A "smart" LP perturbation was implemented in the primal and
dual simplex solvers. Now LP is perturbed only if it is
necessary, and by default perturbation is not activated.
The sum of primal infeasibilites that appears in the terminal
output of the primal simplex solver (as "inf = ...") now
corresponds to the original bounds of variables. This allows to
see how much perturbed solution violates the original bounds.
The long-step technique was implemented for phase I of the
primal simplex solver. This feature can be enabled by
specifying --flip option for glpsol or by specifying
glp_smcp.r_test = GLP_RT_FLIP on api level. For many LP
instances the long-step technique allows reducing the number
of simplex iterations to 30-70%. Please note that unlike the
dual simplex, where this technique can be used on both phase I
and II, for the primal simplex it can be used only on phase I,
where the sum of primal infeasibilities (which is a convex
piecewise linear function) is minimized.
An internal objective scaling was included in both primal and
dual simplex solvers. For many LP/MIP instances this feature
improves numerical stability (for the dual solver) and prevents
cycling near the optimum (for the primal solver).
The Posix version of glp_time (glpk/src/env/time.c) was changed
to resolve time_t issue on msys2.
Three new example models in MathProg were added:
life_goe.mod (Conway's Game of Life garden of eden checker);
tiling.mod (Rectifiable polyomino tilings generator);
toto.mod (Covering code generator).
The following modules were renamed to simplify maintenance.
* src/prob.h RENAMED -> src/api/prob.h
* src/glpapi01.c RENAMED -> src/api/prob1.c
* src/glpapi02.c RENAMED -> src/api/prob2.c
* src/glpapi03.c RENAMED -> src/api/prob3.c
* src/glpapi04.c RENAMED -> src/api/prob4.c
* src/glpapi05.c RENAMED -> src/api/prob5.c
* src/env/tls.c
TLS (thread local storage class specifier) option was added;
see comments in tls.c for details.
* configure.ac, config.h.in
Test for TLS was added.
* src/env/tls.c
Dll support was added.
The following modules were changed to add __cdecl specifier
for functions passed to qsort (this is needed only on compiling
GLPK with MSVC to run under MS Windows).
* src/api/cpxbas.c
* src/cglib/cfg1.c
* src/cglib/gmigen.c
* src/cglib/mirgen.c
* src/misc/wclique1.c
* src/simplex/spychuzc.c
* src/glpios10.c
* src/glpios11.c
* examples/glpsol.c
* src/glpk.h, src/env/env.c
The API routine glp_version was changed to avoid initialization
of the GLPK environment. The new API routine glp_config was
added (but not documented yet).
* INSTALL
Description of the configure option '--with-zlib' was removed.
Some improvements were made in the primal and dual simplex
solvers to make the solution process more numerically stable.
An experimental long-step ratio test feature was added to the
dual simplex. On API level this feature is available thru the
GLP_RT_FLIP option. For glpsol it is available thru the options
--flip (for MIP) or --flip and --dual (for LP). This feature is
not documented yet.
Additional check was added to reject wrong solutions sometimes
reported by the PROXY heuristic.
A bug (memory leak) was fixed in the FPUMP heuristic routine.
The header sql.h was renamed to avoid conflicts with standard
ODBC headers.
A new, more efficient implementation of the dual simplex method
was included in the package. This new implementation replaces
the old one, which was removed.
Option sr_heur was added to struct glp_iocp to enable/disable
the simple rounding heuristic used by the MIP solver.
New API routine glp_at_error was added and documented.
Some minor typos were corrected in the GLPK documentation.
An example application program TSPSOL was added. It uses the
GLPK MIP optimizer to solve the Symmetric Traveling Salesman
Problem and illustrates "lazy" constraints generation. For more
details please see glpk/examples/tsp/README.
Some internal (non-API) routines to estimate the condition of
the basis matrix were added. These routines are mainly intended
to be used by the simplex-based solvers.
Two open modes "a" and "ab" were added to GLPK I/O routines.
Minor bug was fixed in the solver glpsol (command-line options
--btf, --cbg, and --cgr didn't work properly).
A serious bug was fixed in a basis factorization routine used
on the dense phase. (The bug might appear only if the number of
rows exceeded sqrt(2**31) ~= 46,340 and caused access violation
exception because of integer overflow.)
Two API routines glp_alloc and glp_realloc were documented.
Translation of the document "Modeling Language GNU MathProg"
to Spanish was included (in LaTeX and pdf formats).
Block-triangular LU-factorization was implemented to be used
on computing an initial factorization of the basis matrix.
A new version of the Schur-complement-based factorization
module was included in the package. Now it can be used along
with plain as well as with block-triangular LU-factorization.
Currently the following flags can be used to specify the type
of the basis matrix factorization (glp_bfcp.type):
GLP_BF_LUF + GLP_BF_FT LUF, Forrest-Tomlin update (default)
GLP_BF_LUF + GLP_BF_BG LUF, Schur complement, Bartels-Golub
update
GLP_BF_LUF + GLP_BF_GR LUF, Schur complement, Givens rotation
update
GLP_BF_BTF + GLP_BF_BG BTF, Schur complement, Bartels-Golub
update
GLP_BF_BTF + GLP_BF_GR BTF, Schur complement, Givens rotation
update
In case of GLP_BF_FT the update is applied to matrix U, while
in cases of GLP_BF_BG and GLP_BF_GR the update is applied to
the Schur complement.
Corresponding new options --luf and --btf were added to glpsol.
For more details please see a new edition of the GLPK reference
manual included in the distribution.
A minor bug (in reporting the mip solution status) was fixed.
A call to "iodbc-config --cflags" was added in configure.ac
to correctly detect iodbc flags.
