changelog

著名的大规模线性规划求解器源码GLPK.C语言版本,可以修剪.内有详细帮助文档.

        lpx_load_mat3          lpx_load_matrix
        lpx_unmark_all         (no more supported)
        lpx_mark_row           (no more supported)
        lpx_mark_col           (no more supported)
        lpx_clear_mat          (no more supported)
        lpx_del_items          lpx_del_rows, lpx_del_cols
        lpx_get_row_bnds       lpx_get_row_type, lpx_get_row_lb,
                               lpx_get_row_ub
        lpx_get_col_bnds       lpx_get_col_type, lpx_get_col_lb,
                               lpx_get_col_ub
        lpx_get_obj_c0         lpx_get_obj_coef
        lpx_get_row_coef       (no more supported)
        lpx_get_col_coef       lpx_get_obj_coef
        lpx_get_row_mark       (no more supported)
        lpx_get_col_mark       (no more supported)
        lpx_get_row_info       lpx_get_row_stat, lpx_get_row_prim,
                               lpx_get_row_dual
        lpx_get_col_info       lpx_get_col_stat, lpx_get_col_prim,
                               lpx_get_col_dual
        lpx_get_ips_stat       lpx_ipt_status
        lpx_get_ips_row        lpx_ipt_row_prim, lpx_ipt_row_dual
        lpx_get_ips_col        lpx_ipt_col_prim, lpx_ipt_col_dual
        lpx_get_ips_obj        lpx_ipt_obj_val
        lpx_get_mip_stat       lpx_mip_status
        lpx_get_mip_row        lpx_mip_row_val
        lpx_get_mip_col        lpx_mip_col_val
        lpx_get_mip_obj        lpx_mip_obj_val

        Obsolete API routines were kept for backward compatibility,
        however, they will be removed in the future.

        * doc/refman.*
        New edition of the GLPK reference manual containing description
        of all new API routines was included.

        * contrib/glpkmex/*.*
        GLPKMEX, a Matlab MEX interface to GLPK package, contributed by
        Nicolo Giorgetti <[e-mail]> was included.

        * doc/GLPK_FAQ.txt
        GLPK FAQ contributed by Harley Mackenzie <[e-mail]> was
        included.

Fri Dec 12 12:00:00 2003 Andrew Makhorin <[e-mail]>

        * GLPK 4.3 has been released.

        * configure.in
        The bug, due to which the standard math library is not linked on
        some platforms, was fixed.

        * glpmpl3.c
        The bug (0 ** y) was fixed in the routine fp_power.

        * glpmpl.h, glpmpl1.c, glpmpl3.c
        Some new built-in functions (round, trunc, Irand224, Uniform01,
        Uniform, Normal01, Normal) were added to the MathProg language.

        * glpmpl1.c
        The MathProg syntax was changed to allow writing 'subj to'.

        * glplpx.h, glplpx1.c, glplpx2.c
        The new api routine lpx_get_ray_info was added.

        * glplpx8a.c
        The api routine lpx_print_sol was changed to print the number of
        non-basic variable, which causes primal unboundness.

        * glpmps.c
        The code was changed to avoid errors on compiling the package on
        Mac OS X. Thanks to Andre Girard <[e-mail]> for
        the bug report.

        * doc/lang.*, doc/refman.*
        Several typos were fixed and some new material was added in the
        glpk documentation.

Fri Nov 14 12:00:00 2003 Andrew Makhorin <[e-mail]>

        * GLPK 4.2 has been released.

        * glpiet.h, glpiet.c, glpios.h, glpios1.c, glpios2.c, glpios3.c
        A preliminary implementation of the Integer Optimization Suite
        (IOS) was included in the package. Eventually IOS will replace
        the Implicit Enumeration Suite (IES).

        * glplpx.h, glplpx6d.c
        A dummy version of the integer optimization routine lpx_intopt
        was included in the package. Later this routine will replace the
        routine lpx_integer.

        * examples/glpsol.c
        A new command-line option --int-opt was added to the solver to
        call lpx_intopt rather than lpx_integer.

        * glpbcs.h, glpbcs1.c, glpbcs2.c
        Being replaced by IOS routines (see above) the Branch-and-Cut
        Framework (BCS) routines were removed from the package.

        * examples/tspsol.c
        Stand-alone Symmetric TSP solver was completely re-programmed
        using IOS routines.

        * glplib.h, glplib2.c, glplib4.c
        The random-number generator was implemented. It is based on the
        module GB_FLIB from the Stanford GraphBase originally developed
        by Donald Knuth.

        * glphbsm.c, glplpx8a.c, glpmps.c
        All calls to fopen/fclose were replaced by corresponding calls
        to ufopen/ufclose due to bug reports provided by Morten Welinder
        <[e-mail]> and <[e-mail]>.

        * glpmps.c
        The code was made re-entrant.

        * glplpx8b.c
        API routine lpx_print_sens_bnds for bounds sensitivity analysis
        contributed by Brady Hunsaker <[e-mail]> was added
        to the package. This feature is also available in glpsol via the
        command-line option --bounds.

        * contrib/jni/*.*
        New version of GLPK JNI (Java Native Interface) contributed by
        Chris Rosebrugh <[e-mail]> was added to the package.

        * contrib/deli/*.*
        GLPK DELI (Delphi Interface) contributed by Ivo van Baren
        <[e-mail]> was added to the package.

        * glplpx3.c
        Default method to scale the problem was changed to equilibration
        scaling (lp->scale = 1 in lpx_reset_parms).

        * glplpx6a.c
        Two minor (non-critical) typos were fixed due to report provided
        by Andrew Hamilton-Wright <[e-mail]>.

        * glplpp2.c
        An untested case (line 941) had been tested due to bug report
        provided by Jiri Spitz <[e-mail]>.

        * w32bc5.mak, w32vc6.mak, w32vc6d.mak, d32dmc.mak
        Several makefiles were added to allow building GLPK library for
        some non-GNU 32-bit platforms.

Sat Aug 23 12:00:00 2003 Andrew Makhorin <[e-mail]>

        * GLPK 4.1 has been released.

        * glpmpl1.c, glpmpl3.c
        Some bugs were fixed in the MathProg translator due to the bug
        reports provided by Giles Thompson <[e-mail]>:
        conditional set expressions were incorrectly parsed;
        dimen attribute was not set by default when a set was used
        recursively in its own declaration;
        logical expressions ... in if ... then ... else ... did not
        work;
        displaying set expressions did not free memory allocated for
        temporary results.

        * glpmpl3.c (reduce_terms)
        Implementation of summation of linear forms over domain was
        improved to reduce complexity of that operation from O(n*n) to
        O(n*log n). The improvement was made due to a report provided
        by Sebastien de Menten <[e-mail]>.

        * glplpx6a.c (line 1056), glpmip1.c (line 641)
        Two minor bugs were fixed due to the bug report provided by
        Kendall Demaree <[e-mail]>.

        * glplpx.h, glplpx6a.c
        The method of one artificial variable implemented in the routine
        lpx_prim_art and used on the phase I in the glpk simplex solver
        has a serious defect: for some lp instances it erroneously
        reports that the problem has no primal feasible solution. This
        error appears when the column of the artificial variable, which
        enters the basis to make it primal feasible, has large
        constraint coefficients, that leads to small reduced costs of
        non-basic variables and premature termination of the search,
        i.e. to wrong conclusion that the problem has no primal feasible
        solution. To avoid this defect the routine lpx_prim_feas was
        included. It implements the method of implicit artifical
        variables (based on minimization of the sum of infeasibilities),
        which is a bit slower but much more robust. The routine
        lpx_prim_feas having the same functionality now is used instead
        the routine lpx_prim_art.

        * glpinv.h, glpinv.c
        The test used in the routine inv_update to detect low accuracy
        after updating LU-factorization of the basis matrix was replaced
        by a new, more robust test.

        * glplpx6c.c
        Selecting an active node to be solved next in the routine
        btrack_bestp was changed. Now, if any integer feasible solution
        has not been found yet, the routine chooses an active node which
        has the minimal sum of integer infeasibilities.

        * glpmip.h, glpmip1.c
        The additional flag int_obj was included in the structure
        MIPTREE used by the branch-and-bound. This flag is set in the
        routine mip_create_tree and used in the routine is_better. It
        means that the objective is integral, i.e. depends only on
        integer variables with integer objective coefficients. The test
        used in the routine check_integrality was also replaced by a
        new, more reasonable one.

        * glplpx1.c
        A minor bug was fixed in the routine lpx_check_name.

        * glpmpl.h, glpmpl4.c, glplpx8d.c
        The flag skip_data was added to the parameter list of the
        routine mpl_read_model. If this flag is set, the data section
        in the model file is ignored. Corresponding change was made in
        the routine lpx_read_model. Now, if both model and data files
        are specified, the data section in the model file is ignored.

        * glplpx8c.c
        A minor bug (wrong format used for writing free columns) in the
        routine lpx_write_lpt was fixed due to the bug report provided
        by Bernhard Schmidt <[e-mail]>

        * sample/glpsol.c
        The command-line parameter --tmlim, which allows limiting the
        solution time, was added.

        * doc/lang.*, doc/refman.*
        New edition of the GLPK documentation was included.

        * java-binding/*.*
        New version of the GLPK JNI (Java Native Interface) package was
        included in the distribution.

        * sample/lpglpk40.c
        A non-trivial example was added. It allows using GLPK as a base
        LP solver for Concorde, a program for solving Traveling Salesman
        Problem (TSP). For details see comments in lpglpk40.c.

        * sample/*.mod
        Some examples of LP and MIP models written in GNU MathProg were
        added.

Tue May 06 12:00:00 2003 Andrew Makhorin <[e-mail]>

        * GLPK 4.0 has been released.

        * glpmpl.h, glpmpl1.c, glpmpl2.c, glpmpl3.c, glpmpl4.c
        The model translator for the GNU MathProg modeling language was
        implemented and included in the package.

        * glplpx.h, glplpx8d.c
        The api routine lpx_read_model, which is an interface to the
        MathProg translator, was included in the package.

        * glplpx.h, glplpx8a.c
        The api routine lpx_print_prob for writing LP/MIP problem data
        in plain text format was included in the package.

        * sample/glpsol.c
        New version of the GLPK stand-alone LP/MIP solver that supports
        the GNU MathProg modeling language was implemented.

        * doc/lang.latex, doc/lang.dvi, doc/lang.ps
        The document "GLPK: Modeling Language GNU MathProg" was included
        in the package.

        * doc/refman.latex, doc/refman.dvi, doc/refman.ps
        New edition of the GLPK Reference Manual was included in the
        package.

        * glplpx8c.c
        A bug in the api routine lpx_write_lpt was fixed. Due to that
        bug an addressing error occured in the routine if the objective
        function has the non-zero constant term.

        * glplan.h, glplan1.c, glplan2.c, glplan3.c, glplan4.c,
        * glplan5.c, glplan6.c, glplan7.c, glplan8.c, glplpx8b.c
        All modules of the translator for the GLPK/L modeling language
        were removed from the package, because GLPK/L being completely
        superseded by GNU MathProg is no more supported.

Tue Mar 25 12:00:00 2003 Andrew Makhorin <[e-mail]>

        * GLPK 3.3 has been released.

        * glplpp.h, glplpp1.c, glplpp2.c
        An implementation of the built-in LP presolver was added to the
        package.

        * glplpx.h
        The flag presol was added to the structure LPX. This flag tells
        the lpx_simplex whether the built-in LP presolver should be used
        or not. By default this flag is off. Also three macros (namely
        LPX_E_NOPFS, LPX_E_NODFS, and LPX_K_PRESOL) that concern using
        the LP presolver were introduced.

        * glplpx3.c, glplpx6a.c
        These modules was changed to use the built-in LP presolver.

        * sample/glpsol.c
        Command line options --presol and --nopresol that concern using
        the LP presolver were added to the stand-alone LP/MIP solver.

        * glplan1.c
        This module was changed to allow declaring sets like A[1:10] in
        the models written in the GLPK/L modeling language.

        * doc/refman.latex, doc/lang.latex
        New editions of the documents "GLPK User's Guide" and "GLPK/L
        Modeling Language" were included in the distribution.

        * java-binding/*.*
        The package GLPK JNI (Java Native Interface) implementing Java
        binding for GLPK was included in the distribution. This package
        was developed and programmed by Yuri Victorovich <[e-mail]>.

Tue Feb 18 12:00:00 2003 Andrew Makhorin <[e-mail]>

        * GLPK 3.2.4 has been released.

        * glplpx6b.c
        The code was changed to allow auxiliary variables have non-zero
        objective coefficients.

        Also a minor bug was fixed (the constant term was not considered
        on displaying the objective function value).