Summary of changes for version 1.1.1:
------------------------------------

  * New built-in variables, default_return_value and
    define_all_return_values.

    If define_all_return_values is set to "false", Octave does not do
    anything special for return values that are left undefined, and
    you will get an error message if you try to use them.  For
    example, if the function

      function [x, y] = f ()
        y = 1;
      endfunction

    is called as

      octave:13> [a, b] = f ()

    Octave will print an error message for the attempt to assign an
    undefined value to `a'.

    This is incompatible with Matlab, which will define the return
    variable `x' to be the empty matrix.  To get the Matlab-like
    behavior, you can set the variable define_all_return_values to
    "true" (the default is "false") and default_return_value to `[]'
    (the default).  Then, any return values that remain undefined when
    the function returns will be initialized to `[]'.

    If the function is called without explicitly asking for an output,
    it will succeed.  This behavior is compatible and unchanged from
    previous versions of Octave.

  * New built-in variable suppress_verbose_help_message.  If set to
    "true", Octave will not add additional help information to the end
    of the output from the help command and usage messages for
    built-in commands.  The default value is "false".

  * New built-in variable PS4 is used as the prefix of echoed input
    (enabled with the --echo-input (-x) option).

  * The function size() now accepts an optional second argument.

  * Output from `save - ...' now goes through the pager.

  * The break statement may also be used to exit a function, for
    compatibility with Matlab.

  * The directory tree for installing Octave is now closer to
    conforming with the current GNU standards.

  * More bug fixes.

Summary of changes for version 1.1.0:
------------------------------------

  * Octave now requires g++ 2.6.3 or later.  This change is necessary
    to make template instantiations cleaner, and to avoid having to
    have special cases in the code for earlier versions of gcc.

  * A new data structure type has been added.  The implementation uses
    an associative array with indices limited to strings, but the
    syntax is more like C-style structures.  here are some examples of
    using it.

    Elements of structures can be of any type, including structures:

      octave:1> x.a = 1;
      octave:2> x.b = [1, 2; 3, 4];
      octave:3> x.c = "string";
      octave:4> x
      x =

      <structure: a b c>

      octave:5> x.a
      x.a = 1
      octave:6> x.b
      x.b =

        1  2
        3  4

      octave:7> x.c
      x.c = string
      octave:8> x.b.d = 3
      x.b.d = 3
      octave:9> x.b
      x.b =

      <structure: d>

      octave:10> x.b.d
      x.b.d = 3

    Functions can return structures:

      octave:1> a = rand (3) + rand (3) * I;
      octave:2> function y = f (x)
      > y.re = real (x);
      > y.im = imag (x);
      > endfunction
      octave:3> f (a)
      ans =

      <structure: im re>

      octave:4> ans.im
      ans.im =

        0.093411  0.229690  0.627585
        0.415128  0.221706  0.850341
        0.894990  0.343265  0.384018

      octave:5> ans.re
      ans.re =

        0.56234  0.14797  0.26416
        0.72120  0.62691  0.20910
        0.89211  0.25175  0.21081

    Return lists can include structure elements:

      octave:1> [x.u, x.s, x.v] = svd ([1, 2; 3, 4])
      x.u =

        -0.40455  -0.91451
        -0.91451   0.40455

      x.s =

        5.46499  0.00000
        0.00000  0.36597

      x.v =

        -0.57605   0.81742
        -0.81742  -0.57605

      octave:8> x
      x =

      <structure: s u v>

    This feature should be considered experimental, but it seems to
    work ok.  Suggestions for ways to improve it are welcome.

  * Octave now supports a limited form of exception handling modelled
    after the unwind-protect form of Lisp:

      unwind_protect
        BODY
      unwind_protect_cleanup
        CLEANUP
      end_unwind_protect

    Where BODY and CLEANUP are both optional and may contain any
    Octave expressions or commands.  The statements in CLEANUP are
    guaranteed to be executed regardless of how control exits BODY.

    This is useful to protect temporary changes to global variables
    from possible errors.  For example, the following code will always
    restore the original value of the built-in variable
    do_fortran_indexing even if an error occurs while performing the
    indexing operation.

      save_do_fortran_indexing = do_fortran_indexing;
      unwind_protect
        do_fortran_indexing = "true";
        elt = a (idx)
      unwind_protect_cleanup
        do_fortran_indexing = save_do_fortran_indexing;
      end_unwind_protect

    Without unwind_protect, the value of do_fortran_indexing would not
    be restored if an error occurs while performing the indexing
    operation because evaluation would stop at the point of the error
    and the statement to restore the value would not be executed.

  * Recursive directory searching has been implemented using Karl
    Berry's kpathsea library.  Directories below path elements that
    end in // are searched recursively for .m files.

  * Octave now waits for additional input when a pair of parentheses
    is `open' instead of giving an error.  This allows one to write
    statements like this

      if (big_long_variable_name == other_long_variable_name
          || not_so_short_variable_name > 4
          && y > x)
        some (code, here);

    without having to clutter up the if statement with continuation
    characters.

  * Continuation lines are now allowed in string constants and are
    handled correctly inside matrix constants.

  * Both `...{whitespace}\n' and `\{whitespace}\n' can be used to
    introduce continuation lines, where {whitespace} may include
    spaces, tabs and comemnts.

  * The script directory has been split up by topic.

  * Dynamic linking mostly works with dld.  The following limitations
    are known problems:

    -- Clearing dynamically linked functions doesn't work.

    -- Dynamic linking only works with dld, which has not been ported
       to very many systems yet.

    -- Configuring with --enable-lite-kernel seems to mostly work to
       make nonessential built-in functions dynamically loaded, but
       there also seem to be some problems.  For example, fsolve seems
       to always return info == 3.  This is difficult to debug since
       gdb won't seem to allow breakpoints to be set inside
       dynamically loaded functions.

    -- Octave uses a lot of memory if the dynamically linked functions
       are compiled with -g.  This appears to be a limitation with
       dld, and can be avoided by not using -g to compile functions
       that will be linked dynamically.

  * fft2 and ifft2 are now built-in functions.

  * The `&&' and `||' logical operators are now evaluated in a
    short-circuit fashion and work differently than the element by
    element operators `&' and `|'.  See the Octave manual for more
    details.

  * Expressions like 1./m are now parsed as 1 ./ m, not 1. / m.

  * The replot command now takes the same arguments as gplot or
    gsplot (except ranges, which cannot be respecified with replot
    (yet)) so you can add additional lines to existing plots.

  * The hold command has been implemented.

  * New function `clearplot' clears the plot window.  The name `clg'
    is aliased to `clearplot' for compatibility with Matlab.

  * The commands `gplot clear' and `gsplot clear' are equivalent to
    `clearplot'.  (Previously, `gplot clear' would evaluate `clear' as
    an ordinary expression and clear all the visible variables.)

  * The Matlab-style plotting commands have been improved.  They now
    accept line-style arguments, multiple x-y pairs, and other plot
    option flags.  For example,
 
      plot (x, y, "@12", x, y2, x, y3, "4", x, y4, "+")

    results in a plot with

      y  plotted with points of type 2 ("+") and color 1 (red).
      y2 plotted with lines.
      y3 plotted with lines of color 4.
      y4 plotted with points which are "+"s.

    the help message for `plot' and `plot_opt' provide full
    descriptions of the options.

  * NaN is now dropped from plot data, and Inf is converted to a
    very large value before calling gnuplot.

  * Improved load and save commands:

    -- The save and load commands can now read and write a new binary
       file format.  Conversion to and from IEEE big and little endian
       formats is handled automatically.  Conversion for other formats
       has not yet been implemented.

    -- The load command can now read Matlab .mat files, though it is
       not yet able to read sparse matrices or handle conversion for
       all data formats.

    -- The save command can write Matlab .mat files.

    -- The load command automatically determines the save format
       (binary, ascii, or Matlab binary).

    -- The default format for the save command is taken from the
       built-in variable `default_save_format'.

    -- The save and load commands now both accept a list of globbing
       patterns so you can easily load a list of variables from a
       file.

    -- The load command now accepts the option -list, for listing the
       variable names without actually loading the data.  With
       -verbose, it prints a long listing.

    -- The load command now accepts the option -float-binary, for
       saving floating point data in binary files in single precision.

  * who and whos now accept a list of globbing patterns so you can
    limit the lists of variables and functions to those that match a
    given set of patterns.

  * New functions for manipulating polynomials
      
      compan     -- companion matrix corresponding to polynomial coefficients
      conv       -- convolve two vectors
      deconv     -- deconvolve two vectors
      roots      -- find the roots of a polynomial
      poly       -- characteristic polynomial of a matrix
      polyderiv  -- differentiate a polynomial
      polyinteg  -- integrate a polynomial
      polyreduce -- reduce a polynomial to minimum number of terms
      polyval    -- evaluate a polynomial at a point
      polyvalm   -- evaluate a polynomial in the matrix sense
      residue    -- partial fraction expansion corresponding to the ratio
                    of two polynomials

  * New functions for manipulating sets

      create_set   -- create a set of unique values
      complement   -- find the complement of two sets
      intersection -- find the intersection of two sets
      union        -- find the union of two sets

  * New elementary functions:

      acot   acoth   acsc   acsch
      asec   asech   cot    coth
      csc    csch    log2   sec
      sech

  * New special functions:

      beta   -- beta function
      betai  -- incomplete beta function
      gammai -- incomplete gamma function

  * New image processing functions:

      colormap  -- set and return current colormap
      gray      -- set a gray colormap
      gray2ind  -- image format conversion
      image     -- display an image
      imagesc   -- scale and display an image
      imshow    -- display images
      ind2gray  -- image format conversion
      ind2rgb   -- image format conversion
      loadimage -- load an image from a file
      ntsc2rgb  -- image format conversion
      ocean     -- set a color colormap
      rgb2ind   -- image format conversion
      rgb2ntsc  -- image format conversion
      saveimage -- save an image to a file

  * New time and date funcitons:

      tic          -- set wall-clock timer
      toc          -- get elapsed wall-clock time, since timer last set
      etime        -- another way to get elapsed wall-clock time
      cputime      -- get CPU time used since Octave started
      is_leap_year -- is the given year a leap year?

  * Other new functions:

      bug_report -- submit a bug report to the bug-octave mailing list

      toascii -- convert a string to a matrix of ASCII character codes

      octave_tmp_file -- generate a unique temporary file name

      undo_string_escapes -- replace special characters in a string by
                             their backslash forms

      is_struct -- determine whether something is a structure data type

      feof   -- check EOF condition for a specified file
      ferror -- check error state for a specified file
      fread  -- read binary data from a file
      fwrite -- write binary data to a file

      file_in_path -- check to see if named file exists in given path

      kbhit  -- get a single character from the terminal

      axis   -- change plot ranges
      hist   -- plot histograms

      diary  -- save commands and output to a file

      type   -- show the definition of a function
      which  -- print the type of an identifier or the location of a
                function file

      isieee  -- Returns 1 if host uses IEEE floating point
      realmax -- Returns largest floating point number
      realmin -- Returns smallest floating point number

      gcd     -- greatest common divisor
      lcm     -- least common multiple

      null    -- orthonormal basis of the null space of a matrix
      orth    -- orthonormal basis of the range space of a matrix

      fft2    -- two-dimensional fast fourier transform
      ifft2   -- two-dimensional inverse fast fourier transform
      filter  -- digital filter
      fftfilt -- filter using fft
      fftconv -- convolve to vectors using fft
      sinc    -- returns sin(pi*x)/(pi*x)
      freqz   -- compute the frequency response of a filter

  * The meaning of nargin (== args.length ()) in built-in functions
    has been changed to match the meaning of nargin in user-defined
    functions.

  * Variable return lists.  Octave now has a real mechanism for
    handling functions that return an unspecified number of values,
    so it is no longer necessary to place an upper bound on the number
    of outputs that a function can produce.

    Here is an example of a function that uses the new syntax to
    produce n values:

      function [...] = foo (n)
        for i = 1:n
          vr_val (i * x);
        endfor
      endfunction

  * New keyword, all_va_args, that allows the entire list of va_args
    to be passed to another function.  For example, given the functions

      function f (...)
        while (nargin--)
          disp (va_arg ())
        endwhile
      endfunction
      function g (...)
        f ("begin", all_va_args, "end")
      endfunction

    the statement

      g (1, 2, 3)

    prints

      begin
      1
      2
      3
      end

    all_va_args may be used more than once, but can only be used
    within functions that take a variable number of arguments.

  * If given a second argument, svd now returns an economy-sized
    decomposition, eliminating the unecessary rows or columns of U or
    V.

  * The max and min functions correctly handle complex matrices in
    which some columns contain real values only.

  * The find function now handles 2 and 3 output arguments.

  * The qr function now allows computation of QR with pivoting.

  * hilb() is much faster for large matrices.

  * computer() is now a built-in function.

  * pinv() is now a built-in function.

  * The output from the history command now goes through the pager.

  * If a function is called without assigning the result, nargout is
    now correctly set to 0.

  * It is now possible to write functions that only set some return
    values.  For example, calling the function

      function [x, y, z] = f () x = 1; z = 2; endfunction

    as

      [a, b, c] = f ()

    produces:

      a = 1

      b = [](0x0)

      c = 2

  * The shell_cmd function has been renamed to system (the name
    shell_cmd remains for compatibility).  It now returns [output, status].

  * New built-in variable `OCTAVE_VERSION'.  Also a new function,
    version, for compatibility with Matlab.

  * New built-in variable `automatic_replot'.  If it is "true", Octave
    will automatically send a replot command to gnuplot each time the
    plot changes.  Since this is fairly inefficient, the default value
    is "false".

  * New built-in variable `whitespace_in_literal_matrix' allows some
    control over how Octave decides to convert spaces to commas in
    matrix expressions like `[m (1)]'.

    If the value of `whitespace_in_literal_matrix' is "ignore", Octave
    will never insert a comma or a semicolon in a literal matrix list.
    For example, the expression `[1 2]' will result in an error
    instead of being treated the same as `[1, 2]', and the expression

      [ 1, 2,
        3, 4 ]

    will result in the vector [1 2 3 4] instead of a matrix.

    If the value of `whitespace_in_literal_matrix' is "traditional",
    Octave will convert spaces to a comma between identifiers and `('.
    For example, given the matrix

      m = [3 2]

    the expression

      [m (1)]

    will be parsed as

      [m, (1)]

    and will result in

      [3 2 1]

    and the expression

      [ 1, 2,
        3, 4 ]

    will result in a matrix because the newline character is converted
    to a semicolon (row separator) even though there is a comma at the
    end of the first line (trailing commas or semicolons are ignored).
    This is apparently how Matlab behaves.

    Any other value for `whitespace_in_literal_matrix' results in
    behavior that is the same as traditional, except that Octave does
    not convert spaces to a comma between identifiers and `('.
    For example, the expression

      [m (1)]

    will produce 3.  This is the way Octave has always behaved.

  * Line numbers in error messages for functions defined in files and
    for script files now correspond to the file line number, not the
    number of lines after the function keyword appeared.

  * Octave now extracts help from script files.  The comments must
    come before any other statements in the file.

  * In function files, the first block of comments in the file will
    now be interpreted as the help text if it doesn't look like the
    Octave copyright notice.  Otherwise, Octave extracts the first set
    of comments after the function keyword.

  * The function clock is more accurate on systems that have the
    gettimeofday() function.

  * The standard output stream is now automatically flushed before
    reading from stdin with any of the *scanf() functions.

  * Expanded reference card.

  * The Octave distribution now includes a frequently asked questions
    file, with answers.  Better answers and more questions (with
    answers!) are welcome.

  * New option --verbose.  If Octave is invoked with --verbose and not
    --silent, a message is printed if an octaverc file is read while
    Octave is starting.

  * An improved configure script generated by Autoconf 2.0.

  * Lots of bug fixes.

Summary of changes for version 1.0:
----------------------------------

  * C-style I/O functions now handle files referenced by name or by
    number more consistently.

Summary of changes for version 0.83:
-----------------------------------

  * Loading global symbols should work now.

  * Clearing the screen doesn't reprint the prompt unnecessarily.

  * The operations <complex scalar> OP <real matrix> for OP == +, -,
    *, or ./ no longer crash Octave.

  * More portability and configuration fixes.

Summary of changes for version 0.82:
-----------------------------------

  * Octave now comes with a reference card.

  * The manual has been improved, but more work remains to be done.

  * The atanh function now works for complex arguments.

  * The asin, acos, acosh, and atanh functions now work properly when
    given real-valued arguments that produce complex results.

  * SEEK_SET, SEEK_CUR, and SEEK_END are now constants.

  * The `using' qualifier now works with gplot and gsplot when the
    data to plot is coming directly from a file.

  * The strcmp function now works correctly for empty strings.

  * Eliminated bogus parse error for M-files that don't end with `end'
    or `endfunction'.

  * For empty matrices with one nonzero dimension, the +, -, .*, and
    ./ operators now correctly preserve the dimension.

  * Octave no longer crashes if you type ^D at the beginning of a line
    in the middle of defining a loop or if statement.

  * On AIX systems, Back off on indexing DiagArray via Proxy class to
    avoid gcc (or possibly AIX assembler?) bug. 

  * Various other bug and portability fixes.

Summary of changes for version 0.81:
-----------------------------------

  * Octave no longer dumps core if you try to define a function in
    your .octaverc file.

  * Fixed bug in Array class that resulted in bogus off-diagonal
    elements when computing eigenvalue and singular value
    decompositions.

  * Fixed bug that prevented lsode from working on the SPARCstation,
    at least with some versions of Sun's f77.  This bug was introduced
    in 0.80, when I changed LSODE to allow the user to abort the
    integration from within the RHS function.

  * Fixed bug that prevented global attribute of variables from being
    saved with save(), and another that prevented load() from working
    at all.

Summary of changes for version 0.80:
-----------------------------------

  * I have started working on a manual for the C++ classes.  At this
    point, it is little more than a list of function names.  If you
    would like to volunteer to help work on this, please contact
    bug-octave@bevo.che.wisc.edu.

  * The patterns accepted by the save and clear commands now work like
    file name globbing patterns instead of regular expressions.  I
    apologize for any inconvenience this change may cause, but file
    name globbing seems like a more reasonable style of pattern
    matching for this purpose.

  * It is now possible to specify tolerances and other optional inputs
    for dassl, fsolve, lsode, npsol, qpsol, and quad.  For each of
    these functions, there is a corresponding function X_options,
    which takes a keyword and value arguments.  If invoked without any
    arguments, the X_options functions print a list of possible
    keywords and current values.  For example,

      npsol_options ()

    prints a list of possible options with values, and

      npsol_options ("major print level", 10)

    sets the major print level to 10.

    The keyword match is not case sensitive, and the keywords may be
    abbreviated to the shortest unique match.  For example,

      npsol_options ("ma p", 10)

    is equivalent to the statement shown above.

  * The new built-in variable save_precision can be used to set the
    number of digits preserved by the ASCII save command.

  * Assignment of [] now works in most cases to allow you to delete
    rows or columns of matrices and vectors.  For example, given a
    4x5 matrix A, the assignment

      A (3, :) = []

    deletes the third row of A, and the assignment

      A (:, 1:2:5) = []

    deletes the first, third, and fifth columns.

  * Variable argument lists.  Octave now has a real mechanism for
    handling functions that take an unspecified number of arguments,
    so it is no longer necessary to place an upper bound on the number
    of optional arguments that a function can accept.

    Here is an example of a function that uses the new syntax to print
    a header followed by an unspecified number of values:

      function foo (heading, ...)
        disp (heading);
        va_start ();
        while (--nargin)
          disp (va_arg ());
        endwhile
      endfunction

    Note that the argument list must contain at least one named
    argument (this restriction may eventually be removed), and the
    ellipsis must appear as the last element of the argument list.

    Calling va_start() positions an internal pointer to the first
    unnamed argument and allows you to cycle through the arguments
    more than once.  It is not necessary to call va_start() if you
    do not plan to cycle through the arguments more than once.

  * Recursive functions should work now.

  * The environment variable OCTAVE_PATH is now handled in the same
    way as TeX handles TEXINPUTS.  If the path starts with `:', the
    standard path is prepended to the value obtained from the
    environment.  If it ends with `:' the standard path is appended to
    the value obtained from the environment.

  * New functions, from Kurt Hornik (hornik@neuro.tuwien.ac.at) and
    the Department of Probability Theory and Statistics TU Wien,
    Austria:

     corrcoef    -- corrcoef (X, Y) is the correlation between the i-th
                    variable in X and the j-th variable in Y
                    corrcoef (X) is corrcoef (X, X)
     cov         -- cov (X, Y) is the covariance between the i-th
                    variable in X and the j-th variable in Y
                    cov (X) is cov (X, X)
     gls         -- generalized least squares estimation
     kurtosis    -- kurtosis(x) = N^(-1) std(x)^(-4) SUM_i (x(i)-mean(x))^4 - 3
                    If x is a matrix, return the row vector containing
                    the kurtosis of each column
     mahalanobis -- returns Mahalanobis' D-square distance between the
                    multivariate samples X and Y, which must have the
                    same number of components (columns), but may have
                    a different number of observations (rows)
     ols         -- ordinary least squares estimation
     pinv        -- returns the pseudoinverse of X; singular values
                    less than tol are ignored
     skewness    -- skewness (x) = N^(-1) std(x)^(-3) SUM_i (x(i)-mean(x))^3
                    if x is a matrix, return the row vector containing
                    the skewness of each column

  * Errors in user-supplied functions called from dassl, fsolve,
    lsode, npsol, and quad are handled more gracefully.

  * Programming errors in the use of the C++ classes within Octave
    should no longer cause Octave to abort.  Instead, Octave's error
    handler function is called and execution continues as best as is
    possible.  This should result in eventually returning control to
    the top-level Octave prompt.  (It would be nice to have a real
    exception handling mechanism...)

  * A number of memory leaks have been eliminated.  Thanks to
    Fong Kin Fui <fui@ee.nus.sg> for reporting them.

  * The C++ matrix classes are now derived from a generic
    template-based array class.

  * The readline function operate-and-get-next (from bash) is now
    available and bound to C-O by default.

  * Octave now uses the version of readline currently distributed with
    bash-1.13.  On some systems, interactive invocations of Octave
    will now blink the cursor to show matching parens.

  * By default, include files are now installed in
    $prefix/include/octave instead of $prefix/include.

  * Octave now uses a config.h file instead of putting all defines on
    the compiler command line.

Summary of changes for version 0.79:
-----------------------------------

  * New control systems functions:

     dgram -- Returns the discrete controllability and observability gramian.
     dlqr  -- Discrete linear quadratic regulator design.
     dlqe  -- Discrete linear quadratic estimator (Kalman Filter) design.
     c2d   -- Convert continuous system description to discrete time
              description assuming zero-order hold and given sample time.

  * The max (min) functions can now return the index of the max (min)
    value as a second return value.

Summary of changes for version 0.78:
-----------------------------------

  * Octave's handling of global variables has been completely
    rewritten.  To access global variables inside a function, you must
    now declare them to be global within the function body.  Likewise,
    if you do not declare a variable as global at the command line,
    you will not have access to it within a function, even if it is
    declared global there.  For example, given the function

      function f ()
        global x = 1;
        y = 2;
      endfunction

    the global variable `x' is not visible at the top level until the
    command

      octave:13> global x

    has been evaluated, and the variable `y' remains local to the
    function f() even if it is declared global at the top level.

    Clearing a global variable at the top level will remove its global
    scope and leave it undefined.  For example,

      octave:1> function f ()   # Define a function that accesses
      >  global x;              #   the global variable `x'.
      >  x
      > endfunction
      octave:2> global x = 1    # Give the variable `x' a value.
      octave:3> f ()            # Evaluating the function accesses the
      x = 1                     #   global `x'.
      octave:4> clear x         # Remove `x' from global scope, clear value.
      octave:5> x = 2           # Define new local `x' at the top level
      x = 2
      octave:6> f               # The global `x' is no longer defined.
      error: `x' undefined near line 1 column 25
      error: evaluating expression near line 1, column 25
      error: called from `f'
      octave:7> x               # But the local one is.
      x = 2

  * The new function, `is_global (string)' returns 1 if the variable
    named by string is globally visible.  Otherwise, returns 0.

  * The implementation of `who' has changed.  It now accepts the
    following options:

      -b -builtins   -- display info for built-in variables and functions
      -f -functions  -- display info for currently compiled functions
      -v -variables  -- display info for user variables
      -l -long       -- display long info

    The long output looks like this:

      octave:5> who -l

      *** currently compiled functions:

      prot  type               rows   cols  name
      ====  ====               ====   ====  ====
       wd   user function         -      -  f

      *** local user variables:

      prot  type               rows   cols  name
      ====  ====               ====   ====  ====
       wd   real scalar           1      1  y

      *** globally visible user variables:

      prot  type               rows   cols  name
      ====  ====               ====   ====  ====
       wd   complex matrix       13     13  x

    where the first character of the `protection' field is `w' if the
    symbol can be redefined, and `-' if it has read-only access.  The
    second character may be `d' if the symbol can be deleted, or `-'
    if the symbol cannot be cleared.

  * The new built-in variable ignore_function_time_stamp can be used
    to prevent Octave from calling stat() each time it looks up
    functions defined in M-files.  If set to "system", Octave will not
    automatically recompile M-files in subdirectories of
    $OCTAVE_HOME/lib/VERSION if they have changed since they were last
    compiled, but will recompile other M-files in the LOADPATH if they
    change.  If set to "all", Octave will not recompile any M-files
    unless their definitions are removed with clear.  For any other
    value of ignore_function_time_stamp, Octave will always check to
    see if functions defined in M-files need to recompiled.  The
    default value of ignore_function_time_stamp is "system".

  * The new built-in variable EDITOR can be used to specify the editor
    for the edit_history command.  It is set to the value of the
    environment variable EDITOR, or `vi' if EDITOR is not set, or is
    empty.

  * There is a new built-in variable, INFO_FILE, which is used as the
    location of the info file.  Its initial value is
    $OCTAVE_HOME/info/octave.info, so `help -i' should now work
    provided that OCTAVE_HOME is set correctly, even if Octave is
    installed in a directory different from that specified at compile
    time.

  * There is a new command line option, --info-file FILE, that may be
    used to set Octave's idea of the location of the info file.  It
    will override any value of OCTAVE_INFO_FILE found in the
    environment, but not any INFO_FILE="filename" commands found in
    the system or user startup files. 

  * Octave's Info reader will now recognize gzipped files that have
    names ending in `.gz'.

  * The save command now accepts regular expressions as arguments.
    Note that these patterns are regular expressions, and do not work
    like filename globbing.  For example, given the variables `a',
    `aa', and `a1', the command `save a*' saves `a' and `aa' but not
    `a1'.  To match all variables beginning with `a', you must use an
    expression like `a.*' (match all sequences beginning with `a'
    followed by zero or more characters).

  * Line and column information is included in more error messages.

Summary of changes for version 0.77:
-----------------------------------

  * Improved help.  The command `help -i topic' now uses the GNU Info
    browser to display help for the given topic directly from the
    Texinfo documenation.

  * New function: chol -- Cholesky factorization.

Summary of changes for version 0.76:
-----------------------------------

  * Better run-time error messages.  Many now include line and column
    information indicating where the error occurred.  Octave will also
    print a traceback for errors occurring inside functions. If you
    find error messages that could use improvement, or errors that
    Octave fails to catch, please send a bug report to
    bug-octave@bevo.che.wisc.edu.

  * If gplot (or gsplot) is given a string to plot, and the string
    does not name a file, Octave will pass the string along to gnuplot
    directly.  This allows commands like

      gplot "sin (x)" w l, data w p

    to work (assuming that data is a variable containing a matrix of
    values).

  * Long options (--help, --version, etc.) are supported.

Summary of changes for version 0.75:
-----------------------------------

  * The documentation is much more complete, but still could use a lot
    of work.

  * The history function now prints line numbers by default.  The
    command `history -q' will  omit them.

  * The clear function now accepts regular expressions.

  * If gplot (or gsplot) is given a string to plot, and the string
    names a file, Octave attempts to plot the contents of the file.

  * New functions:

    history:

      run_history  -- run commands from the history list.
      edit_history -- edit commands from the history list with your
                      favorite editor.

    linear algebra:

      balance         -- Balancing for algebraic and generalized
                         eigenvalue problems.
      givens          -- Givens rotation.
      is_square       -- Check to see if a matrix is square.
      qzhess          -- QZ decomposition of the matrix pencil (a - lambda b).
      qzval           -- Generalized eigenvalues for real matrices.
      syl             -- Sylvester equation solver.

    control systems:

      is_symmetric    -- Check to see if a matrix is symmetric.
      abcddim         -- Check dimensions of linear dynamic system [A,B,C,D].
      is_controllable -- Check to see if [A,B,C,D] is controllable.
      is_observable   -- Check to see if [A,B,C,D] is observable.
      are             -- Solve algebraic Ricatti equation.
      dare            -- Solve discrete-time algebraic Ricatti equation.
      lqe             -- Kalman filter design for continuous linear system.
      lqr             -- Linear Quadratic Regulator design.
      lyap            -- Solve Lyapunov equation.
      dlyap           -- Solve discrete Lyapunov equation.
      tzero           -- Compute the transmission zeros of [A,B,C,D].

Summary of changes for version 0.74:
-----------------------------------

  * Formal parameters to functions are now always considered to be
    local variables, so things like

      global x = 0
      global y = 0
      function y = f (x) x = 1; y = x; end
      f (x)

    result in the function returning 1, with the global values of x
    and y unchanged.

  * Multiple assignment expressions are now allowed to take indices,
    so things like

      octave:13> [a([1,2],[3,4]), b([5,6],[7,8])] = lu ([1,2;3,4])

    will work correctly.

Summary of changes for version 0.73:
-----------------------------------

  * Saving and loading global variables works correctly now.

  * The save command no longer saves built-in variables.

  * Global variables are more reliable.

  * Matrices may now have one or both dimensions zero, so that
    operations on empty matrices are now handled more consistently.

    By default, dimensions of the empty matrix are now printed along
    with the empty matrix symbol, `[]'.  For example:

      octave:13> zeros (3, 0)
      ans = 

      [](3x0)

    The new variable `print_empty_dimensions' controls this behavior.
    
    See also Carl de Boor, An Empty Exercise, SIGNUM, Volume 25,
    pages 2--6, 1990, or C. N. Nett and W. M. Haddad, A
    System-Theoretic Appropriate Realization of the Empty Matrix
    Concept, IEEE Transactions on Automatic Control, Volume 38,
    Number 5, May 1993.

  * The right and left division operators `/' and `\' will now find a
    minimum norm solution if the system is not square, or if the
    coefficient matrix is singular.

  * New functions:

      hess   -- Hessenberg decomposition
      schur  -- Ordered Schur factorization
      perror -- print error messages corresponding to error codes
                returned from the functions fsolve, npsol, and qpsol
                (with others to possibly be added later).

  * Octave now prints a warning if it finds anything other than
    whitespace or comments after the final `end' or `endfunction'
    statement.

  * The bodies of functions, and the for, while, and if commands are
    now allowed to be empty.

  * Support for Gill and Murray's QPSOL has been added.  Like NPSOL,
    QPSOL is not freely redistributable either, so you must obtain
    your own copy to be able to use this feature.  More information
    about where to find QPSOL and NPSOL are in the file README.NLP.

Summary of changes for version 0.72:
-----------------------------------

  * For numeric output, columns are now lined up on the decimal point.
    (This requires libg++-2.3.1 or later to work correctly).

  * If octave is running interactively and the output intended for the
    screen is longer than one page and a pager is available, it is
    sent to the pager through a pipe.  You may specify the program to
    use as the pager by setting the variable PAGER.  PAGER may also
    specify a command pipeline.

  * Spaces are not always significant inside square brackets now, so
    commands like

      [ linspace (1, 2) ]

    will work.  However, some possible sources of confusion remain
    because Octave tries (possibly too hard) to determine exactly what
    operation is intended from the context surrounding an operator.
    For example:

    -- In the command 

         [ 1 - 1 ]

       the `-' is treated as a binary operator and the result is the
       scalar 0, but in the command

         [ 1 -1 ]

       the `-' is treated as a unary operator and the result is the
       vector [ 1 -1 ].

    -- In the command

         a = 1; [ 1 a' ]

       the single quote character `'' is treated as a transpose operator
       and the result is the vector [ 1 1 ], but in the command

         a = 1; [ 1 a ' ]

       an error message indicating an unterminated string constant is
       printed.

  * Assignments are just expressions now, so they are valid anywhere
    other expressions are.  This means that things like

      if (a = n < m) ... endif

    are valid.  This is parsed as:  compare `n < m', assign the result
    to the variable `a', and use it as the test expression in the if
    statement.

    To help avoid errors where `=' has been used but `==' was
    intended, Octave issues a warning suggesting parenthesis around
    assignments used as truth values.  You can suppress this warning
    by adding parenthesis, or by setting the value of the new built-in
    variable `warn_assign_as_truth_value' to 'false' (the default
    value is 'true').

    This is also true for multiple assignments, so expressions like

      [a, b, c] = [u, s, v] = expression

    are now possible.  If the expression is a function, nargout is set
    to the number of arguments for the right-most assignment.  The
    other assignments need not contain the same number of elements.
    Extra left hand side variables in an assignment become undefined.

  * The default line style for plots is now `lines' instead of
    `points'.  To change it, use the `set data style STYLE' command.

  * New file handling and I/O functions:

      fopen    -- open a file for reading or writing
      fclose   -- close a file
      fflush   -- flush output to a file
      fgets    -- read characters from a file
      frewind  -- set file position to the beginning of a file
      fseek    -- set file position
      ftell    -- tell file position
      freport  -- print a report for all open files
      fscanf   -- read from a file
      sscanf   -- read from a string
      scanf    -- read from the standard input

  * New built-in variables for file and I/O functions:

      stdin    -- file number corresponding to the standard input stream.
      stdout   -- file number corresponding to the standard output stream.
      stderr   -- file number corresponding to the standard error stream.

    The following may be used as the final (optional) argument for
    fseek: 

      SEEK_SET -- set position relative to the beginning of the file.
      SEEK_CUR -- set position relative to the current position.
      SEEK_END -- set position relative to the end of the file.

  * New function: setstr -- convert vectors or scalars to strings
    (doesn't work for matrices yet).

  * If possible, computer now prints the system type instead of
    always printing `Hi Dave, I'm a HAL-9000'.

  * Octave now properly saves and restores its internal state
    correctly in more places.  Interrupting Octave while it is
    executing a script file no longer causes it to exit.

  * Octave now does tilde expansion on each element of the LOADPATH.

  * A number of memory leaks have been plugged.

  * Dependencies for C++ source files are now generated automatically
    by g++.

  * There is a new command line option, -p PATH, that may be used to
    set Octave's loadpath from the command line.  It will override any
    value of OCTAVE_PATH found in the environment, but not any
    LOADPATH="path" commands found in the system or user startup files.

  * It is now possible to override Octave's default idea of the
    location of the system-wide startup file (usually stored in
    $(prefix)/lib/octave/octaverc) using the environment variable
    OCTAVE_HOME.  If OCTAVE_HOME has a value, Octave will look for
    octaverc and its M-files in the directory $OCTAVE_HOME/lib/octave.

    This allows people who are using binary distributions (as is
    common with systems like Linux) to install the real octave binary
    in any directory (using a name like octave.bin) and then install
    a simple script like this

      #!/bin/sh
      OCTAVE_HOME=/foo/bar/baz
      export OCTAVE_HOME
      exec octave.bin

    to be invoked as octave.


Summary of changes for version 0.71:
-----------------------------------

  * Much improved plotting facility.  With this release, Octave does
    not require a specially modified version of gnuplot, so gnuplot
    sources are no longer distributed with Octave.  For a more
    detailed description of the new plotting features, see the file
    PLOTTING.

  * New plotting commands:

      plot            -- 2D plots
      semilogx        -- 2D semilog plot with logscale on the x axis
      semilogy        -- 2D semilog plot with logscale on the y axis
      loglog          -- 2D log-log plot
      mesh            -- 3D mesh plot
      meshdom         -- create matrices for 3D plotting from two vectors
      contour         -- contour plots of 3D data
      bar             -- create bar graphs
      stairs          -- create stairstep plots
      polar           -- 2D plots from theta-R data
      grid            -- turn plot grid lines on or off
      xlabel, ylabel  -- place labels on the x and y axes of 2D plots
      sombrero        -- demonstrate 3D plotting
      gplot           -- 2D plot command with gnuplot-like syntax
      gsplot          -- 3D plot command with gnuplot-like syntax
      set             -- set plot options with gnuplot syntax
      show            -- show plot options with gnuplot syntax
      closeplot       -- close stream to gnuplot process
      purge_tmp_files -- delete temporary files created by plot command

  * Other new commands:

      ls, dir         -- print a directory listing
      shell_cmd       -- execute shell commands
      keyboard        -- get input from keyboard, useful for debugging
      menu            -- display a menu of options and ask for input
      fft             -- fast fourier transform
      ifft            -- inverse fast fourier transform

  * Strings may be enclosed in either single or double quote
    characters.  Double quote characters are not special within single
    quote strings, and single quotes are not special within double
    quote strings.

  * Command name completion now works for M-file names too.

  * Better help and usage messages for many functions.

  * Help is now available for functions defined in M-files.  The first
    block of comments is taken as the text of the help message.

  * Numerous changes in preparation to support dynamic loading of
    object files with dld.

  * Bug fixes to make solving DAEs with dassl actually work.

  * The command `save file' now saves all variables in the named file.

  * If do_fortran_indexing is 'true', indexing a scalar with
    [1,1,1,...] (n times) replicates its value n times.  The
    orientation of the resulting vector depends on the value of
    prefer_column_vectors.

  * Things like [[1,2][3,4]] no longer cause core dumps, and invalid
    input like [1,2;3,4,[5,6]] now produces a diagnositic message.

  * The cd, save, and load commands now do tilde expansion.

  * It's now possible to clear global variables and functions by name.

  * Use of clear inside functions is now a parse error.

Summary of changes for version 0.70:
-----------------------------------

  * Better parse error diagnostics.  For interactive input, you get
    messages like

      octave:1> a = 3 + * 4;

      parse error:

          a = 3 + * 4;
                  ^

    and for script files, the message includes the file name and input
    line number:

      octave:1> foo

      parse error near line 4 of file foo.m:

          a = 3 + * 4;
                  ^

  * New built-in variable PS2 which is used as the secondary prompt.
    The default value is '> '.

  * New file, octave-mode.el, for editing Octave code with GNU Emacs.
    This is a modified version of Matthew R. Wette's matlab-mode.el.

  * Better support for missing math functions.

  * User preferences are now cached in a global struct so we don't
    have to do a symbol table lookup each time we need to know what
    they are.  This should mean slightly improved performance for
    evaluating expressions.

Summary of changes for version 0.69:
-----------------------------------

  * Multiple assignments are now possible, so statements like

      a = b = c = 3;
      a = b = c = [1,2;3,4];

    or

      c = (a = (b = 2) * 3 + 4) * 5

    are legal, as are things that have even more bizarre effects, like

      a(4:6,4:6) = b(2:3,2:3) = [1,2;3,4];

    (try it).

  * Improved parsing of strings (but they still don't work as matrix
    elements).

  * An M-file may now either define a function or be a list of
    commands to execute.

  * Better detection and conditional compilation of IEEE functions
    isinf, finite, and isnan.

  * Replacements for acosh, asinh, atanh, and gamma from the BSD math
    library for those systems that don't have them.

Summary of changes for version 0.68:
-----------------------------------

  * New functions:

      eval  -- evaluate a string as a sequence of Octave commands. 
      input -- print a prompt and get user input.

Summary of changes for version 0.67:
-----------------------------------

  * New functions:

      find -- return the indices of nonzero elements.

  * Zero-one style indexing now works.  For example,

      a = [1,2,3,4];
      b = a([1,0,0,1])

    sets b to the first and fourth elememnts of a.

    Zero-one style indexing also works for indexing the left hand side
    of an assignment.  For example,

      a = rand (1,2;3,4);
      a([0,1],:) = [-1,-2]

    sets the second row of a to [-1 -2]

    The behavior for the ambiguous case

      a = [1,2,3,4];
      b = a([1,1,1,1]);

    is controlled by the new global variable `prefer_zero_one_indexing'.
    If this variable is equal to 'true', b will be set to [1 2 3 4].
    If it is false, b will be set to [1 1 1 1].  The default value is
    'false'.

  * Using the new global variable `propagate_empty_matrices', it is
    possible to have unary andy binary operations on empty matrices
    return an empty matrix.  The default value of this variable is
    'warn', so that empty matrices are propagated but you get a
    warning.  Some functions, like eig and svd have also been changed
    to handle this.

  * Empty matrices can be used in conditionals, but they always
    evaluate to `false'.  With propagate_empty_matrices = 'true', both
    of the following expressions print 0: 

      if  [], 1, else 0, end
      if ~[], 1, else 0, end

  * Octave no longer converts input like `3.2 i' or `3 I' to complex
    constants directly because that causes problems inside square
    brackets, where spaces are important.  This abbreviated notation
    *does* work if there isn't a space between the number and the i,
    I, j, or J.

Summary of changes for version 0.66:
-----------------------------------

  * Logical unary not operator (~ or !) now works for complex.

  * Left division works.

  * Right and left element by element division should work correctly
    now.

  * Numbers like .3e+2 are no longer errors.

  * Indexing a matrix with a complex value doesn't cause a core dump.

  * The min and max functions should work correctly for two arguments.

  * Improved (I hope!) configuration checks.

  * Octave is now installed as octave-M.N, where M and N are version
    numbers, and octave is a link to that file.  This makes it
    possible to have more than one version of the interpreter installed.

Summary of changes for version 0.63:
-----------------------------------

  * The reshape function works again.

  * Octave now converts input like `3.2i' or `3 I' or `2.3e5 j' to be 
    complex constants directly, rather than requiring an expression
    like `3.3 * i' to be evaluated.

Summary of changes for version 0.61:
-----------------------------------

  * Octave has been successfully compiled using gcc 2.3.3 and libg++ 2.3.
    on a 486 system running Linux.

  * The win_texas_lotto function is now called texas_lotto (it's a
    script file, and win_texas_lotto.m is too long for some Linux and
    System V systems).

Summary of changes for version 0.57:
------------------------------------

  * The C-like formatted print functions printf, fprintf, and sprintf
    finally work. 

Summary of changes for version 0.56:
------------------------------------

  * By default, octave prints a short disclaimer when it starts.
    (You can suppress it by invoking octave with -q).

  * You can keep octave from reading your ~/.octaverc and .octaverc
    files by invoking it with -f.

  * When returning two values, eig now returns [v, d] instead of
    [lambda, v], where d is a diagonal matrix made from lambda.

  * The win_texas_lotto function now produces a sorted list.

  * New functions:

      expm -- matrix exponential.
      logm -- matrix logarithm.

Summary of changes for version 0.55:
------------------------------------

  * The following (C-style) backslash escape sequences work in quoted
    strings (useful(?) with printf()):

      \a  bell         \r  carriage return
      \b  backspace    \t  horizontal tab
      \f  formfeed     \v  vertical tab
      \n  newline      \\  backslash

  * Use of `...' at the end of a line will allow a statement to
    continue over more than one line.

  * The names `inf' and `nan' are now aliases for `Inf' and `NaN',
    respectively.

  * New functions:

      casesen -- print a warning if the luser tries to turn off case
                 sensitivity.
      median  -- find median value.
      norm    -- compute the norm of a matrix.
      sort    -- sort columns.

  * New variable, `silent_functions'.  If silent_functions == 'true',
    the results of expressions are not printed even if they are not
    followed by a semicolon.  The disp() and printf() functions still
    result in output.  The default value for this variable is 'false'.

  * New variable `return_last_value_computed'.  If it is 'true',
    functions defined in script files return the last value computed
    if a return value has not been explicitly declared.  The default
    value for this variable is 'false'.

Summary of changes for version 0.52:
------------------------------------

  * Name completion works for function and variable names currently in
    the symbol tables.  Coming soon: completion for names of functions
    defined in script files but not yet compiled. 

  * The initial value of do_fortran_indexing is now false, and the
    initial value of prefer_column_vectors is now true.  Swap the
    values of these variables if you want behavior that is more like
    Matlab.

  * All script files check the number of input arguments before doing
    much real work.

  * The identifiers `i' and `j' are now also names for sqrt(-1).
    These symbols may be used for other purposes, but their original
    definition will reappear if they are cleared.

  * The symbol tables are now implemented with hash tables for faster
    searching. 

  * A small amount of help is now available for most built-in
    operators, keywords and functions.  Coming soon: help for script
    files.

  * Without any arguments, the help command now lists all known
    built-in operators, keywords and functions.

  * Generic parse errors are now signalled by `Eh, what's up doc?',
    which is closer to what Bugs actually says.

  * The who command now only prints variable names by default.
    Use the -fcn (or -fcns, or -functions) switch to print the names of
    built-in or currently compiled functions.

Summary of changes for version 0.51:
------------------------------------

  * Major overhaul of array indexing.

  * The colloc function actually works now.

Summary of changes for version 0.50:
------------------------------------

  * The lsode and dassl functions now return the states only,
    instead of the time and the states, so you must keep track of
    the corresponding times (this is easy though, because you have
    to specify a vector of desired output times anyway).

  * Solution of NLPs with NPSOL now works on the SPARC.

  * New keywords `endif', `endfor', `endfunction', `endif', and
    `endwhile', which allow for better diagnostics.  The `end' keyword
    is still recognized.  All script files have been changed to use
    these new keywords in place of `end'.

  * It is now possible to uninstall Octave by doing a `make uninstall'
    in the top level directory.

  * The Makefiles are much closer to conforming with GNU coding standards.

  * New functions:

      win_texas_lotto  -- produce six unique random numbers between 1 and 50.
      quad             -- numerical integration.
      lu               -- LU factorization
      qr               -- QR factorization
      dassl            -- Solution of DAEs using DASSL.

  * New files:

      THANKS -- A list of people and organazations who have supported
                the development of Octave.

      NEWS   -- This file, listing recent changes.

  * Help is now available at the gnuplot prompt.