Richard Kreckel [Tue, 27 Apr 2010 20:44:35 +0000 (22:44 +0200)]
Fix weird ctor invocation syntax.
This patch is required for GCC-4.5.
Jens Vollinga [Mon, 29 Mar 2010 00:34:52 +0000 (02:34 +0200)]
Preparing for release 1.5.7.
Jens Vollinga [Sun, 28 Mar 2010 22:21:18 +0000 (00:21 +0200)]
Products (class mul) now answer correctly to info_flags::negative and
info_flags::negint.
(cherry picked from commit
8a30acc990818792ec4e8f8f4d48f5dd8286dbed)
Richard Kreckel [Thu, 25 Mar 2010 22:08:54 +0000 (23:08 +0100)]
Really fixed the atan2(-Pi, 0) bug.
The problem was that atan2_eval assumed that if y is real and not
positive, it must be negative. But this neglects the fact that in
symbolic compution it may just not be possible to determine the
sign. Now, the expression is returned as it is. Ugly, but correct.
Jens Vollinga [Thu, 25 Mar 2010 09:36:41 +0000 (10:36 +0100)]
Fixed a bug in atan2. It gave a division by zero error for calls like
atan2(-Pi,0), because arguments like -Pi were not recognized (via
info_flags) as negative but as real nevertheless.
(cherry picked from commit
9e13d46552bb7852399867b9eb355732b9ded59e)
Jens Vollinga [Fri, 5 Mar 2010 20:37:19 +0000 (21:37 +0100)]
Copyright goes 2010.
(cherry picked from commit
95ede03f00106be5fb087d76c86ca5c27edb9122)
Alexei Sheplyakov [Tue, 23 Feb 2010 10:01:23 +0000 (12:01 +0200)]
pgcd(): detect relatively prime polynomials properly...
... so pgcd() doesn't loop forever any more. Division test does not handle
relatively prime polynomials (because C = 1 divides any polynomial). So we
should stop interpolation (and return gcd of contents) if we got relatively
prime images (we should do that for efficiency reasons too).
Thanks to Jörg Arndt for a bug report.
(cherry picked from commit
6afbb9793e359cf388462c471ea256a6662b0cd4)
Jens Vollinga [Thu, 28 Jan 2010 14:25:01 +0000 (15:25 +0100)]
Preparing for release 1.5.6.
Alexei Sheplyakov [Wed, 6 Jan 2010 17:55:43 +0000 (19:55 +0200)]
Use C style cast when converting void* into function pointer.
Building GiNaC 1.5.5 with GCC 3.4 fails with the following error:
libtool: compile: ccache g++-3.4 -DHAVE_CONFIG_H -I. -I../../ginac -I../config -I/home/pc7135/varg/target/x86_64-linux-gnu/include -O2 -g -Wall -pipe -MT builtin_fcns.lo -MD -MP -MF .deps/builtin_fcns.Tpo -c ../../ginac/parser/builtin_fcns.cpp -fPIC -DPIC -o .libs/builtin_fcns.o
../../ginac/parser/builtin_fcns.cpp: In function `GiNaC::ex (* GiNaC::encode_serial_as_reader_func(unsigned int))(const GiNaC::exvector&)':
/home/pc7135/varg/tmp/build/GiNaC/build-linux-gcc-3.4/ginac/../../ginac/parser/builtin_fcns.cpp|67| error: ISO C++ forbids casting between pointer-to-function and pointer-to-object
make[2]: *** [builtin_fcns.lo] Error 1
The C++98 standard [expr.reinterpret.cast] does not allow
reinterpret_cast<function_pointer>(void*)
reinterpret_cast<void*>(function_pointer)
But the ability to do so is important for a lot of practical uses. So soon
after the C++98 standard was approved, a language defect report was filed
on this topic, see
http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html#195
The result is that compilers will be allowed to support reinterpret_cast
conversions of function pointers to other (pointers) types, and vice a versa.
Such conversions work with *current* compilers (GCC 4.x), but don't work
with older ones, hence this patch.
Alexei Sheplyakov [Wed, 6 Jan 2010 17:56:04 +0000 (19:56 +0200)]
Install the ginac-excompiler script iff excompiler is supported and enabled.
(cherry picked from commit
25c7a8c09f7db73b48675777aa805e788f2308be)
Jens Vollinga [Wed, 4 Nov 2009 22:12:57 +0000 (23:12 +0100)]
Preparing for release.
Richard Kreckel [Wed, 4 Nov 2009 20:58:26 +0000 (21:58 +0100)]
Fix libtool versioning fiasco.
Fix
f2f67dde5c9caca, which caused the soname to be bumped too frequently.
Jens Vollinga [Mon, 2 Nov 2009 22:29:21 +0000 (23:29 +0100)]
Preparing for release.
Richard Kreckel [Sun, 11 Oct 2009 21:19:21 +0000 (23:19 +0200)]
Add mul::info() and add::info() support for numeric info_flags.
This way, GiNaC knows that the sum or product of a number of positive
expressions is again a positive expression. [by Vladimir Kisil]
Richard Kreckel [Sun, 4 Oct 2009 20:34:33 +0000 (22:34 +0200)]
Merge branch 'ginac_1-5' of ssh://ginac.de/srv/git/ginac into ginac_1-5
Richard Kreckel [Sun, 4 Oct 2009 20:30:45 +0000 (22:30 +0200)]
Output LaTeX-symbols using \mathrm, instead of \mbox.
Using \mathrm improves printing in complex formulae.
One reason is that it scales in fractions. [by Vladimir Kisil]
Jens Vollinga [Sun, 4 Oct 2009 20:15:13 +0000 (22:15 +0200)]
Moved header includes to source files to avoid polluting the namespace
for library users.
Richard Kreckel [Tue, 29 Sep 2009 20:28:47 +0000 (22:28 +0200)]
Remove debian/ directory.
The lifecycle of the Debian packaging files is better maintained in
the Debian pool's diff file. Removing these files makes the life of
the GiNaC releaser easier.
Richard Kreckel [Tue, 29 Sep 2009 20:08:11 +0000 (22:08 +0200)]
Properly document how to deal with libtool versioning.
The old description was referring to ginac_lt_age which should always
stay 0, because we promised to mark incompatible binary interfaces by
bumping ginac_minor_version (or even ginac_major_version). Let's get
rid of ginac_lt_age.
Alexei Sheplyakov [Sat, 8 Aug 2009 10:03:38 +0000 (13:03 +0300)]
ncmul::eval(): don't write beyond the end of the vector.
Alexei Sheplyakov [Sat, 8 Aug 2009 10:03:15 +0000 (13:03 +0300)]
modular_matrix: don't use STL iterators in {mul, sub}_col and friends.
Not using STL iterators makes the code simpler and cleaner. Also we don't
have to argue whether the standard ([lib.random.access.iterators]) imposes
any pre/post-condictions on operator+=.
(cherry picked from commit
59ec13895c97ffd74979a0b811a7571f6de56386)
Jens Vollinga [Sun, 9 Aug 2009 21:38:48 +0000 (23:38 +0200)]
Added get_builtin_reader() that parses only the builtin GiNaC functions
and pow, sqrt, and power.
Jens Vollinga [Sun, 9 Aug 2009 21:27:10 +0000 (23:27 +0200)]
Fixed include of stdint.h (parser.cpp needs the header as well).
Alexei Sheplyakov [Sat, 8 Aug 2009 10:03:48 +0000 (13:03 +0300)]
shaker_sort, permutation_sign: fix invalid use of STL iterators.
According to the standard incrementing end(), and decrementing begin()
are ill-defined operations (see [lib.forward.iterators],
[lib.bidirectional.iterators]).
(cherry picked from commit
694f839947982f5b12b6c629d5bab522c801630d)
Alexei Sheplyakov [Sat, 8 Aug 2009 10:07:14 +0000 (13:07 +0300)]
symmetry::calchash(): be careful to not dereference past-the-end iterator.
(cherry picked from commit
526825bcee294cb20e37d8db6d5040ba0f8c428f)
Alexei Sheplyakov [Sat, 8 Aug 2009 10:06:36 +0000 (13:06 +0300)]
check_parameter_G: fix pontential increment of end().
Incrementing past-the-end iterator is not permitted by the standard, see
[lib.input.iterators].
(cherry picked from commit
49a44f7d55ec0d6686d3c32c7081a07d10c93274)
Alexei Sheplyakov [Sat, 8 Aug 2009 10:02:47 +0000 (13:02 +0300)]
G_eval: fix incorrect use of STL iterator.
According to [lib.bidirectional.iterators] it's not OK to decrement
an iterator pointing to the beginning of the sequence. Fortunately random
access iterators provided by (current versions of) gcc/libstdc++ don't have
this silly limitation, so the code which works with pointers works with
iterators without any changes at all. However,
- there's no guarantee that the current behavior won't change in the future
- some non-GNU compilers are not that smart (i.e. the program segfaults
upon when decrementing beginning-of-the-sequence iterator).
(cherry picked from commit
dda45abd8a2c408f8b3eb7959a10dfb2468ecc3a)
Alexei Sheplyakov [Sat, 8 Aug 2009 08:43:12 +0000 (11:43 +0300)]
Fix the compliation error *for real* ... and restore performance
Commit
8bf0597dde55e4c94a2ff39f1d8130902e3d7a9b (titled as 'Fixed the parser
such that it can read in user defined classes again.') made the parser a bit
slower, especially if the input contains many terms of user-defined type.
The reason for that is quite simple: we throw and catch an exception every
time we construct an object of user-defined type:
// dirty hack to distinguish between serial numbers of functions and real
// pointers.
GiNaC::function* f = NULL;
try {
unsigned serial = (unsigned)(unsigned long)(void *)(reader->second);
f = new GiNaC::function(serial, args);
}
catch ( std::runtime_error ) {
if ( f ) delete f;
ex ret = reader->second(args);
return ret;
}
Fortunately functions are aligned and we can use much more efficient
technique to distinguish between serial and pointers to functions.
Alexei Sheplyakov [Fri, 7 Aug 2009 20:22:18 +0000 (23:22 +0300)]
Fix the compliation error *for real*
Jens Vollinga [Fri, 31 Jul 2009 15:54:16 +0000 (17:54 +0200)]
Fixed memory leak.
Jens Vollinga [Fri, 31 Jul 2009 13:29:17 +0000 (15:29 +0200)]
Jens Vollinga [Fri, 31 Jul 2009 12:41:08 +0000 (14:41 +0200)]
Fixed dirty hack in parser to distinguish between serial numbers and pointers.
Jens Vollinga [Fri, 31 Jul 2009 10:48:58 +0000 (12:48 +0200)]
Fixed the parser such that it can read in user defined classes again.
Fixed default reader to parse also pow, sqrt and power.
Jens Vollinga [Fri, 31 Jul 2009 09:14:01 +0000 (11:14 +0200)]
Fixed cast that caused compile error on 64bit machines.
Jens Vollinga [Thu, 30 Jul 2009 17:04:32 +0000 (19:04 +0200)]
Preparing for release 1.5.3.
Jens Vollinga [Wed, 15 Jul 2009 03:19:07 +0000 (05:19 +0200)]
Allow underscores in identifiers.
(cherry picked from commit
619879b5bd86a99334affe1f05d17616cc2c1898)
Jens Vollinga [Wed, 15 Jul 2009 06:26:33 +0000 (08:26 +0200)]
Allow user defined functions to be parsed.
This patch is an ugly hack that does the same as the commit
f38cbcd651246fb5c1294705d29399f3cbfddaf5
but without changing the ABI (so it can be used in ginac_1-5).
Jens Vollinga [Fri, 10 Jul 2009 13:21:09 +0000 (15:21 +0200)]
Preparing for release 1.5.2.
Jens Vollinga [Fri, 10 Jul 2009 07:54:09 +0000 (09:54 +0200)]
Fixed manual to correctly explain return_type_tinfo().
(cherry picked from commit
feed241b95f7dbd6294795b4afc2bcae41880c31)
Jens Vollinga [Wed, 10 Jun 2009 09:28:34 +0000 (11:28 +0200)]
Raised required version of CLN to 1.2.2.
Jens Vollinga [Fri, 24 Apr 2009 21:37:44 +0000 (23:37 +0200)]
Added include for cstdio header.
gcc 4.0 does not include cstdio as part of other headers like string
anymore. As a result, gcc 4.0 complained about EOF being undefined.
Jens Vollinga [Wed, 25 Feb 2009 14:03:38 +0000 (15:03 +0100)]
Repairing the preparing of the release.
Jens Vollinga [Wed, 25 Feb 2009 12:07:18 +0000 (13:07 +0100)]
Preparing for release.
Alexei Sheplyakov [Mon, 23 Feb 2009 12:08:42 +0000 (14:08 +0200)]
calchash(): work around broken RTTI.
Due to the strange (although permitted by the standard) behaviour of C++
RTTI on woe32 calchash() returns different hash values for equal objects.
As a result automatic evaluation gets spectacularly broken:
examining clifford objects.....({1+t,2+x,3+y,4+z}) - ({1+t,2+x,3+y,4+z}) erroneously returned -{1+t,2+x,3+y,4+z}+{1+t,2+x,3+y,4+z} instead of 0
({1+t,2+x,3+y,4+z}) - ({1+t,2+x,3+y,4+z}) erroneously returned -{1+t,2+x,3+y,4+z}+{1+t,2+x,3+y,4+z} instead of 0
.({1+t,2+x,3+y,4+z}) - ({1+t,2+x,3+y,4+z}) erroneously returned -{1+t,2+x,3+y,4+z}+{1+t,2+x,3+y,4+z} instead of 0
({1+t,2+x,3+y,4+z}) - ({1+t,2+x,3+y,4+z}) erroneously returned {1+t,2+x,3+y,4+z}-{1+t,2+x,3+y,4+z} instead of 0
[skipped]
.......FAIL: exam_clifford.exe
This patch works around `features' of woe32 RTTI, so calchash() works
properly.
Alexei Sheplyakov [Mon, 23 Feb 2009 12:03:16 +0000 (14:03 +0200)]
Fix compilation failure due to (template) operator- defined in factor.cpp
GiNaC 1.5.0 fails to compile with g++ 3.4.
The fix is simple: declare (and define) operator- (and operator+) only
for (univariate) polynomials.
Alexei Sheplyakov [Mon, 16 Feb 2009 14:23:16 +0000 (16:23 +0200)]
Univariate GCD timing: use sr_gcd when appropriate.
The benchmark consists of two parts:
1) timing of different GCD calculation methods (i.e. subresultant PRS,
heuristic, chinese remaindering).
2) timing of different implementations of the same method. The purpose
is to find out how (in)efficient GiNaC::ex is as a representation
of (univariate) polynomials (as a side note, the result is a bit
depressing -- using coefficient vector instead of GiNaC:ex makes
GCD calculation 50x -- 1000x faster).
Now GiNaC uses modular (chinese remaindering) GCD by default, so part 2)
got broken, i.e. instead of (intended) timings
a) (heuristic, GiNaC::ex) versus (heuristic, coefficient vector)
b) (PRS, GiNaC::ex) versus (PRS, coefficient vector)
one gets
a') (heuristic, GiNaC::ex) versus (heuristic, coefficient vector)
b') (chinese remaindering, GiNaC::ex) versus (PRS, coefficient vector)
Set the gcd_options::use_sr_gcd to restore the meaning of the benchmark.
Note: this patch does not affect the library proper.
Alexei Sheplyakov [Mon, 16 Feb 2009 14:24:15 +0000 (16:24 +0200)]
Polished NEWS a little bit.
Jens Vollinga [Tue, 17 Feb 2009 13:38:17 +0000 (14:38 +0100)]
Added release date.
Jens Vollinga [Fri, 6 Feb 2009 15:49:47 +0000 (16:49 +0100)]
Adjusted NEWS and INSTALL documentation.
Jens Vollinga [Fri, 6 Feb 2009 15:35:22 +0000 (16:35 +0100)]
Missed to change one date in function.pl.
Jens Vollinga [Fri, 6 Feb 2009 15:24:58 +0000 (16:24 +0100)]
Changed name of debugging macro (to avoid warning cause by multiple definition).
Jens Vollinga [Fri, 6 Feb 2009 15:22:58 +0000 (16:22 +0100)]
Fixed returning of test result (++result was +result).
Jens Vollinga [Fri, 6 Feb 2009 15:07:10 +0000 (16:07 +0100)]
Prettified source code.
- Added copyright and GPL licencing to new files.
- Increased year to 2009.
- Changed guarding macros in header to uniform pattern without leading or
trailing __ (double underscores).
- Put includes of system wide header consistently below own includes (help
a tiny bit to clarify dependencies).
Alexei Sheplyakov [Fri, 6 Feb 2009 06:57:32 +0000 (08:57 +0200)]
Fix chinese_remainder() so modular GCD actually works.
Jens Vollinga [Fri, 6 Feb 2009 12:13:06 +0000 (13:13 +0100)]
Automake now needs to be version >=1.8. Older versions like 1.7.9 do not create
the directory m4 in a VPATH build.
Jens Vollinga [Thu, 5 Feb 2009 09:54:21 +0000 (10:54 +0100)]
Renamed files *.tcc and *.hpp to *.h.
Jens Vollinga [Tue, 3 Feb 2009 12:50:33 +0000 (13:50 +0100)]
Fixed compile errors introduced in previous commit
45b1e47372090352ac5af655b32473df2abab23b.
Alexei Sheplyakov [Mon, 19 Jan 2009 06:45:38 +0000 (08:45 +0200)]
Implement modular multivariate GCD (based on chinese remaindering algorithm).
Both algorithm and implementation are a bit inefficient:
* algorithm does not make use of sparseness of inputs (and most of
multivariate polynomials are quite sparse)
* GiNaC's expressions are essentially immutable. Thus some simple
operations (i.e. multiplying a polynomial by an element of the base ring)
are prohibitively expensive.
* All numbers (i.e. GiNaC::numeric objects) are heap allocated.
Still it's much faster (~5x on bivariate polynomials, ~100x on 3-variate
ones) than (subresultant) PRS algorithm, so gcd() uses modular algorithm
by default.
Alexei Sheplyakov [Mon, 19 Jan 2009 06:44:36 +0000 (08:44 +0200)]
Improve (fix?) smod: now it really converts into symmetric representation...
... instead of clumsy convention inspirited by some proprietary CAS.
Alexei Sheplyakov [Mon, 19 Jan 2009 06:43:50 +0000 (08:43 +0200)]
Fix copy-paste error in GINACLIB_MICRO_VERSION.
Jens Vollinga [Thu, 8 Jan 2009 09:26:50 +0000 (10:26 +0100)]
Fixed bug in binomial(). binomial(n,0) erroneously returned 0.
Jens Vollinga [Thu, 18 Dec 2008 11:19:35 +0000 (12:19 +0100)]
Small fixes to avoid compile warnings (-Wall).
Jens Vollinga [Thu, 18 Dec 2008 10:10:50 +0000 (11:10 +0100)]
Fixed bug in multivariate factorization. Fractional numerical content was not
extracted and led to a not so finite running behavior. Added new checks for that
bug.
Added static keywords to hide debugging output operators.
Jens Vollinga [Mon, 1 Dec 2008 09:26:38 +0000 (10:26 +0100)]
Added new items.
Alexei Sheplyakov [Mon, 1 Dec 2008 09:13:12 +0000 (10:13 +0100)]
[PATCH] [build] Simplify generation of ginac/version.h
Alexei Sheplyakov [Mon, 1 Dec 2008 09:12:25 +0000 (10:12 +0100)]
[PATCH] Fix GiNaC library version information.
The libtool naming scheme cannot guarantee that on all systems, the numbering
is consecutive. It only guarantees that it is increasing. This doesn't matter,
though: there is not incurred cost for numbers that are omitted, except for
shrinking the available space of leftover numbers. Not something we need to
worry about yet. ;-) On the other hand, tricks which we were using to make
version numbers consecutive on GNU/Linux break versioning on other OSes.
Jens Vollinga [Mon, 1 Dec 2008 08:54:53 +0000 (09:54 +0100)]
Fixed a problem in polynomial divide by deactivating a recursion that lead to a
significant slowdown in sqrfree() and sometimes to a infinite loop.
Jens Vollinga [Fri, 14 Nov 2008 11:00:52 +0000 (12:00 +0100)]
Added section about factor(). Made appropriate changes at other places.
Jens Vollinga [Thu, 13 Nov 2008 15:36:02 +0000 (16:36 +0100)]
Added many comments.
Jens Vollinga [Thu, 13 Nov 2008 10:13:05 +0000 (11:13 +0100)]
Fixed lots of bugs in factor_multivariate().
factor_univariate() takes now the content of the polynomial into accout when
choosing a prime. It also returns the chosen prime. This supports the
factor_multivariate() code.
Polynomials are expanded before calling factor_sqrfree(). This avoids problems
with some input polynomials.
Added static qualifiers to all hidden functions.
Jens Vollinga [Tue, 11 Nov 2008 09:33:29 +0000 (10:33 +0100)]
Fixed a bug in squarefree(). Some polynomials were erroneously classified as
square free (e.g. x^prime+1).
Fixed a bug in multivar_diophant() causing sporadically an infinite loop.
Improved lifting bound code.
Univariate lifting can now use a cache for the modular factors. At the moment,
this gives no measurable speed gain.
Jens Vollinga [Mon, 10 Nov 2008 12:38:11 +0000 (13:38 +0100)]
Added modular square free factorization.
Completed distinct degree factorization.
Univariate polynomial factorization uses now upoly.
Merged class Partition and function split into class factor_partition.
Jens Vollinga [Thu, 6 Nov 2008 13:18:01 +0000 (14:18 +0100)]
Stupid error: did the revert on the wrong branch (master instead of ginac_1-4).
Revert "Revert "[portability] GiNaC::compile_ex works on OpenBSD now.""
This reverts commit
a6854a8a0f1e5db722320f5d7ccb9097053500cf.
Jens Vollinga [Thu, 6 Nov 2008 13:13:54 +0000 (14:13 +0100)]
Revert "[portability] GiNaC::compile_ex works on OpenBSD now."
This reverts commit
375a7aa1f13fa6061c278eb2c2447fd91521b45d.
Jens Vollinga [Thu, 6 Nov 2008 13:11:02 +0000 (14:11 +0100)]
Univariate Hensel lifting now uses upoly.
Changed q_matrix code.
Paul Irofti [Tue, 4 Nov 2008 14:55:34 +0000 (17:55 +0300)]
Equip ginac-examples.texi with directory entry for info.
Alexei Sheplyakov [Mon, 3 Nov 2008 16:17:16 +0000 (19:17 +0300)]
[portability] GiNaC::compile_ex works on OpenBSD now.
* acinclude.m4: check for dlopen() in libdl and in libc itself.
Thanks to Paul Irofti for a bug report and a suggestion.
Alexei Sheplyakov [Mon, 3 Nov 2008 14:34:29 +0000 (17:34 +0300)]
[build] Set correct rpath for linking with CLN...
... so users won't get 'error while loading shared libraries: no such file
or directory'. Also embed rpath into pkg-config meta data, so
g++ `pkg-config --cflags --libs ginac` foo.cc
actually works without any additional effort from the user side.
Alexei Sheplyakov [Thu, 23 Oct 2008 13:48:14 +0000 (17:48 +0400)]
[bugfix] integer_cra: check if arguments contain at least 2 moduli
While at it, also fix exam_cra so it does not produce pointless inputs.
Alexei Sheplyakov [Tue, 21 Oct 2008 06:09:33 +0000 (10:09 +0400)]
exam_cra: explicitly #include <limits>. This should make g++ 4.3 happy.
Jens Vollinga [Wed, 5 Nov 2008 12:56:53 +0000 (13:56 +0100)]
Changed debugging facilities in factor.cpp.
Jens Vollinga [Wed, 5 Nov 2008 10:31:35 +0000 (11:31 +0100)]
Fixed some minor spelling errors.
Jens Vollinga [Wed, 5 Nov 2008 10:22:19 +0000 (11:22 +0100)]
Changed code from using cl_UP_MI to using umodpoly. The cl_UP_MI interface was
inconvenient to use and caused several very difficult bugs (some were still
unresolved before changing to umodpoly!).
Fixed severe bug in multivariate factorization. The condition that all modular
factors should be relatively prime in the base ring was violated. This caused
the factorization sometimes to go into an infinite loop.
Jens Vollinga [Mon, 3 Nov 2008 14:50:31 +0000 (15:50 +0100)]
Added code for distinct degree factorization.
Alexei Sheplyakov [Sun, 19 Oct 2008 16:21:33 +0000 (20:21 +0400)]
tutorial: simplify the instructions on writing extension classes.
Don't mention (un)archiving while describing the `mystring' class
(the minimalistic extension class), so readers have less chances to get
confused.
Alexei Sheplyakov [Sun, 19 Oct 2008 15:39:30 +0000 (19:39 +0400)]
Don't force every algebraic class to implement archiving/unarchiving.
So people who don't use (un)archiving don't need to bother with it.
Alexei Sheplyakov [Fri, 17 Oct 2008 13:57:27 +0000 (17:57 +0400)]
tutorial: don't mention custom RTTI any more (as it does not exist).
Alexei Sheplyakov [Thu, 16 Oct 2008 11:43:32 +0000 (15:43 +0400)]
symbol: make get_domain() a virtual method, remove symbol::domain.
This cuts yet another 4 bytes from GiNaC::symbol. While at it, fix a memory
leak in ginsh.
Alexei Sheplyakov [Thu, 16 Oct 2008 11:43:32 +0000 (15:43 +0400)]
symbol: don't bother to set TeX name if user did not specified one.
Same for the usual name. As a result we use less memory (and the output
is exactly the same).
Alexei Sheplyakov [Fri, 17 Oct 2008 10:56:54 +0000 (14:56 +0400)]
symbol: remove return_type_tinfo() and return_type() (shrink symbol by 8 bytes)
Bloating symbol to the state sizeof(symbol) == 64 is not appreciated at all.
If someone needs/wants non-commutative symbols or other fancy stuff, please
derive from class symbol and do whatever you want.
Alexei Sheplyakov [Fri, 17 Oct 2008 09:41:01 +0000 (13:41 +0400)]
Wipe out remnants of custom RTTI.
Custom RTTI considered harmful, final part.
Alexei Sheplyakov [Wed, 15 Oct 2008 11:32:11 +0000 (15:32 +0400)]
registered_class_info: use typeid() instead of tinfo_static.
(Custom RTTI considered harmful, part 5).
Since custom RTTI is going to be removed, registered_class_info needs to be
re-organazied. Now tinfo_key is a (const) pointer to std::type_info.
Alexei Sheplyakov [Fri, 17 Oct 2008 09:15:03 +0000 (13:15 +0400)]
A better return_type_tinfo() mechanism.
return_type_tinfo() is used in order to distingish between non-commutative
objects of different type. However, often it's necessary to distingish
between non-commutative objects of the same type, for example, between
gamma matrices with different representation label. return_type_tinfo()
does not provide any clean way to do that. Hence, introduce return_type_t
type which holds representation label along with type information, and
use it for return_type_tinfo().
Alexei Sheplyakov [Wed, 15 Oct 2008 11:32:11 +0000 (15:32 +0400)]
calchash(): use type_info::name() instead of tinfo().
Custom RTTI considered harmful, part 4.
The hash value of the object of different types should be different whenever
possible. Hence calcash() needs a unique per type number. Previously we used
tinfo_key for this. typeinfo::name() (a *pointer* to implementation dependent
string) is also unique for each class, so it's just as good as tinfo() is.
Alexei Sheplyakov [Wed, 15 Oct 2008 11:32:11 +0000 (15:32 +0400)]
indexed::eval: use standard C++ RTTI.
Custom RTTI considered harmful, part 3.
Alexei Sheplyakov [Wed, 15 Oct 2008 11:32:11 +0000 (15:32 +0400)]
basic, expairseq: use standard C++ RTTI in compare(), is_equal(), match(), etc.
Custom RTTI considered harmful, part 2.
* basic.cpp: use standard C++ RTTI in compare(), is_equal(), operator=(),
match().
* expairseq.cpp: use standard C++ RTTI in match(), make_flat(),
construct_from_2_ex().
Alexei Sheplyakov [Wed, 15 Oct 2008 11:32:11 +0000 (15:32 +0400)]
is_exactly_a: use typeid() to check the type of expression.
Custom RTTI considered harmful, part 1.
Custom run time type information (RTTI) system implemented in GiNaC have
several serious drawbacks, such as
1. It makes writing GiNaC classes unnecessary cumbersome.
2. It wastes sizeof(void *) bytes per object, for small objects like
symbol, numeric, etc. this overhead is considerable.
It turns out that GiNaC's RTTI is not any faster than the standard one
(at least on GNU/Linux and Solaris with GNU C++ library and compiler).
So, GiNaC RTTI have no reasons to exit any more.
Alexei Sheplyakov [Wed, 15 Oct 2008 06:16:07 +0000 (10:16 +0400)]
ptr.h: use unsigned int to store reference counts (in order to save memory).
We'll hardly ever get more than 2^32 references to the same object. So using
size_t to store reference count only wastes 4 bytes per object (on 64-bit
architecture). Use unsigned int instead.
Alexei Sheplyakov [Sun, 19 Oct 2008 17:27:14 +0000 (21:27 +0400)]
Rewritten heuristic and PRS GCD for univariate polynomials, added benchmark.
Using a better data structure for univariate polynomials makes GCD 10 -- 300
times faster (and less memory hungry).
Alexei Sheplyakov [Tue, 14 Oct 2008 06:44:07 +0000 (10:44 +0400)]
polynomial: introduce a helper function to divide polynomial by ring element.