2006-09-05 Matt Newville <firstname.lastname@example.org>
* Version 1.3.0:
- play with max entropy / fit fft components
- many bug fixes for array initialization and data corruption.
- chi_noise() nows works much more reliably
2005-09-28 Matt Newville <email@example.com>
* Version 1.2.9:
- added warning when a user-defined 'fitting variable' is accessed
as a 'global variable' by an internal command by
2005-09-14 Matt Newville <firstname.lastname@example.org>
* Version 1.2.8:
- fixed configure script looking for curses/termcap libraries.
- many changes in Windows distribution.
- fixed bug in iff_shell.c so that it now works correctly on
64-bit linux. Also added '-fPIC' to FFLAGS arguments in
Makefiles so that wrappers build on 64bit linux systems.
- add '--with-64bit' to PGPLOT_install so that pgplot may be
built on 64bit linux.
- changed (fixed??) background refinement in feffit() to have
the spline range extend over a wider k-range.
2005-04-07 Matt Newville <email@example.com>
* Version 1.2.7:
- add &status variable to hold many status flags from commands.
The &status variable will hold an integer:
1 being a warning (including 'unknown keyword' for commands)
2 something caused the command to stop prematurely
3 a more serious error, generally including hitting an
internal limit, but also including 'file not found'
math syntax errors ('a/*b') sets &status to 2
math evaluation erors ('log(0)') sets &status to 2
unknown group or no x-array in most commands sets &status to 2
1. &status value is NOT reset at each command (so that warnings
persist through macros, for example), and need to be reset by
2. the &status value will never be lowered, so that the most
serious of &status levels is reflected.
The whole &status mechanism is designed primarily for
wrappers where this sequence is easy enough to encapsulate:
react to &status (and maybe look at echo buffer at this point)
reset &status to 0
- added math function 'penalty(x,lo_val,hi_val)' which returns:
0 for lo_val <= x <= hi_val
abs(x-lo_val) for x < lo_val
abs(x-hi_val) for x > hi_val
this is useful for setting a restraint on a parameter so that it
is preferred to fall within a given range. The penalty function
gives a more gentle penalty than
and is better suited for chi-square sums (and will allow
uncertainties to be estimated).
- fixed bug in iff_wrdata.f so that up to 99 arrays can be written
with the write_data() command. This value is set in consts.h as
max_outarr and exposed as the scalar '&max_output_cols'.
- upgraded readline library to readline 5.0
2004-07-08 Matt Newville <firstname.lastname@example.org>
* Version 1.2.6: assorted bug fixes, improvements
- fixed important bug in feffit() and ff2chi() that could end up
with polluted path lists. Doing a second fit in a session using
fewer paths would use some leftover paths. The path list is now
- reading of feff.inp in feff6 now uses iread() so that
feff6 no longer fails because of line endings that don't match
the standard for that platform.
- plot markers and arrows now support color, with
- location of the plot legend now more customizable using
one of two equivalent methods:
plot(...., key='data', key_x=0.9, key_y0=0.9,key_dy=0.1)
&plot_key_x = 0.9
&plot_key_y0 = 0.9
&plot_key_dy = 0.1
where 'key_x' is the x-location of the legend 'key_y0' is the
y-location of the first key, and 'key_dy' is the vertical offset
2004-01-22 Matt Newville <email@example.com>
* Version 1.2.5: bug fixes, improved feff path lists, fixed
problems with spline() clamps, restraints in feffit(), altered
screen messages for autobk/feffit/feff6, and added random()
command to generate arrays of random numbers.
- path lists for feffit(), ff2chi(), erase(), and show() can now
range over undefined paths, so that
will use all *currently defined* paths between 1 and 100
(inclusive), even if only paths 2, 8, and 55 are defined. Here
are some related changes:
o path indices are strictly checked to be in the range 1..9999.
o since the maximum path index is 9999, the list '1-9999' means
'all defined paths'. This can be abbreviated as 'all', as in:
which will sum all curently defined paths.
o erase(@paths) and erase(@path=all) are now equivalent, both
erasing all defined paths.
o when erasing a path, the feff data associated with a path will
be freed (if it's not in use by another path) so that new feff
paths can be used.
o the formatting of the output of 'show @path' and 'show @paths'
has changed. R is now reported (not just reff and delr), each
parameter is listed on its own line, and the path parameters
3rd, 4th, ei, and dphase are reported only if defined.
- added random() command, to generate an array of random numbers,
using uniform, gaussian, or normal distributions. syntax is:
random(output=my.random, npts=200, dist=normal, sigma=2.)
where sigma will be (approximately) the standard deviation of the
normal and gaussian distributions, or the scale of the uniform
distribution. Sigma is 1 unless explicitly set. You can optionally
reset the seed of the random number generator with the seed
keyword. The gaussian and normal distribution appear very
similar, and can both be used to simulate noise in data.
- chi_noise() now generates 'kmax_suggest', an estimate of the
maximum usable k-value for the data, based on the estimated noise
in the data. In early tests, this appears to be a reasonable
estimate for 'noisy data', and a bit pessimistic for clean data.
Please use this value as a guide.
- clamps in spline() now try to bring chi() to zero at kmax/emax
and kmin/emin instead af last and first data point.
- an error in feffit() for how multiple restraints were applied
has been fixed. multiple restraints can now be applied and mixed
freely with multiple k-weights and multiple data sets.
- improved screen messages printed by autobk/feffit/feff6:
o replaced all print* statements with 'call echo()'
o messages 'echo'ed to the screen are now 'tee'd to the
run files autobk/feffit/feff6.run
o replace all 'stop' statements with a 'call echo()' and
the creation of an error file autobk/feffit/feff6.err.
If this file exists after a run of autobk/feffit/feff6,
then an error did occur; after successful runs, the error
file will not exist.
2003-10-28 Matt Newville <firstname.lastname@example.org>, with Charlotte's help
* Version 1.2.4: a long list of bug fixes, increases in limits,
and other improvements
- n_varys now correctly reported by feffit()
- increased many limits on string sizes, now generally 256
characters for most internal strings and for text variables.
- increased number of paths to 256 and feff files to 512, and
eliminated the distinction between 'total number of paths'
and 'total number of paths for a particular data set'.
- added program variables &n_guess, &n_scalars, &n_arrays,
&n_scalars_set, &n_scalars_def, &n_arrays_set, &n_arrays_def,
and &n_strings that contain the number of each kind of variable
currently in use, and the variables &max_scalars, &max_arrays,
&max_strings, &max_paths, &max_varys, and &max_data_sets to
contain the maximun number of each of these kinds of data.
- added show @limits to show all the '&n_*' and '&max_*' variables
above plus current memory limits in &maxpts and &heap_free.
- changed autobk, feffit, and feff6 to take command-line argument
for input file:
and so on. autobk and feffit name their log according to the
input file name.
- fixed annoying bug that ignored entire lines that included an
EOF character, which might (on many unixes) be a valid last line
of a file.
- rewrote handling of repeated x-values when automatically sorting
data. Previously, data points with x(i) == x(i+1) were attempted
to be averaged. Now, the value of x(i+1) is increased slightly
(typically 1% of x(i+2)-x(i)). This fixed odd behaviour of
spline() when handed strangely ordered data by athena.
- amended $column_label output of read_data() to be able to handle
column labels longer than a single string can hold (256
characters, currently). The scalar ncolumn_label now holds the
number of text strings used. The first string is held in
both $column_label and $column_label1. Additional text strings
of $column_label2, ..., $column_label9 are used, as needed.
- added provisional f1->f2 capabilities to diffkk. This passes a
very simple test, but is not very well-tested at this point, and
may require some tweaks. Consider it a test version, and report
2003-09-15 Matt Newville <email@example.com>
* Version 1.2.3: bug fixes
- setting path_index now correctly makes the selected
path 'the current path', and causes re-evaluation of
defined variables and path parameters according to
that path's values for reff, degen, etc.
- fixed error with spline() not passing pre1,pre2 to
2003-08-21 Matt Newville <firstname.lastname@example.org>
* Version 1.2.2: bug fixes, build improvements
- fixed error with k-space feffit() fitting so that
k-weighting is now really used!!
- fixed error with using reff in defined math expressions
- increased size of math expressions to 256 operations/values.
- changed restraint mechanism to work slightly better with
multiple data sets
- fixed error in read_data() that allowed core-dumps when
reading files larger than the maximum array size.
- switched configure to use autoconf 2.53, including 'newer'
automated searches for fortran libraries, setting default f77
compiler to g77, which should help builds on Mac OS X.
- removed fortran subdirectory -- an existing, working fortran
compiler is needed to build ifeffit.
- fixed iconf_pgplot to check that PGPLOT directories listed
in the PGPLOT make really exist (and are directories) before
including them in the link arguments.
- upgraded to readline 4.3
- upgraded python installation to use Pmw 1.1
2003-05-19 Matt Newville <email@example.com>
* Version 1.2.1: bug fixes:
- precision error in feff6/rdinp.f prevented the
pathfinder from working (completely MN's fault
-- not original to feff6!!!)
- fixed bug in writing output fit arrays in feffit()
for multiple-k-weight fits, and with do_bkg=true.
- slightly outdated Ifeffit.py had been included in
1.2.0, only noticed when building the Windows version.
2003-05-14 Matt Newville <firstname.lastname@example.org>
* Version 1.2.0: declared this version as the first
'stable' version of the 1.2 series,
bug fixes since 1.0077:
- extraneous print statements in iff_set.f removed.
- background function '*.kbkg' was not correctly computed
for output in feffit()
2003-04-30 Matt Newville <email@example.com>
* Version 1.0077: several changes and improvements, ranging
from bug fixes to significant enhancements:
- plot() now interprets 'key=""' correctly to mean to not show a
- plot(reset) now works, resetting the color and style tables to
the original values.
- read_data() no longer automatically sorts on the first column
(or any other column) by default, but rather leaves data as read
unless explicitly sorted.
- fixed incorrect reporting of a parsing non-error (and some
related mistakes) on several math expression idioms involving
nested multi-argument functions, as in this deglitching gem, which
now works as expected to remove the data point closest to x=badx:
new.x = join(slice(old.x,1,badx-1),slice(old.x,badx+1,npts(old.x)))
new.y = join(slice(old.y,1,badx-1),slice(old.y,badx+1,npts(old.y)))
- spline() and bkg_cl() now uses a new rebinning/interpolation
algorithm to convert data from chi(E) to chi(k). For
finely-spaced energy data such as from QEXAFS measurements, this
algorithm averages over all energy points within the corresponding
k range, given by [k-kgrid/2k,k+kgrid/2). Where no points lie
within the k-range, a 3-point interpolation using the closest
points is used. The 'interp' keyword to spline() and bkg_cl()
have been deprecated and are silently ignored.
- The rebin() function (which works like other interpolation
functions) now uses this rebinning/averaging algorithm as well.
- f1f2() now looks up and uses the 'known core-width of an edge'
by default. That is, if the atom given has an edge energy in the
supplied energy range, the corresponding core-width is used.
- Other minor improvements were made to f1f2(), and especially in
the lorenztian convolution so that it works more reliably for
large, uneven arrays.
- Cromer-Libermann data updated to have slightly more significant
digits, which seems to be important to avoid 'weird points', which
often showed up exacly at absorption edges.
- feffit() and minimize() now list 'bad variables' after a fit for
which error bars could not be estimated.
- feffit() now produces correct chi(R) data when input chi(k) data
is not on a k=0.05Ang^-1 grid.
- feffit(bkg=true) now works with multiple-data-set fits.
- feffit() now "correctly" sums the number of independent points
for multiple-data sets as the sum of independent points in each
data set. There is also a minimal checks whether the chi(k) data
is repeated in a single fit (check== is the name the same or is
the sum of squares of chi(k) over the fit k-range the same?).
- feffit() now supports multiple-k-weights, with syntax
feffit(..., kweight=2, kweight=0, kweight=4,...)
which will use _all_ these k-weights (up to 5 allowed). For
simple implementations of multiple-data-set fits on the same
dataset with only the kweight being different, the gives
numerically equivalent results. Important caveats are that the
repored error bars _always_ reflect the fact that only 1 data set
is really is used, and that the rest of the k- and r-range
parameters are forced to be uniform. There may be some subtle
differences when using other fitting add-ons, such as 'bkg=true'
(ie, only one bkg function will be used) or restraints. Output
arrays generated for this form of multiple-k-weight fits will use
the FIRST kweight listed.
2003-01-21 Matt Newville <firstname.lastname@example.org>
* Version 1.0076: several changes and improvements over the past months
fixed plotting bug in which all legends were not always shown,
especially when making hard copies
added 'unguess' command which simply converts all 'guess'ed
variables to 'set' values. No arguments, no options.
added 'window' command to generate an FT window without actually
doing the FT. requires an x array, but otherwise acts very much
fixed bug in reading 1-row data files. The results are now
converted to scalars, with names of the form $group_prefix
allow (consistently) '&', ':', and '?' in variable and group
added math functions for array manipulations:
x = nofx(m.a,x0) index of m.a with value closest to x0
m.x = join(m.a,m.b) concatenate 2 arrays
m.x = slice(m.a,x0,x1) sub-array m.a[x0:x1], using fortran
array conventions (first element is 1).
many further tweaks to configuration procedure:
- the shell script iconf_iff now automatically builds
src/lib/sys.h and src/lib/maxpts.h at configure time.
- the shell script iconf_flib now fixes the poorly sorted
FLIBS on most Linux systems, that caused some problems with
- there is now a 'configure; make' procedure in src/macosx,
with 'configure' now building a Makefile (which still might
need tweaking). This uses 'uname -s' for setting
darwin-specific code. Aside from GNU Readline issues, this
build procedure should work on many Unix systems.
2002-11-12 Matt Newville <email@example.com>
* Version 1.0075a: tweaked build procedure so that all
system-specific stuff is put into src/lib/sys.h, which is
built by the shell script iff_config.
The MacOS X directory (src/macosx) now has its own version of this
script, and should now work.
2002-11-01 Matt Newville <firstname.lastname@example.org>
* Version 1.0075: switched wrapper generation to swig1.3,
so that perl extension works with gcc3.2 (as on RedHat Linux 8.0)
The Tcl extension has not been updated.
added feff6 to source distribution
reworked meaning of 'erase @path' to only set path parameters
to '', and not destroy corresponding feff data, as this could
trash other paths.
tweaked build procedure a little.
2002-09-18 Matt Newville <email@example.com>
* Version 1.0074: fixed bug in FFTs with phase-correction.
fixed bug in reporting of path parameters that values of sigma2
depending on debye() and eins() functions were not always shown
to be updated.
fixed bug in feffit (now version 2.98) in which parameters
depending on debye(t,th) or eins(t,th) were not always updated
correctly when t and th were literal constants. This
especially affected non-fits.
2002-08-16 Matt Newville <firstname.lastname@example.org>
* Version 1.0073: added 'interpolation options' to spline()
and bkg_cl(), with syntax 'interp=line', 'interp=quad', or
'interp=cubic' to change the final interpolation scheme for
converting data from chi(E) to chi(k)
The k-grid spacing (previously set at 0.05Ang-1) can now be
set as a build time option in config.h with the parameter
'qgrid'. This and the length of the FFT arrays may
eventually become run-time parameters. Changing these values
should be considered as highly experimental.
changed default arraysize to 'medium' (ie, 4096 elements).
fixed bug in call to 'coth' math function.
improved Grwin/PGPLOT on Windows to auto-raise the plotting
window on all plot, cursor, and zoom commands.
improvements to GIFEFFIT.
2002-06-24 Matt Newville <email@example.com>
* Version 1.0072: rewrote macro execution routine to remove
the use of recursion, which was causing nested macros to
not work correctly on windows.
- some major changes to configuration and build procedure,
especially in that perl and python wrappers are no longer built
during 'configure; make', but must be made separately, and after
- added a &heap_free variable, which contains the number of
available array elements. When this gets lower than the maximum
array length (typically 8192), bad things can start to happen.
Normally, &heap_free starts at 2**21 ~= 2million array elements.
- fixed some errors in 'erase()' so that now the constants pi and
etok cannot be erased, and 'erase @scalars' does not erase system
variables (ie, '&*'). Also sped up execution times for
'erase @arrays' and '@groups'.
2002-04-02 Matt Newville <firstname.lastname@example.org>
* Version 1.0071: changed configuration and some plotting
behavior, updated documentation
- PGPLOT_install will now install the PNG driver unless explicitly
told to (with --no-png). Similarly, the GIF driver can be turned
off with --no-gif.
- configure was changed to check which PGPLOT drivers are used and
to build the list of libraries to link with explicitly. This runs
the shell script ./test_pgplot to spit out the 'link libraries'
for PGPLOT. It is not extensively tested across unixes.
Specifically, it works with PGPLOT_install and with pre-built
PGPLOT lib on Mac OS X.
configure also now runs on Mac OS X, though the returned value for
FLIBS seems bad. To fix this,
is necessary. This builds the C, and fortran libs and apps, but
the extensions seem to need to be built after the installation of
the main library. This is still under investigation -- for now,
the build in src/macosx also works.
- the ifeffit variable $plot_devices now contains a list of all
available plotting devices for the installation.
- error bars can be included in a plot:
plot( dat.x, dat.y, dy=dat.dy)
where dat.dy is the uncertainty in dat.y. Error bars in x:
'dx=dat.dx' is also supported.
- the plot command now supports a 'one array' version:
which will plot dat.z as the y- array against the _previous_
x-array, or against an dummy array of (1,2,..) if it is the
first plot in a series. That is,
plot energy,xmu, group='data'
now does what you would expect:
plot (data.energy, data.xmu)
plot (data.energy, data.bkg)
2002-03-05 Matt Newville <email@example.com>
* Version 1.0070: feffit 2.97 now appears to work completely
and is included in build and installation. Like autobk, this
is a 'transitional version' between the classic version and
a full 'ifeffit application', and keeping the outward appearance
of the 'classic version' and using the ifeffit libraries only
for the low-level numeric stuff. Feffit currently uses its own
version of 'the math parser' as well.
- added special functions to make the generation of gaussian,
lorentzian, and pseudo-voight lineshapes easier:
my.y = gauss(my.x, centroid, sigma)
my.y = loren(my.x, centroid, sigma)
my.y = pvoight(my.x, centroid, fwhm,eta)
see documentation (soon to be updated!).
- added 'group' keyword to save to give a specified group to
include in the save file. If used, all groups must be listed:
- fixed zoom(show) to work again.
- minor updates to gifeffit.
2002-02-18 Matt Newville <firstname.lastname@example.org>
* Version 1.0069: added feff tables, made several
changes to aid portability to Win32 and MacOS X(!),
- included fefftab directory of Feff Tables. Currently
these are only read by internal routines, and cannot
(yet) be manipulated in the main interpreter.
- provisional (ie, not very well tested) support for
phase-corrected Fourier Transforms using feff tables.
fftf(xx.chi, pc_edge = 'Ni K')
- added 'force_read' keyword to path() to force the
re-reading of a feffNNNN.dat file (in case it may have
changed since it was last read).
- improved behavior of load() command for files and/or
directories with spaces in their names (use quotes!!),
and force surrounding quotes on startup file (for Win32
startup file of C:\Program Files\ifeffit\startup.iff)
- addded 'macosx' directory and Makefile to build for
MacOS X while avoiding 'configure;make'. This relies
heavily on installin of Fink tools for MacOS X. See
README.MACOSX and src/macosx/README for details.
Build instructions are effectively reduced to
'cd src/macosx; view Makefile; make'
This Makefile might actually be a useful starting point
for other Unix and Unix-like systems that have some Unix
tools in place but does not 'configure' correctly.
- tweaked iff_shell.c to better link against a known
installation of readline 4.2 (for MacOS X). Enough linux
distributions still distribute 2.2 that it's just easier
to always build a known version under 'normal build'.
- feffit is now included in the build tree (src/feffit),
compiles as fully double precision, and uses some of the
routines in src/lib/. But this version of feffit is still
SEVERELY BROKEN at the moment. Please don't use it and
complain about it!!
2002-02-07 Matt Newville <email@example.com>
* Version 1.0068: bug fixes
- fixed severe parse error for expressions containing min()
and max() functions.
- fixed error (yet another!) in setting of path degeneracy.
Really, this should work now.
2002-01-27 Matt Newville <firstname.lastname@example.org>
* Version 1.0067: several additions
- changed cursor / zoom routines from pgplot's pgcurs to pgband,
showing 'zoom box' on zoom, and allowing vertical lines or
cross-hairs for cursor selection. Seems to not work on Win32.
- added additional plotting commands: plot_marker(), plot_text(),
plot_arrow(), each of which draw to the plot window:
which is equivalent to
plot (text_x=x, text_y=y, text=text_string)
arrows can have adjustable size, barb fraction, and angle of
arrowhead, or no head at all:
plot_arrow(x1,y1,x2,y2, size=3, barb=0.4, angle=45)
- added storage of and access to the central-atom phase-shift for
a feff path. The command get_path() will now generate arrays for
k, amp, phase, lambda and cphase with the 'do_arrays' keyword.
- fftf() phase-corrected FTs now use a specified feff path, and
either the full phase-shift or central-atom phase-shift:
fftf(...., pc_path = path_index, pc_caps / pc_full )
- added 'sort' keyword to read_data to sort the data so that a
selected column (by column number only) is monotonically
- improved spline() and pre_edge() handling of unsorted data
(sorting it internally) and for data in keV.
2002-01-07 Matt Newville <email@example.com>
* Version 1.0066: bug fixes
more complete enforcement of '&maxpts' internally. Now
writing 'maxpts.h' at configure time and using as an
include file for almost all array size definitions.
switched python module from using compile.py to using
distutils for build process.
changed return of the C library function iff_scaval from *double
to a plain double. ifftest2.c changed accordingly.
changed rule for determining comment lines in data files from "a
comment character in the first character on a line" to "a comment
character in the first non-blank and non-tab character on a line".
comment characters can be '#%;' or a user-defined character.
2002-01-01 Matt Newville <firstname.lastname@example.org>
* Version 1.0065: provisional fixing of multiple bugs and annoyances:
added '&maxpts' internal scalar to give maximum array size. This
can now be used by perl/python modules to generate arrays.
fixed severe bug in Ifeffit.pm causing get_scalar to not work
(exit status of C's get_scalar had changed to 0 for success).
fixed errors in ifftest2.c, and in iff_scaval()
added initialization string to perl/python modules so that the
interpreter is ensured to be initialized when loaded.
improved the auto-scaling of the optional standard used in
spline() -- it was essentially hardwired previously. Also added
a keyword for 'r1st' -- distance beyond rbkg for first-shell
amplitude matching. Added simple examples for using standard in
spline() and for using end-point clamps.
moved this ChangeLog back to root directory, altered BuildBuild to
use it from root position for library build.
updated athena version to
removed plotting from diffkk executable, so that it can compile
simply under --with-f2c compile option.
2001-10-17 Matt Newville <email@example.com>
* Version 1.0064: provisional (tested, but not completely) support
for multiple-data-set fitting in feffit(). This appears easier
than I originally thought, and more useful than trying to support
a hybrid stand-alone feffit. Fixed lots of little problems
with path definitions along the way.
diffkk() command still incomplete
2001-10-05 Matt Newville <firstname.lastname@example.org>
* Version 1.0063: preliminary support for diffkk standalone
(unstable but can be made to work) and for dkk() function in
math parser (the diffkk command inside ifeffit proper is not
yet working right).
fixed a few bugs, including 'show @paths' possibly showing
the wrong reff/degen values.
several updates to reference doc
improved _fortran_ interface with ifeffit.inc file.
2001-09-07 Matt Newville <email@example.com>
* Version 1.0062: several minor changes, mostly to improve
documentation and python interface.
updated athena version to B Ravel's latest.
added provisional inclusion of diffkk into the ifeffit build
system (working, poorly tested -- no plotting as of yet).
2001-07-09 Matt Newville <firstname.lastname@example.org>
* Version 1.0061: added B Ravel's athena perl GUI program,
fixed bug in handling screen echo with log files.
2001-07-06 Matt Newville <email@example.com>
* Version 1.0060: improved C interface, adding functions
iff_strval() and iff_scaval() to return pointers to ifeffit
strings and scalars. Also, now allowing '#' characters in
improved lorenztian and gaussian convolution functions
to allow non-evenly spaced data to be properly convolved.
turned off default startup message from library -- now
assumed to be handled by calling program
eliminated need for PGPLOT_DEV environmental variable, by
defaulting to the '/xserve' persistent window and to '/gw'
for Win32 systems.
added --with-arraysize=[tiny, small, medium, large] to
configure for customizing the maximum array size (which
dominates the amount of memory required). The default is
'medium', which allows arrays of up to 8192 elements, and
uses around ~16Mb of RAM for a typical 'spline()' and ~22Mb
for a typical 'feffit()' execution. For now, I consider this
to be much more efficient than necessary.
2001-06-08 Matt Newville <firstname.lastname@example.org>
* Version 1.0059: added options for different sizes for
different plot quantities (labels, markers, on-panel
2001-06-05 Matt Newville <email@example.com>
* Version 1.0058: fixed deep, weird parse error
essentially rewrote subroutine bkeys() to be somewhat
more sensible at dealing with nested parens and quotes.
print now treats '...' and "..." differently: "..."
will evaluate the enclosed string, '...' will not. No
other commands are affected by this change.
added 'key' keyword to plot() which gives a primitive
worked on reference manual some.
added '.norm' output to spline(), which was redefining
edge_step after a pre_edgg() call, but not updating the
2001-06-01 Matt Newville <firstname.lastname@example.org>
* Version 1.0057: several bug fixes and changes, including:
updated PAD library to avoid sign errors on crays, and to
be in sync with version used in Feff
updated use of restraints in feffit(), added keywords
restraint (and restraint_01 through restraint_09). Not fully
tested or documented, but appears to work in test cases.
added command 'bkg_cl()' to do pre-edge subtraction, and
normalization based on fit of data to Cromer-Libermann data.
Not fully tested or documented, but appears to work.
upgraded to GNU Readline4.2, made several improvements
in build system for non-linux Unix systems.
still puzzling over bug report for improper path parameter
values (based on old definitions?)
2001-04-13 Matt Newville <email@example.com>
* Version 1.0056: debye bugfix, win32 update.
fixed several related bugs associated with debye(), eins(),
and 'path_index' that were preventing debye() and eins()
to be correctly evaluated at all times. 'path_index'
stores user-chosen path index for a path, and is used by many
internal functions and commands, including debye() and eins().
switched Win32 to GrWin 0.997, which now builds and runs
correctly, and fixes problems with cursor()/zoom() so that
these now work on Win32.
Makefile is now generating installation location for gifeffit,
instead of relying on a hardwired /usr/local/bin.
2001-04-11 Matt Newville <firstname.lastname@example.org>
* Version 1.0055:
fixed several bugs discovered by building Win32 DLL,
mostly due to array-bounds. Win32 DLL and command-line
program are both now working again.
vastly improved mechanism to add an additional scalar
value to function-to-minimize in feffit/minimize.
2001-03-25 Matt Newville <email@example.com>
* Version 1.0054: bug fixes, improved build system
fixed bug in minimize() that broke many fits.
fixed bug recently added to write_data() that
prevented 'string globs' ('$title_*') from working
fixed 'TERMLIB' setting for linking against readline.
reworked building of perl Makefile.PL, now using
ExtUtils::F77 (included with distribution).
added --enable-python, --enable-perl, --enable-tcl,
and --enable-all switches to configure to automate the
building of the optional wrappers for these interpreted
languages. All are enabled by default. So farw this
works, but there are still a few issues to work out:
(support for python2.0, some hardwired install
directories, untest for old installations, and no
installation directory of tcl extension).
2001-03-22 Matt Newville <firstname.lastname@example.org>
* Version 1.0053: added range() function in math parser
my.x = range(0.0, 1.0, 0.05)
now creates an array of 21 elements: 0.0, 0.05, ..., 1.00
array will go up to but not exceed the value stop:
range(1,2.2, 0.5) -> 1.0, 1.5, 2.0
step can be negative, but it is an error for it to be zero.
minor fixes to script configurations and PGPLOT_install,
tweaks of Ifeffit.pm.
2001-03-10 Matt Newville <email@example.com>
* Version 1.0052: misc fixes (working on tutorial)
improved behavior of 'print' so that it doesn't
chokes on unparenthesized lists. improved behavior
of 'show' so that output is better aligned.
fixed bug preventing k-space feffit() from working
fixed bug in fft due to unitialized array data,
(seemed to only show up when mixing fftf and fftr),
set default limit of output for q-space data to 30Ang^-1.
added support for scalars to write_data(), so that
scalar names and values are written to header lines.
worked on tutorial to point of calling it a completed
2001-02-27 Matt Newville <firstname.lastname@example.org>
* Version 1.0051: fixed minor bugs:
1. save files got broken in 1.0050: fixed.
2. one column data files can now be read.
3. added log10() function.
4. improved initialization of pre-edge/normalization routine.
2001-02-26 Matt Newville <email@example.com>
* Version 1.0050: fixed several bugs in implementation and
manipulation of single flat array structure for program data,
after a good bit (but not exhaustive) testing.
Arrays can now have up to 16384 elements in length, and smaller
arrays only use up as much room as they need (well maybe with a
little padding). Erasure of arrays and re-defining arrays (the
tricky parts, for sure) seem to working fine, after a few days of
Slight change in behavior for 'erased arrays': if an array is
defined to depend soley on another array, and that dependency
is erased, the dependent array will now be erased as well (since
it's size is now 1: all 1-element arrays are now erased).
a.x = indarr(1000)
a.y = sin(a.x/20)
will also cause a.y to be erased. Note that
a.z = sin((a.x + b.x)/30)
will not be erased -- it still depends on the living b.x.
add &sync_level to control how often sync() is called behind
the scenes == still experimental and not fully implemented, but
with the larger array sizes, repeated definitions of arrays
can be notably slow.
allow 'load save_file' to work as expected (that is to restore
the file) -- this allows save_files to be specified on the
command line list of files to load at start up:
~> ifeffit data.sav process.iff
2001-02-22 Matt Newville <firstname.lastname@example.org>
* Version 1.0049: non-release version, converted to a single
flat array / linked-list structure for program array data.
This is partially tested, but seems to work well, and allows
arrays up to 16384 elements. Not well-tested for complex
dependencies (with lots of sync() calls) or numerous arrays,
for a full buffer (2**20 doubles in all of arrays).
added src/test directory with test programs in Fortran and C
src/test/ifftest1 (Fortran) and src/test/ifftest2 (C)
2001-02-20 Matt Newville <email@example.com>
* Version 1.0048: altered read_data to use a temporary
buffer and then set program arrays from this. This allows
the number of arrays read in to be different than the number
of arrays on a single line, which is useful for large data
sets or much of the data in the Lytle database.
will force two arrays to be filled, even if each line contains
x y x y x y x y
x y x y x y x y
2001-02-20 Matt Newville <firstname.lastname@example.org>
* Version 1.0047: fixed several minor problems,
especially determination of path list of feffit/ff2chi
and feffit()'s auto-execution of chi_noise()
added no_header keyword to write_data() to suppress all
2001-02-19 Matt Newville <email@example.com>
* Version 1.0046: moving steadily towards internal
rearrangement of array data to allow arbitrary length
arrays, moved nearly all array access in iff_* commands
to function calls, changing many of the routines in
re-implemented use of CVS (and now checked-in properly!)
partial (incomplete) rearrangement of examples/
several minor code tweaks.
2001-02-12 Matt Newville <firstname.lastname@example.org>
* Version 1.0045: fixed bug in call of chi_noise
fixed bug in isnum() incorrectly evaluating some
expressions as numerical constants.
fixed bug in minimize() (genfun.f) recently introduced
that killed minimization of data that did not have
estimated uncertainties included
2001-01-29 Matt Newville <email@example.com>
* Version 1.0044: some progress on improving
array layout and implementation
tested (and several buglets fixed) portability of
build to Solaris and Alpha (with f2c), neither
using PGPLOT yet. Also worked on improving the
PGPLOT_install script as well.
2001-01-27 Matt Newville <firstname.lastname@example.org>
* Version 1.0043: several minor changes to installation,
configuration and scripting front-ends. Most notably,
PGPLOT_install greatly expanded.
changed ifeffit() startup behavior to be able to
completely suppress screen echo's, even of copyright
notice, by giving initial call as ifeffit('&quiet').
2000-12-12 Matt Newville <email@example.com>
* Version 1.0042: fixed bug in iff_fft code of
unitialiazed arrays that could, under certain
circumstances, give bad results.
add '&pause_ignore' system variable for whether to
ignore pause statements -- the '-x' command-line
switch in the ifeffit shell now uses this.
improved support and compatiblity with stand-alone
autobk code (autobk version 2.92).
2000-12-08 Matt Newville <firstname.lastname@example.org>
* Version 1.0041: put environmental variables for history
file and size properly into shell command file.
Improved several base-installation instruction files.
Expanded and organized macros and examples in examples/
fixed serious bug in arrays.h (tmparr dimensioned wrong).
2000-12-07 Matt Newville <email@example.com>
* Version 1.0040: fixed bug in spline() preventing reliable
user setting of e0
- fixed bug in autobk's call to pre_edge
- made tmparr() a global variable in arrays.h
- added check for ncurses,termlib, etc in configuration
2000-10-27 Matt Newville <firstname.lastname@example.org>
* Version 1.0039: reworked startup and share directory
so that the core interpreter loads
itself on startup (if they exist). This improves
flexibility of configuration and consistency between
different uses of ifeffit.
- added 'width' keyword to f1f2() command, and added
convolve() function to math engine for convolution with
- made python module class based so that it can initiate
the core interpreter, thereby avoiding the bug of calling
put_arr() or put_scalar() before ifeffit() has been
- put earlier checks on the use of trivial constants
in set/guess/define and iff_eval, to avoid filling
the math engine's constant buffer
- added splint() function to math engine for natural
cubic spline interpolation. Should probably adjust
fefinp and chipth to use cubic spline interpolation of
feff data, but not implemented yet.
2000-10-22 Matt Newville <email@example.com>
* Version 1.0038: added cromer-libermann calculations of f', f''
-command f1f2() calculates f'(e) and f''(e) given an atomic
number and an array of energies. The calculation uses small
data files of the Cromer-Libermann parameters (not interpolation
tables). The data files (in src/cldata) are installed to
and are read from the installed disk location by f1f2() for
use in the calculation. The CL method is borrowed from
Cowan/Brennan's version, but altered significantly.
2000-10-15 Matt Newville <firstname.lastname@example.org>
* Version 1.0037: macro execution for each fitting loop
-feffit and minimize now supports a 'macro' argument
which names a user-defined macro to execute at the end of each
fitting iteration. This allows the progress of the fit to be
monitored by printing the current values of the variables or
by plotting some part of data/fit.
2000-10-12 Matt Newville <email@example.com>
* Version 1.0036: improved logging capabilities, correl()
-log() command now opens/closes a log file to which
all echos can be sent:
log(file = log_file) to open log file
.... all screen echos go to file
log(close) to close log file
screen_echo = 0,1,2,3, 'tee', 'both', 'file'
to set the value of &screen_echo:
0 save to echo buffer (for get_echo() from scripts)
1 normal write to screen
2 write to log file, not screen
3 write to log file and screen
-correl() command turns elements of the fit correlation
matrix into scalars:
where the first 2 arguments are the names of fitting
variables and the third argument is the name of the scalar
to create holding this value (with default name correl_x_y)
-added 'norm_order' keyword to pre_edge() to set order of
polynomial for normalization constant.
2000-10-10 Matt Newville <firstname.lastname@example.org>
* Version 1.0035: several minor fixes:
-gifeffit -- several tweaks, mostly about failing
to properly load and execute command files with blank lines.
-fixed BuildBuild to allow extra ":" on Version line (as above!)
-fixed iff_show to show variable type for unfound variables
-improved some warning messages
-replaced repeated calls to lintrp() in chipth() with
explicit call to hunt() and linear interpolation
-first working implementation of 'do_bkg=true' in command
feffit() to fit 'additional background'.
2000-10-03 Matt Newville <email@example.com>
* Version 1.0034: fixed debye() and eins() to work correctly
with both scalar and array contexts -- both use the concept
of 'current path' from program variable 'path_index'.
- fixed 'show @paths' and 'show @path=N' to always (??)
show the real path information
2000-10-03 Matt Newville <firstname.lastname@example.org>
* Version 1.0033: added erf(), erfc() and erfcx(),
loggamma() special functions to math parser, and
revised implementation of gamma() to be more
consistent with other intrinsic functions.
- improved exit behavior: retracted message from
'quit', made iff_load a function returning 1 immediately
upon seeing 'quit' or 'exit' in a script, and improved
command-line program to honor such exits in scripts
either loaded from the ifeffit prompt or named as a
command-line argument. That is, a 'quit' or an 'exit'
in a script will always truly exit the library and the
command-line program, and return 1 to a calling program.
- encapsulated most [rdffdt not done yet] read() calls
for reading from files calls with a call to function
iread() that returns line length and includes a call to
sclean to clean line-ending problems (dos/unix/mac) and
other C v. fortran read problems.
2000-09-16 Matt Newville <email@example.com>
* Version 1.0032: added get_path() command for a
simple conversion of path data to Program Variables.
Not yet complete.
2000-09-16 Matt Newville <firstname.lastname@example.org>
* Version 1.0031: added gamma function to math expressions.
also delayed putting up plot window until the first plot()
command. altered 'quit' to print exit message. Other:
- updated python installation to auto-install Pmw if needed
- updated command-line ifeffit to accept '!ls' as a valid
shell escape, and to invoke a bash shell on a bare '!'.
2000-09-11 Matt Newville <email@example.com>
* Version 1.0030: added initialization of nerstp to
iff_minimize and added maximum loop-count to fiterr to
avoid infinite-loop bug in minimize().
2000-09-10 Matt Newville <firstname.lastname@example.org>
* Version 1.0029: changed iff_plot to use pgopen/pgslct
and no longer pgbeg/pgend. 'new' plots are now handled very
differently. This has several consequences -- most notably
faster plotting. The default-interactive-device (read from
PGPLOT_DEV) is assumed to be the normal plotting device, and
any other device makes a single hardcopy (gif, ps, ...) of
the current plot with a single 'plot' command. The hardcopy
file is closed immediately, and the default plot is
There is no longer a need to re-select the '/XSERVE' device.
This also means that plotting works correctly on Win32 systems
using the GrWin graphics libraray, at least with Cygwin
(see README.CYGWIN for details). Testing of making an
executable with DVF and linking as a Win32 library is still
2000-09-07 Matt Newville <email@example.com>
* Version 1.0028: several minor fixes
iff_show // macro show fixed small core-dump bug
added 'show @macro=full' to list full macros
added commmand log() which is show() with one
pre-argument (optional keyword = 'file') of a file
name to write results to:
log(file= Macros.iff, @macros=full)
will write all macros to Macros.iff
changed fiterr/gaussj call to try to stave off
a bug pre-report.
2000-09-05 Matt Newville <firstname.lastname@example.org>
* Version 1.0027: altered subroutine preedg() to pass back
coefficients of post-edge normalization curve. the pre_edge(),
now saves thses to 'norm_c0, norm_c1, norm_c2', so that
the curve can be reconstructed as:
data.norm_curve = norm_c0 + norm_c1 * data.energy
+ norm_c2 * data.energy^2
autobk: (version 292a now) now writes these coefficients
out to autobk.log.
ifeffit (version 1.0a): several improvements to this
command-line program (This is the C/ReadLine version that
better support for tab-completion
command history (including between sessions)
startup file ($HOME/.ifeffit: ifeffit commands loaded
before anything else)
command line arguments:
files listed are 'load'ed as command files
-x option means 'execute and exit'
help more complete (though still not complete enough!)
2000-09-02 Matt Newville <email@example.com>
* Version 1.0026: fixed bug in iff_save -- crashed on
restore of a non-existent file.
2000-08-30 Matt Newville <firstname.lastname@example.org>
* Version 1.0025: fixed bug in iff_rddata -- data files with
more that 2048 points were causing severe errors. No truncate
data -- should work on array() as a pointer-driven array.
2000-08-24 Matt Newville <email@example.com>
* Version 1.0024: fixed bug in fefinp that made defining
a path with a non-existent feff.dat file cause _no_ feff
files from ever being read in.
2000-08-23 Matt Newville <firstname.lastname@example.org>
* Version 1.0023: add sorting of energy/xmu arrays for spline,
with routine 'sort_xy'. sort_xy is called in routine spline(),
and having repeated points causes a warning -- but the fit of
the spline works (more or less: still an issue of how
iff_spline should deal with the resulting bkg array).
Also: added iff_test and command '__test' for general testing
2000-08-21 Matt Newville <email@example.com>
* Version 1.0022: relaxed path parameter keywords in feffit
2000-08-21 Matt Newville <firstname.lastname@example.org>
* Version 1.0021: fixed errors (serious!) in sum_paths.f
and fitfun.f for getting Real(chi) and Imag(chi) for fit.
Also turned back on 'default params' for fft, ff2chi, and
2000-08-19 Matt Newville <email@example.com>
* Version 1.0020: turned off all 'default params' from
program variables in spline, pre_edge, fft, and feffit
commands -- trying out __requiring__ all arguments to be
2000-08-17 Matt Newville <firstname.lastname@example.org>
* Version 1.0019: added 'norm' output to pre-edge
2000-08-15 Matt Newville <email@example.com>
* Version 1.0018: added 'find_e0' to spline
2000-06-12 Matt Newville <firstname.lastname@example.org>
* Version 1.0017: altered iff_color to fix background
2000-06-04 Matt Newville <email@example.com>
* Version 1.0016: altered misc_arr.f again (prenam/fixnam)
2000-06-01 Matt Newville <firstname.lastname@example.org>
* Version 1.0015: added 'show' to zoom, fixed 'show' for cursor
2000-05-27 Matt Newville <email@example.com>
* Version 1.0014: fixed return value for 'end macro' to be 0
in more situations.
2000-05-21 Matt Newville <firstname.lastname@example.org>
* Version 1.0013: fixed bug in 'fixnam' that was allowing
arrays named '1.x' to be created. Now munged to '_1.x'
changed python install procedure to use 'compile.py' and
2000-05-12 Matt Newville <email@example.com>
* Version 1.0012: add 'linestyle' command to set plot
line style -- analogous to 'color' command.
rewrote C wrapper: ifeffitc.c and ifeffit.h are now
more complete, and have more C-like interface routines
for iff_exec() [ie, ifeffit()], get_scalar(), etc.
All C/Fortran calling is now confined to ifeffitc.c,
allowing all wrapper modules to use the C interfaces
without dealing with C/Fortran issues. This was mostly
done to ease the Windows port, but may also simplify
porting in general.
2000-05-11 Matt Newville <firstname.lastname@example.org>
* Version 1.0011: updated several things in common blocks
to avoid warnings, changed lm_err() in minpack. Also played
with pgplot calls -- having troubles with GrWin on WinNT --
some pgplot functionality of switching and closing devices
may be broken.
2000-05-05 Matt Newville <email@example.com>
* Version 1.0010: changed array indexing so that the 'magic
zero array' (ie, array that can be relied on to be full of
zeros) is at maxarr, not zero. Changed indexing of arrays
from 0:maxarr to 1:maxarr, and several of the routines that
2000-04-24 Matt Newville <firstname.lastname@example.org>
* Version 1.0009: add '@group' and '@path's to 'erase'
in iff_erase.f. Complete rewrite of iff_erase.f, not
very well tested, but 'erase @group=x' does work (which
is what I needed today!).
2000-04-24 Matt Newville <email@example.com>
* Version 1.0008: add 'notitles' option to iff_rddata,
and made output $column_label string contain only the
2000-04-14 Matt Newville <firstname.lastname@example.org>
* Version 1.0007: rewrote iff_print to use buffer
string before executing echo/chrdmp. Now works
from scripting languages when '&screen_echo = 0'.
added sign() function to math parser, which returns
-1,0,1 based on sign of input.
2000-04-11 Matt Newville <email@example.com>
* Version 1.0006: fixed bug in iff_show, verified build
with f2c. Working of configure/autoconf
2000-04-04 Matt Newville <firstname.lastname@example.org>
* Version 1.0005: fixed 'trivial constant' parsing in
iff_set, removed one more reference to 'info buffer'.
python module reworked, now with shell emulator
2000-04-03 Matt Newville <email@example.com>
* Version 1.0004: removed all references to 'info buffer'
recovering from vacation, laptop theft.
2000-03-11 Matt Newville <firstname.lastname@example.org>
* Version 1.0003: added construction of delta_VARS to
iff_minimize, and tweaked (of all things!) gaussj.f --
minpack/fiterr now completely ported to C.
2000-03-07 Matt Newville <email@example.com>
* Version 1.0002: now auto-building build.h, with version read
from this file, and Keyword/Hints read from Keywords.dat.
Save files upgraded to version 1.01, now recording IFeffit build
info in save files.
2000-03-06 Matt Newville <firstname.lastname@example.org>
* Version 1.0001: Begin using ChangeLog for version numbers
and update notices, working towards building source configuration
options into 'build.h'.
From OLD ChangeLog::
2001-05-31 Matt Newville <email@example.com>
* Version 1.0c: upgraded to Readline 4.2
2000-09-07 Matt Newville <firstname.lastname@example.org>
* Version 1.0b: just plain 1 beta -- the real changelog
is in src/lib/ChangeLog
2000-06-12 Matt Newville <email@example.com>
* Version 1.0.2: updated autobk setup and build to be
fully compliant with stand-alone autobk (version 2.92)
2000-06-02 Matt Newville <firstname.lastname@example.org>
* Version 1.0.1: second release of full distribution
- library much improved
- minor improvements to wrappers/python and wrappers/perl
- added examples/ directory with example scripts
2000-04-18 Matt Newville <email@example.com>
* Version 1.0.0: initial release of full distribution