to
     function (method) members of objects.  If you violate this rule,
     your method code may have unexpected side effects, unless you can
     guarantee that it your method will never be invoked by another
     method by means of use_method (or use as a function).

     Additionally, you can use the special forms of the save and
     restore function to explicitly save and restore variables from the
     context object:
       restore, use, var1, var2, ...;
       save, use, var1, var2, ...;
     The use function is normally the better choice; the special forms
     of save and restore are for slightly higher performance in unusual
     situations.  If you mix save,use,var or restore,use,var with use,var,
     you will very likely have trouble with different versions of var
     overwriting each other unintentionally.

   SEE ALSO: oxy, save, restore, openb, createb, noop, closure
 */

extern closure;
/* DOCUMENT f = closure(function, data)
         or f = closure(object, member)
     creates a closure function from FUNCTION and DATA.  Invoking the
     closure function invokes FUNCTION with argument DATA prepended
     to the argument list passed to the closure function.  For example,
        f;        is equivalent to    FUNCTION, DATA;
        f(a1,a2)  is equivalent to    FUNCTION(DATA, a1, a2)
     and so on.  When the first argument is an OBJECT, and the second
     argument a MEMBER, the object is invoked with the member as its
     first parameter.  (Typically the member would be a function.)
     If the first argument is an object, then the second argument
     (MEMBER) has the same semantics as object(member, ...): namely,
     if MEMBER is a simple variable reference, its value is ignored,
     and only its name is used to specify which member of the object.
     Hence, if you want the value of MEMBER to be used, you need to
     be sure the second argument to closure is an expression, such
     as noop(member).  If the first argument is a function, then
     the second argument is always a value.

     Finally, the first argument can be a string in order to specify
     that the returned closure object use the value of the named
     variable at runtime for the function (or object).  If you expect
     the value to be an object, prefix the variable name by "o:" in
     order to prevent the closure object from keeping a use of the
     value of the second argument, if MEMBER is a simple variable
     reference.  For example,
       f = closure("myfunc", data);     // keeps use of data value
       g = closure("o:myobj", member);  // ignores value of member
       h = closure("myobj", noop(member));  // o: unnecessary
     This feature is primarily an aid during debugging; typically
     you would remove the quotes (and o:) once the code was working.

     You can query a closure object using the member extraction
     operator:
       f.function  returns function or object
       f.data      returns data or member
       f.function_name  returns name of function or object
       f.data_name      returns name of data
     The names are string(0) if unknown; function_name is known
     if and only if the first argument to closure was a string;
     data_name is known only if the first argument was a string or
     an object and the second argument a simple variable reference.

   SEE ALSO: oxy, is_func, use
 */

extern gaccess;
/* DOCUMENT flags = gaccess(grp)
 *       or grp = gaccess(grp, flags)
 *   With single GRP argument, return current group object access flags.
 *   With second FLAGS argument, set group object access flags, returning
 *   the input GRP to allow constructs like g=gaccess(save(var1,var2), 3);
 *   The access flags bits are:
 *     1  set if no new members may be created
 *     2  set if existing members cannot change data type or dimensions
 *          (that is, they behave as x(..)=expr, rather than as x=expr)
 * SEE ALSO: oxy, save, restore
 */

/*= SECTION(system) interacting with system ================================*/

extern quit;
/* DOCUMENT quit
     Exit YMainLoop when current task finishes.
     Normally this terminates the program.
 */

extern system;
/* DOCUMENT system, "shell command line"
     Passes the command line string to a shell for execution.
     If the string is constant, you may use the special syntax:
         $shell command line
     (A long command line may be continued by ending the line with \
     as usual.)  The system function syntax allows Yorick to compute
     parts of the command line string, while the simple $ escape
     syntax does not.  In either case, the only way to get output
     back from such a command is to redirect it to a file, then
     read the file.  Note that Yorick does not regain control
     until the subordinate shell finishes.  (Yorick will get control
     back if the command line backgrounds the job.)
     WARNING: If Yorick has grown to a large size, this may crash
     your operating system, since the underlying POSIX fork function
     first copies all of the running Yorick process before the exec
     function can start the shell.  See Y_SITE/sysafe.i for a fix.
   SEE ALSO: popen
 */

extern yorick_init;
/* xxDOCUMENT yorick_init
     Re-initializes all of the built-in functions for this version
     of Yorick.  To be used in desperation if you overwrite some
     critical built-in function by mistake.  Of course, if you
     redefine yorick_init, you won't be able to recover anything.
 */

extern set_path;
/* DOCUMENT set_path, "dir1:dir2:dir3:..."
         or set_path
     sets the include file search path to the specified list of
     directories.  The specified directories are searched left to
     right for include files specified as relative file names in
     #include directives, or to the include or require functions.
     If the argument is omitted, restores the default search path,
     ".:~/yorick:~/Yorick:Y_SITE/i:Y_SITE/contrib:Y_SITE/i0:Y_HOME/lib",
     where y_site is the main Yorick directory for this site.
     The Y_LAUNCH directory is the directory which contains the
     executable; this directory is omitted if it is the same as
     Y_SITE.

     Only the "end user" should ever call set_path, and then only in
     his or her custom.i file, for the purpose of placing a more
     elaborate set of personal directories containing Yorick procedures.
     For example, if someone else maintains Yorick code you use, you
     might put their ~/yorick on your include path.

   SEE ALSO: Y_LAUNCH, Y_SITE, include, require, get_path
 */

extern get_path;
/* DOCUMENT get_path()
     returns the current include file search path.
   SEE ALSO: set_path, get_pkgnames, split_path
 */

extern set_site;
/* xxDOCUMENT set_site, site_directory
     sets Y_LAUNCH, Y_SITE as a side effect.  Should only be called from
     paths.i.  See paths.i.  */

extern yorick_stats;
/* DOCUMENT yorick_stats
     returns an array of longs describing Yorick memory usage.
     For debugging.  See ydata.c source code.
 */

extern cd;
/* DOCUMENT cd, directory_name
         or cd(directory_name)
     change current working directory to DIRECTORY_NAME, returning
     the expanded path name (i.e.- with leading environment variables,
     ., .., or ~ replaced by the actual pathname).  If called as a
     function, returns nil to indicate failure, otherwise failure
     causes a Yorick error.
   SEE ALSO: lsdir, mkdir, rmdir, get_cwd, get_home, get_env, get_argv
 */

extern lsdir;
/* DOCUMENT files = lsdir(directory_name)
         or files = lsdir(directory_name, subdirs)
     List DIRECTORY_NAME.  The return value FILES is an array of
     strings or nil; the order of the filenames is unspecified;
     it does not contain "." or "..".  If present, SUBDIRS must be
     a simple variable reference, and is set to a list of subdirectory
     names (or nil if none).  If SUBDIRS is not present (first form),
     the return value of lsdir includes both files and subdirectories.
     If DIRECTORY_NAME does not exist or is not a directory, the return
     value is the integer 0 rather than nil.  Hence:
       files = lsdir(dirname, subdirs);
       if (structof(files) == long) {
         directory does not exist
       } else {
         for (i=1 ; i<=numberof(files) ; ++i) {...use files(i)...}
         for (i=1 ; i<=numberof(subdirs) ; ++i) {...use subdirs(i)...}
       }
   SEE ALSO: cd, mkdir, rmdir, get_cwd, get_home, filepath
 */

extern mkdir;
extern rmdir;
/* DOCUMENT mkdir, directory_name
         or rmdir, directory_name 
     Create DIRECTORY_NAME with mkdir, or remove it with rmdir.  The rmdir
     function only works if the directory is empty.  An error is raised if
     DIRECTORY_NAME is not a non-nil scalar string.  If mkdir or rmdir are
     called as subroutines and the operation fails, no error is raised
     (so you can use this form even when the directory already exists for
     mkdir, or already is missing for rmdir).  Otherwise, if
     DIRECTORY_NAME is a non-nil scalar string and if mkdir and rmdir are
     called as a function, they return an integer: 0 to indicate success and
     -1 to indicate failure.
     
   SEE ALSO: mkdirp, cd, lsdir, get_cwd, get_home, filepath
 */

func mkdirp(dir)
/* DOCUMENT mkdirp, directory_name
     Create DIRECTORY_NAME, creating any missing parent directories
     (like UNIX utility mkdir -p).  Unlike mkdir, signals error if
     the creation is unsuccessful.  If DIRECTORY_NAME already exists
     and is a directory, mkdirp is a no-op.
   SEE ALSO: mkdir
 */
{
  dir = strtrim(dir, 2, blank="/") + "/";
  list = strfind("/", dir, n=1024);  /* assume <1024 components in dir */
  i = list(1:-1:2);
  list = i(where((i>0) & (list(2:0:2)>0)));
  for (i=numberof(list) ; i>=1 ; i--) {
    name = strpart(dir, [0,list(i)]);
    if (lsdir(name) != 0) break;
  }
  for (i++ ; i<=numberof(list) ; i++)
    mkdir, strpart(dir, [0,list(i)]);
  if (lsdir(dir) == 0) error, "mkdirp: failed to create "+dir;
}

extern get_cwd;
extern get_home;
/* DOCUMENT get_cwd()
         or get_home()
     returns the pathname of the current working directory or of your
     home directory.
   SEE ALSO: cd, lsdir, get_env, get_argv
 */

extern get_env;
/* DOCUMENT get_env(environment_variable_name)
     returns the environment variable (a string) associated with
     ENVIRONMENT_VARIABLE_NAME (calls ANSI getenv routine).
   SEE ALSO: cd, get_cwd, get_home, get_env, get_argv
 */

extern get_argv;
/* DOCUMENT get_argv()
     returns string array containing the argv from the command line.
     The -batch and batch_include.i arguments are removed (not returned).
   SEE ALSO: process_argv, cd, get_cwd, get_home, get_env, batch
 */

func process_argv(msg)
/* DOCUMENT remaining= process_argv()
         or remaining= process_argv("your startup message")
     Performs standard command line processing.  This function is
     invoked by the default custom.i file (in $Y_SITE/i); you
     can also invoke it from your personal ~/yorick/custom.i file.
     The process_argv calls get_argv, removes any arguments of
     the form "-ifilename" or "-i filename" (the latter is a pair of
     arguments.  It returns any arguments not of this form as its
     result, after including any filenames it found in the order
     they appeared on the command line.
     The optional string argument may be an array of strings to print
     a multi-line message.

     A Yorick package may define the function get_command_line in
     order to feed process_argv something other than get_argv.

   SEE ALSO: batch
 */
{
  if (get_command_line == process_argv) return command_line;
  if (is_void(get_command_line)) command_line = get_argv();
  else command_line = get_command_line();
  get_command_line = process_argv;  /* try to avoid infinite loops */
  if (numberof(command_line)>=2) {
    command_line = command_line(2:);
    mask = (strpart(command_line, 1:2) == "-i");
    j = (command_line(0) == "-i");
    if (j) mask(0) = 0;
    list = where(mask);
    n = numberof(list);
    if (n) {
      file = strpart(command_line(list), 3:);
      i = where(file == "");
      if (numberof(i)) {
        list = list(i) + 1;
        file(i) = command_line(list);
        mask(list) = 1;
      }
      /* push onto stack in reverse order, to include in given order */
      for (i=n ; i>=1 ; --i) include, file(i);
    }
    if (j) mask(0) = 1;
    command_line= command_line(where(!mask));
  } else {
    command_line= [];
  }
  mask = (command_line == "-q");
  if (noneof(mask)) {
    if (is_void(msg)) {
      v = Y_VERSION;
      msg = [
" Copyright (c) 2005.  The Regents of the University of California.",
" All rights reserved.  Yorick "+v+" ready.  For help type 'help'"];
    }
    write, msg, format="%s\n";
  } else if (numberof(command_line)) {
    command_line = command_line(where(!mask));
  }
  return command_line;
}

extern batch;
/* DOCUMENT batch, 1
            batch, 0
            batch()
     turns on, turns off, or tests for batch mode, respectively.
     If yorick is started with the command line:
        yorick -batch batch_include.i ...
     then batch mode is turned on, the usual custom.i startup file is
     skipped, and the file batch_include.i is parsed and executed.  The
     -batch and batch_include.i command line arguments are removed from
     the list returned by get_argv().  These must be the first two
     arguments on the command line.

     In batch mode, any error will terminate Yorick (as by the quit
     function) rather than entering debug mode.  Also, any attempt to
     read from the keyboard is an error.

   SEE ALSO: process_argv, get_argv, set_idler, after_error
 */

extern set_idler;
/* DOCUMENT set_idler, idler_function
         or set_idler, idler_function, errflags
     sets the idler function to IDLER_FUNCTION.  Instead of waiting
     for keyboard input when all its tasks are finished, the interpreter
     will invoke IDLER_FUNCTION with no arguments.  The idler function
     is normally invoked only once, so input from the keyboard resumes
     after one call to the idler.  Of course, an idler is free to call
     set_idler again before it returns, which will have the effect of
     calling that function in a loop.
     If present, the ERRFLAGS argument changes the way errors are processed:
       0  - default processing, add any combination of:
       1  - suppress printing error messages
       2  - append [pc] relative program counter to function name in error
            message (use disassemble to find corresponding instruction)
       4  - call any after_error function in dbug mode (rather than clearing
            stack), so it is responsible for calling dbexit
   SEE ALSO: batch, maybe_prompt, after, after_error
 */

local after_error;
/* DOCUMENT after_error = error_handler_func
     If the variable AFTER_ERROR is set to an interpreted function
     with no parameters, that function will be invoked after an error,
     before the next prompt, instead of entering or offering to enter
     debug mode.  The error message will be printed, and also will be
     stored in the catch_message variable.  A fault during the execution
     of the after_error function will not invoke after_error, but
     otherwise after_error is persistent (unlike set_idler).  An error
     resets any functions scheduled using after or set_idler, so the
     after_error function must reschedule these if necessary.
     The catch function is a more appropriate way to recover from
     some errors.
   SEE ALSO: set_idler, catch, after
 */

extern maybe_prompt;
/* DOCUMENT maybe_prompt
     Issue prompt for keyboard input if appropriate.
     This command only makes sense (I think) as the final statement
     of a function invoked as an idler (via set_idler), when yorick is
     in a loop with an idler function that continuously re-installs
     itself.  Yorick ordinarily issues a prompt only just before it
     stops to wait for keyboard input, it will never prompt in this
     situation, even though it would accept keyboard input if it
     were typed.
   SEE ALSO: set_idler
*/

extern spawn;
/* DOCUMENT process = spawn(argv, on_stdout)
         or process = spawn(argv, on_stdout, on_stderr)
     starts the process named in ARGV(1) with additional arguments
     in any subsequent elements of ARGV (which is a scalar or 1D
     array of strings).  The ON_STDOUT and optional ON_STDERR
     are interpreted functions declared like this:
       func ON_STDOUT(msg)
       {
         commands to process msg on stdout from process
       }

     Yorick will invoke ON_STDOUT asynchronously if process
     emits text to its stdout.  Yorick includes the process in
     the list of event sources, which it polls whenever it waits
     for input.  If the optional ON_STDERR is provided, it is
     called asynchronously whenever process emits a line to stderr;
     with no ON_STDERR, the process will share yorick's stderr,
     which generally means the process stderr prints at the terminal.
     (Note that you can make the third argument the same as the second
     if you want to use the same function to handle stdout and stderr.)
     When the process terminates, ON_STDOUT is invoked with
     string(0) and the process object becomes inactive.  Note that
     ON_STDOUT and ON_STDERR are invoked via the name they were
     originally defined with (in the func or extern statement for
     interpreted and compiled functions, respectively).

     The object returned by spawn, process, can be used to send input
     or signals to the process:
       process, msg;
     where msg is a string, sends msg to the process's stdin.
       process, signum;
     sends process the specified signal (e.g.- signum=2 sends SIGINT,
     like hitting control-C, while signum=9 kills the process), if
     signum is an integer (as opposed to a string).  (Normally you
     should not send signals to a process.)  If you redefine the
     final reference to process, for example by
       process = [];
     yorick will disconnect from the process, closing its end of
     the stdin, stdout, and, optionally, stderr pipes.  For many
     programs, this will stop the program, but if the program can
     continue running without stdin and stdout, it will continue
     running.  (If yorick were a shell, the process would be running
     in the background; if the process would live beyond the shell
     which created it, it will also survive its process variable
     being freed.)

     Note: funcdef may be extremely useful for writing ON_STDOUT.

   SEE ALSO: popen, system, suspend, funcdef, after, spawn_callback
*/

func spawn_callback(&prev, line)
/* DOCUMENT spawn_callback -->
            func on_stdout(msg) {
              extern fragment; <or otherwise manage fragment>
              lines = spawn_callback(fragment, msg);
              for (i=1 ; i<=numberof(lines) ; i++) {
                line = lines(i);
                if (!line) {
                  <handle process has died>
                } else {
                  <handle complete line emitted by process>
                }
              }
            }
     Here is a template for a callback function to be passed to spawn.
     The spawn_callback function buffers any fragmentary lines,
     delivering only complete lines as output.  Note that FRAGMENT
     must somehow be managed between calls to on_stdout; it should
     be intialized to [] before calling spawn.

   SEE ALSO: spawn
 */
{
  dead = !line;  /* spawned process has died */

  /* must be prepared for process output to dribble back a fraction of
   * a line at a time, or multiple lines at a time
   * prev holds the most recent incomplete line,
   *   assuming the the remainder will arrive in future callbacks
   */
  if (is_void(prev)) prev = string(0);
  prev += line;
  selist = strword(prev, "\r\n", 256);
  line = strpart(prev, selist);
  line = line(where(line));
  n = numberof(line);
  if (n && selist(2*n)==strlen(prev)) {
    /* final character of input not \n, store fragment in prev */
    prev = line(0);
    line = (n==1)? [] : line(1:-1);
  } else {
    prev = string(0);
  }

  if (dead) {
    if (is_void(line) || !line(0)) grow, line, [string(0)];
  }
  return line;
}

extern suspend;
extern resume;
/* DOCUMENT suspend
            resume
     Stop execution of the current interpreted program with suspend.
     It resumes at the instruction following suspend when yorick
     becomes idle after another interpreted task has called resume.
     Note that the task which calls resume must be triggered by an
     input stream other than stdin, such as the on_stdout or on_stderr
     function of a spawned process or the on_elapse of an after.
     Use control-c to escape from a hung suspend state.
   SEE ALSO: spawn, funcdef, after
*/

extern _after_func;  /* worker functions for after */
func _after_work { _after_func; }

extern after;
/* DOCUMENT after, secs, f
         or after, secs, f, arg
         of after, -, f, arg
         of after, -
     Execute yorick statement
       F;
     or
       F, ARG;
     when yorick becomes idle, but at least SECS seconds from now.
     SECS may be type double to specify fractions of a second.
     With SECS = 0.0, this is the same as set_idler, except that
     while you may have only a single idler function, you may have
     many after functions.  F may be either a function (is_func(f)
     non-zero), or an oxy object (is_obj(f) non-zero).  For example,
       after, 0.1, include, ["fma; plg, y, x;"];
     can obviously be modified to do anything you want, although you
     are probably better off writing a function containing the
     executable line, rather than putting it into a string.
     As another example,
       after, 0.1, object, method;
     invokes the object method after a delay of a tenth of a second.
     (See help,oxy for more on objects.)  If F is an object, and method
     is a simple variable reference, the special semantics of object
     arguments apply; that is, only the name "method" is significant,
     not its value.
     In the third form, with the pseudo-index - as the first argument,
     cancels the specified after call(s).  The ARG, if specified, must
     be the same variable, not just the same value.  If no ARG is specified,
     all pending after callbacks with the given F are cancelled.  If
     neither ARG nor F is specified, all after callbacks are cancelled.
   SEE ALSO: spawn, set_idler, after_error
*/

/*--------------------------------------------------------------------------*/

extern timestamp;
/* DOCUMENT timestamp()
         or timestamp(utime)
         or timestamp, utime
     returns string of the form "Sun Jan  3 15:14:13 1988" -- always
     has 24 characters.  If a simple variable reference UTIME is supplied,
     it will be set to the number of seconds since 1970 Jan 1 0000 UT.
   SEE ALSO: timer
 */

extern timer;
/* DOCUMENT timer, elapsed
         or timer, elapsed, split
     updates the ELAPSED and optionally SPLIT timing arrays.  These
     arrays must each be of type array(double,3); the layout is
     [cpu, system, wall], with all three times measured in seconds.
     ELAPSED is updated to the total times elapsed since this copy
     of Yorick started.  SPLIT is incremented by the difference between
     the new values of ELAPSED and the values of ELAPSED on entry.
     This feature allows for primitive code profiling by keeping
     separate accounting of time usage in several categories, e.g.--
        elapsed= total= cat1= cat2= cat3= array(double, 3);
        timer, elapsed0;
        elasped= elapsed0;
        ... category 1 code ...
        timer, elapsed, cat1;
        ... category 2 code ...
        timer, elapsed, cat2;
        ... category 3 code ...
        timer, elapsed, cat3;
        ... more category 2 code ...
        timer, elapsed, cat2;
        timer, elapsed0, total;
     The wall time is not absolutely reliable, owning to possible
     rollover at midnight.
   SEE ALSO: timestamp, timer_print
 */

func timer_print(label, split, ..)
/* DOCUMENT timer_print, label1, split1, label2, split2, ...
         or timer_print
         or timer_print, label_total
     prints out a timing summary for splits accumulated by timer.
        timer_print, "category 1", cat1, "category 2", cat2,
                     "category 3", cat3, "total", total;
   SEE ALSO: timer
 */
{
  elapsed= s= array(double, 1:3);
  timer, elapsed;
  write,format="%30s     CPU sec  System sec    Wall sec\n","Timing Category";
  if (!is_void(label) && !is_void(split)) {
    s(1:3)= split;
    write,format="%30s %11.3f %11.3f %11.3f\n", label, s(1), s(2), s(3);
  }
  while (more_args()>1) {
    labl= next_arg();
    s(1:3)= next_arg();
    write,format="%30s %11.3f %11.3f %11.3f\n", labl, s(1), s(2), s(3);
  }
  if (is_void(label) || is_void(split)) {
    if (is_void(label)) labl= "-----Total Elapsed Times-----";
    else labl= label;
    s(1:3)= elapsed;
    write,format="%30s %11.3f %11.3f %11.3f\n", labl, s(1), s(2), s(3);
  }
}

_timer_elapsed= [0.,0.,0.];
timer, _timer_elapsed;

/*= SECTION(mesh) functions on mesh arrays =================================*/

func area(y, x)
/* DOCUMENT area(y, x)
     returns the zonal areas of the 2-D mesh (X, Y).  If Y and X are
     imax-by-jmax, the result is (imax-1)-by-(jmax-1).  The area is
     positive when, say, X increases with i and Y increases with j.
     For example, area([[0,0],[1,1]],[[0,1],[0,1]]) is +1.
   SEE ALSO: volume
 */
{ return x(dif,zcen)*y(zcen,dif) - x(zcen,dif)*y(dif,zcen); }

func volume(r, z)
/* DOCUMENT volume(r, z)
     returns the zonal volumes of the 2-D cylindrical mesh (R, Z).
     If R and Z are imax-by-jmax, the result is (imax-1)-by-(jmax-1).
     The volume is positive when, say, Z increases with i and R increases
     with j.  For example, volume([[0,0],[1,1]],[[0,1],[0,1]]) is +pi.
   SEE ALSO: area
 */
{
  s= r*r;
  v= z(dif,zcen)*s(zcen,dif) - z(zcen,dif)*s(dif,zcen);
  s= z*r;
  return (2.0*pi/3.0)*(v+s(dif,zcen)*r(zcen,dif)-s(zcen,dif)*r(dif,zcen));
}

func ptcen(zncen, ireg)
/* DOCUMENT ptcen(zncen)
         or ptcen(zncen, ireg)
     returns point centered version of the 2-D zone centered array ZNCEN.
     The result is imax-by-jmax if ZNCEN is (imax-1)-by-(jmax-1).
     If the region number array IREG is specified, zones with region
     number 0 are not included in the point centering operation.
     Note that IREG should have dimensions imax-by-jmax; the first
     row and column of IREG are ignored.
     Without IREG, ptcen(zncen) is equivalent to zncen(pcen,pcen).
  SEE ALSO: zncen, uncen
 */
{
  if (is_void(ireg)) return zncen(pcen, pcen, ..);
  void= use_origins(0);
  exist= (ireg(2:,2:)!=0);
  return (exist*zncen)(pcen,pcen,..)/(exist(pcen,pcen)+1.e-35);
}

func zncen(ptcen, ireg)
/* DOCUMENT zncen(ptcen)
         or zncen(ptcen, ireg)
     returns zone centered version of the 2-D point centered array PTCEN.
     The result is (imax-1)-by-(jmax-1) if PTCEN is imax-by-jmax.
     If the region number array IREG is specified, zones with region
     number 0 are not included in the point centering operation.
     Note that IREG should have dimensions imax-by-jmax, like
     the input PTCEN array; the first row and column of IREG are ignored.
     Without IREG, zncen(ptcen) is equivalent to ptcen(zcen,zcen).
  SEE ALSO: ptcen, uncen
 */
{
  if (is_void(ireg)) return ptcen(zcen, zcen, ..);
  void= use_origins(0);
  exist= (ireg(2:,2:)!=0);
  return exist*ptcen(zcen, zcen, ..);
}

func uncen(ptcen, ireg)
/* DOCUMENT uncen(ptcen)
         or uncen(ptcen, ireg)
     returns zone centered version of the 2-D zone centered array PTCEN.
     The result is (imax-1)-by-(jmax-1) if PTCEN is imax-by-jmax.
     If the region number array IREG is specified, zones with region
     number 0 are not included in the point centering operation.
     Note that IREG should have dimensions imax-by-jmax, like
     the input PTCEN array; the first row and column of IREG are ignored.
     Without IREG, uncen(ptcen) is equivalent to ptcen(uncp,uncp).

     Do not use uncen to zone center data which is naturally point
     centered -- use the zncen function for that purpose.  The uncen
     function is the (nearly) exact inverse of the ptcen function,
     so that uncen(ptcen(zncen, ireg), ireg) will return the original
     zncen array.  The uncen reconstruction is as exact as possible,
     given the finite precision of floating point operations.
  SEE ALSO: ptcen, zncen
 */
{
  if (is_void(ireg)) return ptcen(uncp, uncp, ..);
  void= use_origins(0);
  exist= (ireg(2:,2:)!=0);
  return (exist(pcen,pcen)*ptcen)(uncp, uncp, ..);
}

/*--------------------------------------------------------------------------*/

extern noop;
/* DOCUMENT noop(x)
     returns X.  Use to make simple variable references into expressions.
     The noop function is a builtin, which runs much faster than the
     interpreted "call" function.  Also, if X is an array reference for
     a file handle, "call" performs the read, while "noop" does not.
   SEE ALSO: call
 */

func call(void)
/* DOCUMENT call, subroutine(arg1, arg2, arg3, arg4, arg5
                             arg6, arg7, arg8);
     allows a SUBROUTINE to be called with a very long argument list
     as an alternative to:
          subroutine, arg1, arg2, arg3, arg4, arg5,
            arg6, arg7, arg8;
     Note that the statement
          subroutine(arg1, arg2, arg3, arg4, arg5,
                     arg6, arg7, arg8);
     will print the return value of subroutine, even if it is nil.
     If invoked as a function, call simply returns its argument.
   SEE ALSO: noop
 */
{ return void; }

/*= SECTION(checksum) crc, md5, sha1 checksums =============================*/

extern crc_on;
/* DOCUMENT crc = crc_on(x)
         or crc = crc_on(x, crc0)
         or crc_table = crc_on(crc_def, -)
         or crc = crc_on(x, crc_table)
         or crc = crc_on(x, crc_table, crc0)
         or crc_def = crc_on(crc_table, -)
     return a cyclic redundancy check on X.  The crc has type long which
     is very likely (1 chance in 4 billion) to remain unchanged if X is
     corrupted by random noise.  With a non-nil crc0 argument previously
     returned by crc_on, begins with crc0 to yield (roughly speaking) the
     result you would have gotten on a single call if the two X arguments
     had been concatenated (note that the order matters).

     If X is a string array, the strings themselves, not including trailing
     '\0', are concatenated; string(0) is indistinguishable from "".
     If X is a struct instance array or a pointer array, the checksum
     returns an error.  X must be an array or nil [].

     Note that the crc value for any data type other than char depends
     on the native binary format on the platform where crc_on runs; the
     value will be different on other formats, just as it will be different
     for an array cast to a different data type.

     There are many different CRC algorithms, which can be parameterized
     by five integer values:
       CRC_DEF = [width, poly, init, reflect, xor]
     Here width is the width in bits, reflect is either 0 or 1 (false or
     true), and poly, init, and xor are zero except for at most their
     width least significant bits.  The returned crc is also zero except
     for its width least significant bits.  The parameterization is
     described in "A Painless Guide to CRC Error Detection Algorithms" at
       http://www.ross.net/crc/
     (The reflect parameter corresponds to refin and refot, which must
     be equal for crc_on to work, and the xor parameter corresponds to
     xorot.)  You can find a list of popular parameter values at
       http://regregex.bbcmicro.net/crc-catalogue.htm
     Do not try to "roll your own" parameters; let the experts do it.
     Here are some popular choices (crc_on requires width>=8):
       crc_def = [32, 0x04C11DB7, 0xFFFFFFFF, 1, 0xFFFFFFFF]  ("pkzip")
       crc_def = [32, 0x04C11DB7, 0, 0, 0xFFFFFFFF]           ("cksum")
       crc_def = [24, 0x864CFB, 0xB704CE, 0, 0]               ("crc24")
       crc_def = [16, 0x8005, 0, 1, 0]                        ("arc")
       crc_def = [16, 0x1021, 0, 1, 0]                        ("kermit")
     The default is "pkzip".  You can pass any of these five strings
     instead of an array of five numbers as CRC_DEF.
     
     To use a CRC algorithm other than "pkzip", you must first generate
     a CRC_TABLE by calling crc_on(crc_def,-), then pass the CRC_TABLE
     as the second argument with X as the first to compute the CRC.
     Finally, crc_on(crc_table,-) returns the corresponding CRC_DEF;
     crc_on(,-) returns the CRC_DEF for the default "pkzip" algorithm.
   SEE ALSO: md5, sha1
 */

extern md5;
extern sha1;
/* DOCUMENT digest = md5(data)         compute digest of DATA array
            state = []                 initialize STATE
            md5, state, data           process DATA updating STATE
            digest = md5(state, data)  return DIGEST from STATE
            sha1 function has same semantics as md5 function
     The md5 and sha1 functions compute message digests or hashes.  The
     digest returned by md5 is an array of 16 char (128 bits); the digest
     returned by sha1 is an array of 20 char (160 bits).  There is a
     single call form, in which the input DATA array comprises the entire
     "message" to be digested.  There is also a multi-call sequence, in
     which you invoke md5 or sha1 with two arguments: The first argument
     is a STATE variable, while the second argument is the DATA to be
     appended to the "message" in that call.  DATA may be nil [], which
     causes no change in the STATE.  STATE must be a simple variable
     reference, which is updated and returned with each call.  As long
     as you call md5 or sha1 as a subroutine, STATE continues to be updated;
     call md5 or sha1 as a function to return the final digest of the
     concatenated "message".  Before the first call, you must initialize
     STATE to [].  The final function call destroys STATE, returning it
     as [].  (To generate the final digest, both the md5 and sha1 algorithms
     append some padding to the input message, which destroys STATE.)

     If DATA is a string array, the strings themselves, not including trailing
     '\0', are concatenated; string(0) is indistinguishable from "".
     If DATA is a struct instance array or a pointer array, the digesting
     function returns an error.  DATA must be an array or nil [].

     Although STATE is an array of char, it is platform dependent even
     though the final digest is not.  Do not attempt to save STATE or
     to use STATE itself as a digest.

     The advantage of md5 or sha1 over the crc_on function is that
     the resulting digest will be unique; no two DATA streams will
     ever produce the same digest for nearly all practical purposes.
     For the 32-bit crc result, you will get different data streams
     that have identical crc values after you process a modest number
     of data streams (tens of thousands).  Thus, md5 or sha1 can be used
     to "fingerprint" data streams; if the fingerprints of two streams
     agree, you can be practically sure the streams themselves agree.

     Both md5 and sha1 have now been "broken" cryptographically, which
     means that it is possible by heroic effort to create two different
     streams with the same digest.  (At this writing, many examples of
     streams with identical md5 digests exist.  No such examples exist
     yet for sha1, but it is clear that a few will appear fairly soon.)
     However, absent malicious intent and huge levels of effort, both
     md5 and sha1 are perfectly useful for fingerprinting data.  These
     algorithms give the same results as the md5sum and sha1sum
     utilities, widely used to fingerprint files on the Web.

   SEE ALSO: crc_on
 */

/*= SECTION(debug) debug commands ==========================================*/

extern error;
extern exit;
/* DOCUMENT exit, msg
            error, msg
     Exits the current interpreted *main* program, printing the MSG.
     (MSG can be omitted to print a default.)
     In the case of exit, the result is equivalent to an immediate
     return from every function in the current calling chain.
     In the case of error, the result is the same as if an error had
     occurred in a compiled routine.
   SEE ALSO: print, write, batch, catch
 */

extern catch;
/* DOCUMENT catch(category)
     Catch errors of the specified category.  Category may be -1 to
     catch all errors, or a bitwise or of the following bits:

        0x01 math errors (SIGFPE, math library)
        0x02 I/O errors
        0x04 keyboard interrupts (e.g.- control C interrupt)
        0x08 other compiled errors (YError)
        0x10 interpreted errors (error)

     Use catch by placing it in a function before the section of code
     in which you are trying to catch errors.  When catch is called,
     it always returns 0, but it records the virtual machine program
     counter where it was called, and longjumps there if an error is
     detected.  The most recent matching call to catch will catch the
     error.  Returning from the function in which catch was called
     pops that call off the list of catches the interpreter checks.

     To use catch, place the call near the top of a function:

        if (catch(category)) {
          ...<code to execute if error is caught>...
        }
        ...<code "protected" by the catch>...

     If an error with the specified category occurs in the "protected"
     code, the program jumps back to the point of the catch and acts
     as if the catch function had returned 1 (remember that when catch
     is actually called it always returns 0).

     In order to lessen the chances of infinite loops, the catch is
     popped off the active list if it is actually used, so that a
     second error will *not* be caught.  Often, this is only desirable
     for the error handling code itself -- if you want to re-execute
     the "protected" code, do this, and take care of the possibility
     of infinite loops in your interpreted code:

        while (catch(category)) {
          ...<code to execute if error is caught>...
        }
        ...<code "protected" by the catch>...

     After an error has been caught, the associated error message
     (what would have been printed had it not been caught) is left
     in the variable catch_message.

     ***WARNING***
     If the code protected by the catch contains include or require
     calls, or function references which force autoloads, and the
     fault occurs while yorick is interpreting an included file,
     catch will itself fault, and the error code will not execute.
     If a fault occurs after an include has pushed a file onto
     the include stack for delayed parsing and you catch that fault,
     the include stack will not unwind to its condition at the time
     catch was called.  That is, catch is incapable of protecting
     you completely during operations involving nested levels of
     include files.

     In some cases, after_error is a more appropriate way to recover
     from errors.

   SEE ALSO: error, after_error
 */

extern dbexit;
extern dbcont;
extern dbret;
extern dbskip;
extern dbup;
extern dbinfo;
extern dbdis;
extern dbauto;
extern dbwhere;
/* DOCUMENT Debug mode.

   Yorick errors fall into two general categories: Syntax errors discovered
   during parsing, and runtime errors discovered when a Yorick program is
   actually running.  When a runtime error occurs, Yorick offers the
   choice of entering "debug mode", which you can do by typing the <RETURN>
   key immediately after the error occurs.  Typing a non-blank line exits
   debug mode automatically by default.  In debug mode, the Yorick prompt
   becomes "dbug>" instead of the usual ">".  When you see this prompt,
   Yorick has halted "in the middle of" the function in which the error
   occurred, and you can print, plot, modify, or save the local variables
   in that function by means of ordinary Yorick commands.  Debug mode is
   recursive; that is, you can debug an error which occurred during
   debugging to any number of levels.

   You can exit from debug mode in several ways:

      dbexit            -- exit current debug level, discarding all
                           active functions and their local variables
      dbexit, 0         -- exit all debug levels
      dbexit, n         -- exit (at most) N debug levels

      dbcont            -- continue execution of the current function
         Continuing is useful if you have managed to repair the
         problem which caused the error.  The expression in which the
         error occurred will be evaluated a second time, so beware of
         side effects.

      dbret, value      -- continue execution by returning VALUE (which
                           may be nil or omitted) to the caller of the
                           function in which the error occurred.
         This is useful if the function in which the error occurred is
         hopelessly confounded, but you know the value it should return.

   Yorick does not allow "single stepping" directly, although you can
   execute the statements in a function by copying them, then tell
   Yorick to skip those statements you have executed "by hand".  There
   are two functions for skipping execution:

      dbskip            -- skip the next logical line (This will be only
                           a portion of a source line if several statements
                           are stacked on the source line.)
      dbskip, n         -- skip next N (positive or negative) logical lines

      dbup              -- discard the current function, so that you are
                           debugging its caller -- there is no way to go
                           back "down", so be careful

   There are two functions which print information (like other print
   functions, if called as functions instead of subroutines, their
   result is returned as a string array with one line per string):

      dbinfo            -- returns current function and source line

      dbdis             -- returns disassembled virtual machine code
                           for the next line (use the disassemble function
                           to get the entire function)
         This allows you to see exactly where in a line the error occurred.
      dbwhere           -- returns entire calling stack as func[pc] strings
         The dbwhere function also works outside of dbug mode.

   Finally,

      dbauto            -- toggles whether debug mode will be entered
                           automatically when a runtime error occurs
      dbauto, 1         -- enter debug mode automatically after an error
      dbauto, 0         -- type <RETURN> after error to enter debug mode
 */

extern disassemble;
/* DOCUMENT disassemble(function)
         or disassemble, function
     Disassembles the specified function.  If called as a function, the
     result is returned as a vector of strings; if called as a subroutine,
     the disassembly is printed at the terminal.  If the function is nil,
     the current *main* program is disassembled -- you must include the
     call to disassemble in the main program, of course, NOT on its own
     line as a separate main program.
 */

/*= SECTION(list) list objects (deprecated, use oxy) =======================*/

extern _lst;
extern _cat;
extern _car;
extern _cdr;
extern _cpy;
extern _len;
/* DOCUMENT list= _lst(item1, item2, item3, ...)
            list= _cat(item_or_list1, item_or_list2, item_or_list3, ...)
            list= _cpy(list)
              list= _cpy(list, i)
            length= _len(list)
            item= _car(list)
              item_i= _car(list, i)
              _car, list, i, new_item_i
            list= _cdr(list)
              list= _cdr(list, i)
              _cdr, list, i, new_list_i

     **** DEPRECATED, object extensions in new code, see help,oxy 

     implement rudimentary Lisp-like list handling in Yorick.
     However, in Yorick, a list must have a simple tree structure
     - no loops or rings are allowed (loops break Yorick's memory
     manager - beware).  You need to be careful not to do this as
     the error will not be detected.

     Lists are required in Yorick whenever you need to hold an
     indeterminate amount of non-array data, such as file handles,
     bookmarks, functions, index ranges, etc.  Note that Yorick
     pointers cannot point to these objects.  For array data, you have
     a choice between a list and a struct or an array of pointers.
     Note that a list cannot be written into a file with the save
     function, since it may contain unsaveable items.

     The _lst (list), _cat (catenate), and _cpy (copy) functions
     are the principal means for creating and maintaining lists.
     _lst makes a list out of its arguments, so that each argument
     becomes one item of the new list.  Unlike Yorick array data
     types, a statement like x=list does not make a copy of the
     list, it merely makes an additional reference to the list.
     You must explicitly use the _cpy function to copy a list.  Note
     that _cpy only copies the outermost list itself, not the items
     in the list (even if those items are lists).  With the second
     argument i, _cpy copies only the first i items in the list.
     The _cat function concatentates several lists together,
     "promoting" any arguments which are not lists.  This operation
     changes the values of list arguments to _cat, except for the
     final argument, since after _cat(list, item), the variable list
     will point to the new longer list returned by _cat.

     Nil, or [], functions as an empty list.  This leads to ambiguity
     in the argument list for _cat, since _cat "promotes" non-list
     arguments to lists; _cat treats [] as an empty list, not as a
     non-list item.  Also, _lst() or _lst([]) returns a single item list,
     not [] itself.

     The _len function returns the number of items in a list, or 0
     for [].

     The _car and _cdr functions (the names are taken from Lisp,
     where they originally stood for something like "address register"
     and "data register" of some long forgotten machine) provide
     access to the items stored in a list.  _car(list,i) returns the
     i-th item of the list, and i defaults to 1, so _car(list) is the
     first item.  Also, _car,list,i,new_item_i sets the i-th item
     of the list.  Finally, _cdr(list,i) returns a list of all the
     items beyond the i-th, where i again defaults to 1.  The form
     _cdr,list,i,new_list_i can be used to reset all list items
     beyond the i-th to new values.  In the _cdr function, i=0 is
     allowed.  When used to set values, both _car and _cdr can also
     be called as functions, in which case they return the item or
     list which has been replaced.  The _cdr(list) function returns
     nil if and only if LIST contains only a single item; this is
     the usual means of halting a loop over items in a list.

   SEE ALSO: array, grow, _prt, _map, _rev, _nxt
 */

func _prt(x, indent)
/* DOCUMENT _prt, list
     print every item in a list, recursing if some item is itself a list.
   SEE ALSO: _lst
 */
{
  if (is_void(indent)) indent= "";
  if (typeof(x)!="list") {
    write,format="%s\n",indent+print(x);
    return;                       /* exit recursion */
  }
  write,format="%s\n",indent+"list items:";
  do {
    _prt, _car(x), indent+"  ";   /* recurse */
    x= _cdr(x);
  } while (!is_void(x));
}

func _map(f__map, list__map)
/* DOCUMENT _map(f, list)
     return a list of the results of applying function F to each
     element of the input LIST in turn, as if by
       _lst(f(_car(list,1)),f(_car(list,2)),...)
   SEE ALSO: _lst
 */
{
  /* all locals here must have weird names, since the function f will
   * very often rely on external variables for arguments not varying
   * in the input list, or for accumulated outputs */
  if (is_void(list__map)) return [];
  result__map= tail__map= _lst(f__map(_car(list__map)));
  for (list__map=_cdr(list__map) ;
       !is_void(list__map) ; list__map=_cdr(list__map)) {
    _cat, tail__map, _lst(f__map(_car(list__map)));
    tail__map= _cdr(tail__map);
  }
  return result__map;
}

func _rev(list)
/* DOCUMENT _rev(list)
     returns the input list in reverse order
   SEE ALSO: _lst
 */
{
  if (is_void(list)) return;
  prev= [];
  for (;;) {
    tail= _cdr(list, 1, prev);
    if (is_void(tail)) return list;
    prev= list;
    list= tail;
  }
}

func _nxt(&list)
/* DOCUMENT item= _nxt(list)
     return first item in LIST, and set LIST to list of remaining
     items.  If you are iterating through a list, this is the way
     to do it, since a loop on _car(list,i) with i varying from 1
     to _len(list) scales quadratically with the length of the list,
     while a loop on _nxt(list) scales linearly.
   SEE ALSO: _car, _lst
 */
{
  item= _car(list);
  list= _cdr(list);
  return item;
}

/*--------------------------------------------------------------------------*/