
			  Algae Release Notes


Copyright (C) 1994-2004  K. Scott Hunziker.
Copyright (C) 1990-1994  The Boeing Company.

See the file COPYING for license, warranty, and permission details.

Version 4.3.6 (February 22, 2004)

  * Fixed a label handling error in the `set' function that was
    introduced in version 4.3.1.  Also cleaned up and documented
    label handling in `complement', `intersection', and `union'.

Version 4.3.5 (February 11, 2004)

  * Added support for short int endian conversions in `fread'.  This
    was a quick fix -- we really need to go through that code (and
    `fwrite') to clean it up and support a variety of missing cases.

Version 4.3.4 (December 16, 2003)

  * Fixed disassemble bug causing wipeout on `veil' statement.

  * Now use --enable-debug instead of --with-debug in configure.
    (It was and has been broken since version 4.3.0.)

Version 4.3.3 (December 9, 2003)

  * By default, use Doug Lea's dlmalloc package in place of the
    C library's malloc.  On some systems, dlmalloc is far superior.
    For example, Algae's timing suite runs 2.5 times faster with
    dlmalloc than with the AIX 5.1 malloc routines.  (On Linux,
    dlmalloc is the default anyway.)

  * Replaced code for `magic' with faster code from Octave.

  * Supplied some missing prototypes that caused trouble.

  * Quit installing header files.  Not sure how or why that got
    started, but if you have installed previous versions of Algae,
    you may have junk left around in /usr/src/algae.

Version 4.3.2 (November 29, 2003)

  * Ouch!  Fixed a nasty bug in `filter'.

Version 4.3.1 (November 24, 2003)

  * Several efficiency improvements, some pretty dramatic.

  * Short ints supported in `fread' and `fwrite'.

  * New functions `erf', `erfc', `gcd', `lcm', `primef', `primes', and
    `trapz'.

Version 4.3.0 (October 4, 2003)

  * Enable linking to fftw (http://www.fftw.org/) for high-performance
    Fourier transforms.

  * The `fft' and `ifft' functions now handle matrices.  By default
    they transform each column, but that can be changed with an option.

  * The `max' and `min' (and `imax' and `imin') functions now handle
    matrix and table arguments.  Note:  The old versions converted
    matrix arguments to vectors, if possible; the new versions may
    give different results in those cases.

  * The new `circshift' function performs circular shifts of arrays.

  * Minor code changes to support Mac OS X.

  * Fixed several potential buffer overrun problems.

  * Included latest (3 June 2003) SuperLU updates.

  * Included current ARPACK patches.

Version 4.2.1 (August 6, 2003)

  * Added the `$digits' global variable.

Version 4.2.0 (August 5, 2003)

  * New `veil' statement allows temporary changes to global variables.

Version 4.1.4 (August 2, 2003)

  * Improved configuration for floating point exception handling.

  * Added numerous "volatile" qualifiers, to protect against longjmp
    problems with optimized gcc.

  * Fix `select' function for zero-length vectors.

Version 4.1.3 (May 16, 2003)

  * Configure checks for "malloc.h".

  * Fixed configure to work with autoconf-2.57.

Version 4.1.2 (March 19, 2003)

  * Minor installation improvements.

Version 4.1.1 (February 9, 2003)

  * Fixed compile-time problems involving inverse hyperbolic trig.

Version 4.1.0 (January 7, 2003)

  * Changed the installation directories to be in line with the FHS.

  * Added `mult' function for efficient matrix products.

Version 4.0.3 (December 5, 2002)

  * Minor installation improvements, including RPM fix for RedHat 8.0.

Version 4.0.2 (September 30, 2002)

  * Fixed FORTRAN mangling for NPSOL.

Version 4.0.1 (September 21, 2002)

  * Restored optional linking to NPSOL.

Version 4.0.0 (September 15, 2002)

  * Added reserved keywords "try" and "catch".

  * New try/catch mechanism for control of exception handling.

Version 3.7.3 (July 31, 2002)

  * The `iram' (ARPACK) function is now functional and documented.

  * Fixed memory leaks in `solve' and `svd'.

  * Fixed a few missing headers that were causing compile problems on
    some machines.

Version 3.7.2 (May 12, 2002)

  * Builtin `tril' and `triu' functions -- sparse arrays stay sparse.

  * Latest ARPACK patches applied.

Version 3.7.1 (May 6, 2002)

  * Fixed integer overflow problems in `band' and `cram'.

  * Good start at ARPACK, but not yet documented.

Version 3.7.0 (April 24, 2002)

  * Incorporated the SuperLU package for solving sparse linear
    systems.  Previously, sparse systems were converted to dense and
    solved with LAPACK (unless configured with BCSLIB-EXT).

  * Included sparse capability for `equilibrate' from SuperLU.

  * Included support for memory allocation debugging with dmalloc.

  * Fixed label handling errors in `solve.c'.

Version 3.6.2 (November 1, 2001)

  * Minor installation improvements.

Version 3.6.1 (October 26, 2001)

  * Improved the logical operations to do a much better job at
    maintaining sparsity.

  * New standard function `select'.

  * Various other minor improvements.

Version 3.6.0 (October 11, 2001)

  * Changes to the matrix bandwidth minimization code.  The old `band'
    function is now called `gpskca' and can do either profile or
    bandwidth minimization of symmetric or hermitian matrices.  Matrix
    statistics are computed by the new `band' function.

  * New standard functions `linspace' and `logspace'.

  * New standard function `equilibrate' computes scale factors to
    improve the condition of a matrix.  However, sparse code has not
    yet been included.

  * Fixed several deficiencies in the `transform' function.

  * Removed old get_ascii and put_ascii functions.  I doubt if anyone
    has used these in the last decade.  To read and write matrices in
    text files, use the algae MatrixMarket package.

Version 3.5.2 (September 19, 2001)

  * Fixed several errors with hermitian symmetry.

  * Another shot at AIX configuration.

Version 3.5.1 (September 14, 2001)

  * Configuration accommodations for AIX.

  * Fixed mishandling of sparse matrix multiplication case where
    non-zero factors give a zero result.

Version 3.5.0 (August 31, 2001)

  * Added "short-circuit" logical operators && and ||.

  * Added the `cram' builtin function -- it's the same as `form'
    except the output array is sparse when that makes sense.

  * Added the `product' user function -- product of array elements.

  * Added `mksparse' and `exsparse' functions for converting matrices
    to and from coordinate form.

  * Switched to LAPACK's DGGEV now that DGEGV is deprecated.  Also use
    the new workspace size query feature in LAPACK, rather than
    working it out ourselves.

  * Use LAPACK's ZHEGV and ZGGEV for complex generalized eigenvalue
    problems. 

  * Use floating point exception support from glibc 2.2 on Linux.

  * Improved accuracy and behavior for complex arguments in functions
    `log', `asin', and `acos'.

  * Added inverse hyperbolic trig functions `asinh', `acosh', and `atanh'.

  * Various minor bug fixes.

Version 3.4.0 (May 1, 1998)

  * New assignment operators:  `*=', `/=', `@=', and `%='.

Version 3.3.6 (April 1, 1998)

  * Various configuration improvements.

  * New `bliss' function ends want and effects world peace.

Version 3.3.5 (March 3, 1998)

  * Modified configure for hpux 10.  (Eventually, users are going to
    tire of the gratuitous changes we see with every new version of hpux.)

  * Improved `umin' function; now it can pass parameters to the
    objective function.

Version 3.3.4 (October 21, 1997)

  * Fixed a bug introduced in 3.3.3, involving `provide' and `require'.

Version 3.3.3 (October 18, 1997)

  * Standard user functions are now pre-compiled.  This makes startup
    much faster.

  * Minor installation fixes.

Version 3.3.2 (October 5, 1997)

  * New `-x' command line option causes algae not to read stdin by
    default when no file names are given on the command line.

Version 3.3.1 (September 25, 1997)

  * New `-R' command line option causes restricted execution -- no
    process creation is allowed.  This disables the `system' function
    as well as pipes.  Use this mode when executing untrusted Algae
    code.

  * New `-n' command line option changes the way that scalars are
    displayed in printing statements.

Version 3.3.0 (September 2, 1997)

  * Better support for "site" files.  The `src' and `require'
    functions now look in ~/algae and /usr/local/lib/algae/site/A
    (by default), unless overridden by the ALGAE_SRC_PATH environment
    variable.

  * Includes "spec" file for Red Hat Package Manager (RPM).

  * Added the `sign' and `pick' functions.

Version 3.2.9 (August 4, 1997)

  * Improved `info' capability.

Version 3.2.8 (July 28, 1997)

  * Improved EOF handling in `fread'.

  * Fixed error in printing zero-length vectors.

Version 3.2.7 (July 24, 1997)

  * Plugged memory leak in `eig'.

  * Labels on character arrays were being dropped when the arrays were
    added with scalars.

  * Work around texi2html bug dealing with function references in HTML
    documentation.

Version 3.2.6 (June 25, 1997)

  * Faster `get' and `put'.

  * New `char' function.

Version 3.2.5 (May 13, 1997)

  * Fixed serious bug in "improved" sorting.

Version 3.2.4 (May 9, 1997)

  * Improved sorting.

Version 3.2.3 (May 2, 1997)

  * Immediate, clean exit on SIGQUIT.

Version 3.2.2 (April 25, 1997)

  * Fixed installation problems involving f2c.

Version 3.2.1 (March 28, 1997)

  * Float capability in `fread' and `fwrite'.

  * Various bug fixes.

Version 3.2.0 (February 21, 1997)

  * Name changed to "algae"!

Version 3.1.7 (January 18, 1997)

  * `atof' now handles arrays.

  * Minor bug fixes.

Version 3.1.6 (December 12, 1996)

  * Fixed terminal width initialization bug.

Version 3.1.5 (November 2, 1996)

  * Added `arg' function.

Version 3.1.4 (October 22, 1996)

  * Added new `tolower' and `toupper' string functions.

Version 3.1.3 (October 19, 1996)

  * Supply our own BLAS and LAPACK code.

Version 3.1.2 (October 15, 1996)

  * File "examples/numbers.A" now has functions for both cardinal and
    ordinal numbers.

Version 3.1.1 (October 3, 1996)

  * Merge changes to stable version through 3.0.9.

Version 3.0.9 (October 1, 1996)

  * Bug in -e command line option prevented multiple occurrences.

Version 3.0.8 (September 30, 1996)

  * Documentation and installation improvements.

Version 3.0.7 (September 13, 1996)

  * Removed some minor misfeatures.

Version 3.1.0 (August 29, 1996)

  * Added the `leastsq' function.

Version 3.0.6 (August 29, 1996)

  * Link to libUfor on DEC Alpha.

  * Fixed minor bugs in `seq', `eig', and the code that converts
    matrices into vectors.

Version 3.0.5 (June 27, 1996)

  * Keep a history file in "~/.alki_history".  The name may be
    overridden by the ALKI_HISTFILE environment variable.

  * Fixed profiler bug in counting line number hits.

Version 3.0.4 (June 6, 1996)

  * Work around linux problems with dynamically linked Fortran.

  * Various other minor improvements.

Version 3.0.3 (May 28, 1996)

  * User functions now report global variable references.

  * Bug fixes.

Version 3.0.2 (May 21, 1996)

  * Configuration support for Cray TS.

Version 3.0.1 (May 15, 1996)
Version 3.0.0 (May 9, 1996)

  * Syntax change!  The product operator was "#" and is now "@".
    Comments now begin with a single "#".  This change is not backward
    compatible, and it is possible for old files to run but give
    incorrect results.  Please use the "alki2to3" utility (included)
    to convert your old files to the new syntax.

  * Changed the scaling for `fft' and `ifft'.  Previously, both scaled
    by `1/sqrt(n)'.  Although the symmetry of this is appealing, it's
    also unconventional and thus confusing.  The new approach is to
    scale only `ifft', and that by `1/n'.  (There's also a slight
    increase in efficiency.)

  * Significantly improved the efficiency of "randn".

  * Some "improvements" in autoconf-2.9 required work-arounds.

  * More than the usual number of bug fixes and improvements.

Version 2.5.4 (February 15, 1996)

  * Deal with moving targets like linux (new fpe defaults), hpux (it
    gets worse with every version), and cray (T90 machines).

Version 2.5.3 (January 28, 1996)
Version 2.5.2 (December 19, 1995)

  * Support for Linux returns!

  * Bug fixes and installation improvements.

Version 2.5.1 (December 13, 1995)

  * Finally, support for the IBM RS6000.  (I finally found a machine
    that wasn't a total mess.)

  * Installation improvements.

Version 2.5.0 (December 10, 1995)

  * Added new `exec' function.  Besides the additional capability, it
    finally allows an efficient implementation of `eval'.

  * Added the '-e' command line option for providing scripts.

  * Various installation improvements.

Version 2.4.2 (November 20, 1995)

  * Added new `substr' builtin function.

  * Fixed sloppy signal handling.  Profiling is the most obvious
    beneficiary.

  * Fixed memory leaks in scanner.

Version 2.4.1 (November 5, 1995)

  * Improvements for "binary" installation.

Version 2.4.0 (October 14, 1995)

  * Added `fread' and `fwrite' functions for binary i/o.

  * Removed the old `ode' Runge-Kutta integrator.  Use `ode4' instead.

Version 2.3.4 (September 8, 1995)

  * Fixed minor problems introduced in 2.3.3.

Version 2.3.3 (August 24, 1995)

  * Work around a (backwards incompatible) change in gnuplot scanf.

Version 2.3.2 (August 23, 1995)

  * Had trouble reading large binary files in old format.

  * Enable profiler by default.

Version 2.3.1 (August 15, 1995)

  * Fixed obscure bug with `self'.

  * Fixed bug in `atan2' -- array dimensions weren't checked.

Version 2.3.0 (August 15, 1995)

  * New `filter' function.

Version 2.2.0 (August 2, 1995)

  * A new binary file format is used by the `get' and `put' functions.
    This new format enables a significant improvement -- reference
    counts are now preserved within the file so that needless
    duplication of data is avoided.

  * Fixed bug (noticed only on Solaris 2.x) caused by neglecting to
    clear i/o errors on interrupted calls.

Version 2.1.3 (July 29, 1995)
Version 2.1.2 (July 26, 1995)

Version 2.1.1 (July 21, 1995)

  * Fixed installation problems.

Version 2.1.0 (July 20, 1995)

  * Finally completed implementation of `self'.  The `self' keyword is
    an expression that returns the function in which it resides.

  * The Alki documentation is now also provided in HTML form, and the
    `info' function will use an appropriate browser if possible.

Version 2.0.5 (June 15, 1995)

  * Installation improvements.

Version 2.0.1 (May 30, 1995)

  * New `tril' and `triu' functions.

  * Better installation.

Version 2.0.0 (May 29, 1995)

  * Name changed from AMPL (A Matrix Programming Language) to Alki.
    I sure hated to change the name, but there's another AMPL out
    there and it's been distributed widely.

  * New standard functions include `eval' and `string'.

  * Floating point exceptions handled properly on the Cray.

  * Back-substitution no longer botches systems that have a real array
    and a complex right-hand side.

  * Configuration now done with GNU autoconf.

  * Other fixes too numerous to mention.

  * Now licensed under the GNU General Public License.

Version 1.9.1 (January 17, 1995)

  * Fixed memory leaks in element assignments to sparse arrays.

  * Don't pass SIGINT to children.

Version 1.9.0 (December 9, 1994)

  * New home (buxom) for AMPL!

  * Support for UnixWare.

Version 1.8.1 (November 10, 1994)

  * Now use getopt to parse command line options.

  * Profiler now works on Solaris 2.3.

  * Support for Cray Y-MP.  (No signals yet, though.)

  * The usual tweaks.

Version 1.8.0 (October 24, 1994)

  * The factor and solve code is finally presentable.  Condition
    numbers are always checked to identify ill-conditioned systems.
    Sparse problems use the BCSLIB-EXT routines.

  * Major improvements (like reliability, accuracy, and consistency)
    to the complex math functions like square root and exponentiation.

  * Catch errors (like "device full") in output routines.

Version 1.7.2 (October 7, 1994)

  * Fixed bugs in new table overloading code.

Version 1.7.1 (September 13, 1994)

  * Removed a few unintended global variables in standard functions.

Version 1.7.0 (August 28, 1994)

  * Most unary and binary operators are now overloaded to work on
    tables.  For example, `t+1' adds 1 to each member of the table `t'.

  * The new `builtin' function links to user's shared objects.  Now
    you can provide your own C code, for example, and link to it at
    run time.  Reorganization of header files and lots of documentation 
    are still needed to make this fully functional.

Version 1.6.2 (August 13, 1994)

  * Fixed bug in sparse vector assignments introduced in version 1.5.0.

Version 1.6.1 (July 22, 1994)

  * Reduced some waste in error messages.

  * The usual tweaks.

Version 1.6.0 (July 7, 1994)

  * New functions for maintaining source files, including `src',
    `autosrc', `require', and `provide'.

  * Hacked up a script called "atags" to make an Emacs TAGS file for
    AMPL source files.

Version 1.5.1 (July 5, 1994)

  * New `getenv' function to get environment variables.

  * Fixed bug with empty matrices in new assignment code.

Version 1.5.0 (July 4, 1994)

  * Major improvements in element assignment performance.  With sparse
    operands, it may be 100 times faster or more.  When element
    assignment is to a member (like "a.b[c]=d"), there is still much
    room for improvment.

  * New functionality for the `split' function -- delimiters may be
    specified.

  * Fixed bug in addition of zero-length character strings.

  * Fixed dusty old bugs in `diag' involving character arrays.

  * Extra members of arrays were sometimes lost in element assignments.

Version 1.4.6 (May 24, 1994)

  * Fixed inconsistent label behavior with zero row or column matrix
    partitions.

  * Other minor fiddling.

Version 1.4.5 (May 4, 1994)

  * Improved label handling for matrix appends.

  * Fixed MATLAB I/O, I hope.

Version 1.4.4 (May 3, 1994)

  * Fixed bug in `diag' involving sparse matrices.

  * An "escaped newline" may be followed by a comment.

Version 1.4.3 (April 17, 1994)

  * Support for Solaris 2.3.

Version 1.4.2 (April 9, 1994)

  * Improved logical operations on sparse vectors.  For example, `v&v'
    is sparse if `v' is sparse.  Matrices still need this improvement.

Version 1.4.1 (April 1, 1994)

  * Minor installation fixes.

Version 1.4.0 (March 25, 1994)

  * AMPL can now be linked to the BCSLIB-EXT library for solving
    sparse systems of equations.  (Sparse eigenvalue problems will
    follow.)

  * Finally, a man page.

  * The code that decides whether to convert from sparse to dense
    storage had the potential for integer overflow.

  * The usual tweaks.

Version 1.3.12 (March 15, 1994)

  * Improvements (long overdue) in the logical functions.

  * Fixed bug in which a diagonal matrix times a dense symmetric
    matrix gave a symmetric (rather than general) result.

Version 1.3.11 (March 11, 1994)

  * Fixed bugs in `apply' dealing with sparse symmetric and sparse
    hermitian matrices.  This affected functions ranging from `abs' to
    `tanh'.  (I take it that these matrices aren't used much.)

Version 1.3.10 (March 10, 1994)

  * Fixed several symmetric/hermitian mixups for complex matrices.

Version 1.3.9 (March 3, 1994)

  * Fixed bug involving append operation with zero-length vectors.

  * Removed the old Laguerre code.

Version 1.3.8 (February 24, 1994)

  * New `cd' function to change directory.

  * Accommodate tkinfo alternative to standalone info.

Version 1.3.7 (February 17, 1994)

  * Full support for DEC alpha.

Version 1.3.6 (February 16, 1994)

  * New `umin' function for unconstrained minimization.

  * The usual tweaks.

Version 1.3.5 (January 16, 1994)

  * Recent flex update required simple changes to scanner.

  * Fixed label bug in `complement'.

Version 1.3.4 (January 11, 1994)

  * Added `randn' function for normally distributed random numbers.

  * Don't convert integers and reals to complex in `hermitian'.

  * Fixed error handling bug in `fill' and `form'.

Version 1.3.3 (December 6, 1993)

  * Fixed minor bug in inequality with character operands.

Version 1.3.2 (December 6, 1993)

  * Fixed error in Titan installation.

Version 1.3.1 (November 23, 1993)

  * Bug in `plot' and `splot' prevented using percent sign in titles
    or labels.

  * First cut at DEC alpha support.  No fpe traps yet.

  * Added provision to specify Fortran name suffix.

  * Minor improvements to the manual.

  * Fixed bug in installation test for domain error reporting.

  * Improved installation.

Version 1.3.0 (November 5, 1993)

  * Made a significant change to the DYNASTY I/O routines `getdyn' and
    `putdyn'.  They now work with multiple matrices in a file.

Version 1.2.4 (November 2, 1993)

  * Caught the execution profiler in another flub.  Most previous
    profiler results were at least slightly (and sometimes wildly)
    inaccurate.

Version 1.2.3 (October 29, 1993)

  * Fixed some very serious bugs involving sparse, complex matrices.
    Since functions like `diag' return sparse arrays, this should
    concern anyone who uses complex matrices.

  * Added the `label' function for assigning array labels.

Version 1.2.2 (October 22, 1993)

  * Added "scalar" elements in `fem'.  One day soon, I hope to make
    `fem' suitable for general consumption (with documentation and
    everything).

  * Added the `diff' function for differences between array elements.

Version 1.2.1 (October 14, 1993)

  * Added the `sources' function for reading multiple files.

  * Bend over backwards for the poor old Apollo.

Version 1.2.0 (October 12, 1993)

  * Read and write MATLAB binary files with `getmat' and `putmat'.

Version 1.1.4 (September 29, 1993)

  * Fixed faulty logic in execution profiler which sometimes resulted
    in it being turned off at inappropriate times.

  * The transpose operator knows not to bother transposing a symmetric
    matrix, but it mustn't forget to swap the labels.

Version 1.1.3 (September 24, 1993)

  * The `#' operator was dropping labels.

Version 1.1.2 (September 23, 1993)

  * The new `prof' function summarizes output from the execution profiler.

  * Provision is made for disabling the execution profiler when AMPL
    is installed.

  * The `fill' and `form' functions screwed up when they were supposed
    to return scalars.

Version 1.1.1 (September 15, 1993)

  * Fixed bug in member assignments to copies of non-tables.  How
    could this go so long without being noticed?

Version 1.1.0 (September 15, 1993)

  * Execution profiling (-p option) added.

Version 1.0.2 (September 13, 1993)

  * Fixed bug in `get' -- botched reading empty tables.

Version 1.0.1 (September 7, 1993)

  * Straightened out my confusion on left eigenvectors.

  * Enable compiling for old LAPACK 1.0.

Version 1.0.0 (September 6, 1993)

  * The `eig' function can now handle real, nonsymmetric, generalized
    eigenvalue problems.  Also, provision for calculating left
    eigenvectors for nonsymmetric problems has been added.  The option
    to skip computation of the right eigenvectors has been changed
    from "novectors" to "no_right".

  * The `+=' and `-=' operators now work with element references.
    Expressions like `a[b]+=c' are now accepted.  This fixes bug #4.

  * Assignments to the symbol table, like `$$.a=b' or `$$.(a)=b', are
    now accepted.  This fixes bug #12, and opens up a whole new world
    of hacking possibilities.

  * Symbol table references like `$$.a' are now much more efficient.

  * The parser now generates better code for expressions like `(a).b.c'.

  * The `sum' function is new; it sums elements of an array.

  * Fixed some serious density and symmetry bugs in `bdiag'.

Version 0.6.4 (August 31, 1993)

  * New `unlabel' function for removing labels.

Version 0.6.3 (August 25, 1993)

  * The `npsol' function no longer warns you when `inform' is nonzero --
    you'll have to check it yourself.

  * Minor changes and bug fixes in `plot', `splot', and `unplot'.

  * Makefile improvements.

Version 0.6.2 (August 24, 1993)

  * Improvements to `plot' and `splot'.  Tables are no longer
    "flattened"; instead, members of a table within a table are
    given to gnuplot in the same file (which means they get the same
    line and point type).

  * New `seq' function for integer sequences.  Calling `seq(n)' is
    like `1:n' except that it forces integer type and returns a
    zero-length vector if `n' is less than 1.

  * No longer use mktemp(1) in `error_tests' -- not always available.

Version 0.6.1 (August 15, 1993)

  * Restored the "all" option in `unplot'.  (It was in Don's version;
    I don't know why I dropped it.)

  * Fixed memory leak for nonexistent member references.

Version 0.6.0 (August 5, 1993)

  * Surface plots and improved handling of multiple plot processes,
    courtesy of Don Morris.  The plotting functions are better, but
    different.  The `unplot' function is new.

  * Fixed parser bug which allowed vexpr's in "if" conditions but only
    expr's in "elseif" conditions.

  * Added the `reverse' function.

  * Added regression tests for error handling.

  * Improved maintainability of parser code.

Version 0.5.6 (August 2, 1993)

  * The parser occasionally lost track of its "continuing" state after
    a call to `source'.  That's been fixed.

  * No more local symbols in $$.

Version 0.5.5 (July 26, 1993)

  * Parser was missing OP_LINE's in a few places.

Version 0.5.4 (July 25, 1993)

  * Fixed bugs with handling of sparse character arrays.

  * Fixed bug in subvector and submatrix assignments.  If the
    reference count was more than one, members were lost.

Version 0.5.3 (July 22, 1993)

  * Fixed bug in `split'.

Version 0.5.2 (July 21, 1993)

  * Enhanced the `svd' function, providing options for the singular
    vector computation.

  * Fixed bug in `ident'.

  * Improved makefiles.

Version 0.5.1 (July 20, 1993)

  * Fixed minor makefile problems.

Version 0.5.0 (July 19, 1993)

  * Added the `ident' function for identity matrices.

  * No longer source "./.ampl" on startup.

  * Code put under CVS control.

Version 0.4.3 (July 14, 1993)

  * Added the `grep' function.

Version 0.4.2 (July 12, 1993)

  * The `div' function (in "binop_scalar_integer") is undefined for
    zero denominator -- must test before calling it.

  * Added missing math error checking in "apply" and "exp".

Version 0.4.1 (July 8, 1993)

  * In "read_numbers", use atof instead of strtod, and fail instead of
    warn on range errors.

  * Enabled character labels in `putdyn'.

  * Mike radically improved Scott's ugly parser code, cleaned up the
    grammar, and fixed a bug with `while(1){break}'.

Version 0.4.0 (July 5, 1993)

  * Attempt to trap and handle floating point exceptions.  This has
    been a real eye-opener for me -- I've tried four popular machines,
    and not one of them works as its documentation says it should.

  * Added `load()' and `save()' functions.

  * Added `digits()' for user specification of print precision.

  * Added `atan2()' function.

  * Fixed NULL argument bug in `printf()'.

  * Fixed many bugs involving vectors or matrices with 0 dimension.

Version 0.3.0 (June 23, 1993)

  * Added file and line number coding.  Both parse and run time error
    messages give file name and line number.  The "message" function
    provides the same capability to user functions.  New builtin
    functions are "strip", "file", and "line".

  * Changed the binary format used by the "get" and "put" functions.
    Use the "old_get" function to read old files (and convert them --
    "old_get" won't be there forever).  The new format is designed to
    be compatible between machines.  This is not yet the case for VAX
    computers (and may take quite a bit of work).

  * The "get" and "put" functions can now read and write functions.

  * The "real" and "imag" functions now always return real.

  * Improved rounding.  Use rint(3m) on machines that have it.

  * Added "ceil", "floor", and "round" functions.

  * Changed "max", "min", and "sort", to "imax", "imin", and "isort".
    The new functions "max", "min", and "sort" return values, not
    indices.

  * Function definitions are now expressions.

  * Added the "$program" predefined variable.

  * Imply statement terminator at closing brace of "if", "while", etc.

  * No longer assign statement's value to "$".

  * Symmetric partition gives symmetric result.

  * Added error checking on reading numeric constants.

  * All AMPL strings were malloced FOUR TIMES TOO LARGE.

  * Added `atof', `dice', and `split' string manipulation functions.

Version 0.2.3 (March 12, 1993)

  * Scalar rhs in assignments is now expanded to appropriate size.
    For example, "v[1:5]=0" assigns 0 to the first 5 elements of "v".

Version 0.2.2 (February 10, 1993)

  * Fixed a bug in the parser that sometimes left it in the wrong
    state when a parse error was detected.

  * Finished implementing class conversions for partitioning.

  * Assignments now brought into line with our class conversion
    rules.  For example, "x[;1]=1:x.nr" now works just fine.

Version 0.2.1 (January 15, 1993)

  * Mike Brennan showed me how most compilers miss an obvious switch
    optimization.  Simply by reordering our enums (for class, type,
    etc.), our timing tests now run up to 10% faster.  (This probably
    also shows that we do way too much switching.)

  * Added a new "read" builtin, changing the name of the old one to
    "readnum".  The new one reads a line of text.

  * Empty entries accepted in tables, vectors, and matrices.  The
    statements "{x;}", "v[]", and "[]" are all legal now.

  * The "class" function now works for NULL, too.

Version 0.2.0 (December 19, 1992)

  * Changed syntax for "if", "for", "while", and "function"
    statements, replacing the "end" keywords with braces.

  * Improved detection and elimination of "fat" sparse arrays --
    sparse vectors and matrices that are more efficiently stored
    dense.

  * Significantly improved vector and matrix partitioning.  Scalar
    selectors now reduce the dimensionality of the result.  For
    example, "a[1,1]" is a scalar.  Also, v["1"] now works with
    numeric labels.

  * Fixed bug in the "find" function.  (The statement "find(1,1;1,1)"
    gave wrong answers in previous versions.)

  * Plugged several memory leaks.

  * Changed the default AMPL_RCO to "/usr/local/lib/ampl.A".

Version 0.1.9 (September 9, 1992)

  * Added version info in binary files (written by "put" and read by
    "get").  This will let us change the format and still be able to
    read old files.  Unfortunately, it also means that binary files
    written in previous versions can't be read now -- to fix them,
    simply add the first four bytes of any new binary file to the
    beginning of the old ones.

  * Fixed a bug in "set".  Previously it botched character vectors
    with duplicate elements.

  * Made some small improvements to the DYNASTY I/O routines.  The
    "putdyn" function no longer closes its file; this allows multiple
    matrices in the same file.

Version 0.1.7 (July 10, 1992)

Version 0.1.6 (June 23, 1992)

Version 0.1.5 (June 11, 1992)

Version 0.1.3 (May 26, 1992)

    Previous versions lost in the mists of time...
