]> www.ginac.de Git - ginac.git/log
ginac.git
8 years agoAdd rvalue reference ctors from epvectors for add, mul.
Richard Kreckel [Thu, 17 Dec 2015 08:32:15 +0000 (09:32 +0100)]
Add rvalue reference ctors from epvectors for add, mul.

8 years agoMerge some cosmetic patches.
Richard Kreckel [Wed, 16 Dec 2015 21:02:25 +0000 (22:02 +0100)]
Merge some cosmetic patches.

This is a collection of some minor optimizations and indentation fixes.

8 years agoMake .eval() evaluate top-level only.
Richard Kreckel [Wed, 16 Dec 2015 20:22:36 +0000 (21:22 +0100)]
Make .eval() evaluate top-level only.

With the previous patch, some old workarounds have become unnecessary:
If all expressions are evaluated, any object which is an aggregate of
expressions will only ever have to evaluate the top level. As such, it
has become pointless to evaluate child objects of an expression prior
to doing its own term-rewriting. This patch removes the evaluation of
children from all GiNaC objects. It also removes the now superfluous
parameter 'level' of the eval methods.

8 years agoMake add::eval(), mul::eval() work without compromise.
Richard Kreckel [Wed, 16 Dec 2015 12:00:30 +0000 (13:00 +0100)]
Make add::eval(), mul::eval() work without compromise.

If a user asks to evaluate an object, the expectation is that the
library returns an evaluated, canonical expression. Everything else is
a bug. (It doesn't matter whether the user asks explicitly or by
assigning to an ex.) It turns out that it was possible to construct
objects which didn't fully evaluate. This patch fixes this by making
eval() a little bit more careful.

This obsoletes the need to go through combine_ex_with_coeff_to_pair()
when constructing an epvector that is then used to construct an add
or mul. (Alas, this was omitted frequently enough...)

8 years agoImprove method of setting status_flags::dynallocated.
Richard Kreckel [Mon, 30 Nov 2015 22:06:53 +0000 (23:06 +0100)]
Improve method of setting status_flags::dynallocated.

There seems to be no way to obsolete the need to mark an object derived
from basic and handled by ex as being 'on the heap', at least none that
doesn't have significant performance impact. Having said that, this
patch aims at making this process simpler and more intuitive.

Where, before, one would return from a function returning an ex with
        return (new mul(a, b))->setflag(status_flags::dynallocated);
this patch lets us return with
        return dynallocate<mul>(a, b);
which should be much clearer. In any case, it involves less typing.

The two points where the status_flags::dynallocated are set are now
  * the dynallocate<B>(args...) template function and
  * the virtual duplicate() member functions.

This patch rolls out the new functionality throughout the library.

8 years agoRemove accidentally pushed binary.
Richard Kreckel [Sun, 29 Nov 2015 11:42:59 +0000 (12:42 +0100)]
Remove accidentally pushed binary.

In commit 79cad108, a compiled binary from the test suite was accidentally
added and pushed. Sorry.

8 years agoRename array of precomputed data in test suite.
Richard Kreckel [Sun, 29 Nov 2015 11:33:59 +0000 (12:33 +0100)]
Rename array of precomputed data in test suite.

Reason: C++17 may introduce a std::data<> template. Right now, the GCC 6.0
prerelease bails out at this code, when compiler with -std=c++17.
<http://en.cppreference.com/w/cpp/iterator/data>

8 years agoUse neseted initializer lists to construct matrix objects.
Richard Kreckel [Sat, 28 Nov 2015 14:43:46 +0000 (15:43 +0100)]
Use neseted initializer lists to construct matrix objects.

Add constructor of initializer_list<initializer_list<ex>> to matrix.
Use this syntax where, previously, ctor from comma-separated list of
elements was used. Deprecate the ctor from comma-separated list.

Note: The output format '[[a,b],[c,d]]' and ginsh syntax are
unchanged because lists are printed '{a,b,c}' and a matrix is not a
list of lists.

8 years ago[build] Update scripts for correct rpath for linking with CLN.
Richard Kreckel [Fri, 27 Nov 2015 10:49:16 +0000 (11:49 +0100)]
[build] Update scripts for correct rpath for linking with CLN.

Use up-to-date files from GNU gettext 0.19.6.

8 years agoUse initializer lists to construct container<>, lst.
Richard Kreckel [Thu, 26 Nov 2015 20:36:46 +0000 (21:36 +0100)]
Use initializer lists to construct container<>, lst.

Add constructor of initializer_list<ex> to container<C<ex>>. In
particular, this means that we can finally declare lst objects as
lst{a,2*b,2}. Convert GiNaC to this syntax throughout. Deprecate
the old constructors taking 1..16 ex parameters and the ones from
comma seaparated lists (without braces).

8 years agoInfrastructure: Update baseline language check M4 macros.
Richard Kreckel [Thu, 26 Nov 2015 14:04:52 +0000 (15:04 +0100)]
Infrastructure: Update baseline language check M4 macros.

Use the more general ax_cxx_compile_stdcxx.m4 macro from gnu.org
instead of the old ax_cxx_compile_stdcxx_11.m4.

8 years agoMake ample use of the contextual keyword 'override'.
Richard Kreckel [Thu, 26 Nov 2015 11:20:20 +0000 (12:20 +0100)]
Make ample use of the contextual keyword 'override'.

This patch adds the C++11 contextual keyword 'override' to every overridden
virtual function declaration except where that would incur macro duplications.
Along the way, it fixes some comments about member functions 'virtuality'.

8 years agoStandard header cleanup.
Richard Kreckel [Wed, 25 Nov 2015 14:17:20 +0000 (15:17 +0100)]
Standard header cleanup.

Sanitize some standard headers, remove include directives of unused
headers, complete configure check for used standard headers.

8 years agoReplace idx_is_not functor by a C++11 lambda.
Richard Kreckel [Wed, 25 Nov 2015 11:49:31 +0000 (12:49 +0100)]
Replace idx_is_not functor by a C++11 lambda.

8 years agoIn power::expand_add(), don't reserve excess monomial sizes.
Richard Kreckel [Wed, 25 Nov 2015 10:28:10 +0000 (11:28 +0100)]
In power::expand_add(), don't reserve excess monomial sizes.

There is no need to reserve n terms in each of the monomials of the
result of power(+(x,y,z...;0),n): We can compute it exactly as the
number of nonzero exponents in the multinomial expansion. The good
thing is that this counting is the same for each composition of a
partition, so it can be hoisted out of the loop over compositions.

8 years agoMake specialized power::expand() helpers static member fuctions.
Richard Kreckel [Wed, 25 Nov 2015 10:22:34 +0000 (11:22 +0100)]
Make specialized power::expand() helpers static member fuctions.

The class power member functions expand_add(), expand_add_2(), and
expand_mul() do not access any member variable of class power. (In
fact, the only reason not to hide them entirely is that they invoke
protected members of classs expairseq, add, and mul which requires
them to be friends of these classes.)

8 years agoTutorial: how to create noncommutative symbols?
Richard Kreckel [Tue, 24 Nov 2015 12:02:46 +0000 (13:02 +0100)]
Tutorial: how to create noncommutative symbols?

This seems to be frequently asked:
<http://www.ginac.de/pipermail/ginac-devel/2015-November/002213.html>

8 years agoAvoid some explicit copying of STL containers.
Richard Kreckel [Tue, 24 Nov 2015 10:57:35 +0000 (11:57 +0100)]
Avoid some explicit copying of STL containers.

8 years agomul::can_make_flat(): Finish ancient unfinished code.
Richard Kreckel [Tue, 24 Nov 2015 10:14:51 +0000 (11:14 +0100)]
mul::can_make_flat(): Finish ancient unfinished code.

This makes mul(expairseq({a*b,2},{b})) construct the canonical
mul(expairseq({a,3},{b,2})). Normally, such term rewriting is done
by the .eval() methods, but this is not done when a plain object of
a class derived from basic is constructed.

8 years agounit_matrix(r,c) can return an evaluated matrix object.
Richard Kreckel [Mon, 23 Nov 2015 20:38:30 +0000 (21:38 +0100)]
unit_matrix(r,c) can return an evaluated matrix object.

8 years agoMake (a+b+c+...)^n work for huuuge terms.
Richard Kreckel [Mon, 23 Nov 2015 20:21:18 +0000 (21:21 +0100)]
Make (a+b+c+...)^n work for huuuge terms.

This fix is neccessary for results exceeding 2^31 terms. While at it,
also restructured power::expand_add_() a bit to look more like the
more general power::expand_add().

8 years agoConvert some more ctors from copying to moving STL containers.
Richard Kreckel [Mon, 23 Nov 2015 18:36:53 +0000 (19:36 +0100)]
Convert some more ctors from copying to moving STL containers.

9 years agoReplace static const variables with constexpr, where applicable.
Richard Kreckel [Sun, 8 Nov 2015 22:33:33 +0000 (23:33 +0100)]
Replace static const variables with constexpr, where applicable.

9 years agoFix some pedantic compiler warnings.
Richard Kreckel [Sun, 8 Nov 2015 22:18:19 +0000 (23:18 +0100)]
Fix some pedantic compiler warnings.

Found with CLang++ -Wall. (There remain some more, but these are
more ugly to fix.)

9 years agoReplace GiNaC::ToString() with std::to_string().
Richard Kreckel [Sun, 8 Nov 2015 21:19:08 +0000 (22:19 +0100)]
Replace GiNaC::ToString() with std::to_string().

And, while at it, also make CMake exercise the new real_imag exam
introduced last week and document successful build with GCC-4.8.

9 years agoUse C++11 'nullptr' where applicable.
Richard Kreckel [Sat, 7 Nov 2015 21:24:29 +0000 (22:24 +0100)]
Use C++11 'nullptr' where applicable.

9 years agoRemove C++89 workaround for closing angle brackets in nested templates.
Richard Kreckel [Sat, 7 Nov 2015 12:23:26 +0000 (13:23 +0100)]
Remove C++89 workaround for closing angle brackets in nested templates.

9 years agoRemove 'discardable' option from ctors of container and derived classes.
Richard Kreckel [Sat, 7 Nov 2015 12:03:05 +0000 (13:03 +0100)]
Remove 'discardable' option from ctors of container and derived classes.

The whole idea of this was to allow the ctor to pilfer the data from the
constructed-from object, which is precisely the move semantics which
C++11 supports with rvalue references.

9 years agoAdd trivial shortcuts in expair plumbing of class add.
Richard Kreckel [Sat, 7 Nov 2015 00:50:00 +0000 (01:50 +0100)]
Add trivial shortcuts in expair plumbing of class add.

These little shortcuts in add::split_ex_to_pair(ex) and in
add::combine_ex_with_coeff_to_pair(ex, ex) avoid the creation of
new ex objects. These pieces of code are executed so often that this
patch speeds up GiNaC by 5-10%, depending on test.

9 years agoFix little missed optimization.
Richard Kreckel [Fri, 6 Nov 2015 14:04:49 +0000 (15:04 +0100)]
Fix little missed optimization.

9 years agoReplace dynamic throw() with static noexcept operator.
Richard Kreckel [Fri, 6 Nov 2015 00:00:55 +0000 (01:00 +0100)]
Replace dynamic throw() with static noexcept operator.

9 years agoAdd some minor optimizations.
Richard Kreckel [Thu, 5 Nov 2015 12:41:18 +0000 (13:41 +0100)]
Add some minor optimizations.

9 years agoSpeed up power::real_part() and power::imag_part().
Richard Kreckel [Thu, 5 Nov 2015 12:30:48 +0000 (13:30 +0100)]
Speed up power::real_part() and power::imag_part().

Add special case for real base and exponent (not just integer exponent).
Improve special case for integer exponent by explicitly constructing
the result using the Binomial expansion.
Add a test case for real/imaginary part expansion.

9 years agoUse AX_CXX_COMPILE_STDCXX_11 macro to force C++11 compilation.
Richard Kreckel [Tue, 3 Nov 2015 20:23:16 +0000 (21:23 +0100)]
Use AX_CXX_COMPILE_STDCXX_11 macro to force C++11 compilation.

9 years agoFix most remaining GCC compiler warnings.
Richard Kreckel [Tue, 3 Nov 2015 19:53:07 +0000 (20:53 +0100)]
Fix most remaining GCC compiler warnings.

9 years agoMove ctors for classes matrix and pseries.
Richard Kreckel [Tue, 3 Nov 2015 15:46:06 +0000 (16:46 +0100)]
Move ctors for classes matrix and pseries.

9 years agoUse C++11 range-based foor loops and auto, where possible.
Richard Kreckel [Tue, 3 Nov 2015 13:15:57 +0000 (14:15 +0100)]
Use C++11 range-based foor loops and auto, where possible.

9 years agoUpdate documentation of prerequisites.
Richard Kreckel [Sun, 1 Nov 2015 23:00:48 +0000 (00:00 +0100)]
Update documentation of prerequisites.

9 years agoReplace use of NULL by C++11 nullptr.
Richard Kreckel [Sun, 1 Nov 2015 22:48:05 +0000 (23:48 +0100)]
Replace use of NULL by C++11 nullptr.

9 years agoRemove dependence on depreacted std::auto_ptr<T>.
Richard Kreckel [Sun, 1 Nov 2015 19:16:12 +0000 (20:16 +0100)]
Remove dependence on depreacted std::auto_ptr<T>.

Most of the auto_ptr were introduced to avoid copying. C++11 supports
move semantics for this. Some other can be replaced by std::unique_ptr<T>.

9 years agoRemove support for Cint.
Richard Kreckel [Sat, 31 Oct 2015 16:18:55 +0000 (17:18 +0100)]
Remove support for Cint.

9 years agoRemove unfinished code for hash table-based expairseq.
Richard Kreckel [Sat, 31 Oct 2015 16:02:14 +0000 (17:02 +0100)]
Remove unfinished code for hash table-based expairseq.

This is unfinished code by A.F. Later generations don't ever
seem to have tinkered much with it.  ;-)

9 years agoFix make check with g++ -std=c++11.
Richard Kreckel [Thu, 15 Oct 2015 06:59:20 +0000 (08:59 +0200)]
Fix make check with g++ -std=c++11.

Our template function log() returning an object of type GiNaC::function
turns out to compete with a template from a libstdc++ header. We have
to help the compiler a bit.

9 years agoFix some apparent typos and misindentations 2.
Richard Kreckel [Sat, 10 Oct 2015 20:24:32 +0000 (22:24 +0200)]
Fix some apparent typos and misindentations 2.

(Formatting should comply with http://www.cebix.net/indentation.html.)

9 years agoFix algebraic power::has() for larger integer exponents.
Richard Kreckel [Sat, 10 Oct 2015 09:02:17 +0000 (11:02 +0200)]
Fix algebraic power::has() for larger integer exponents.

There is no need to overflow integers in (x^2147483648).has(x^2).

9 years agoInclude crc32.h only if needed.
Richard Kreckel [Mon, 21 Sep 2015 21:04:20 +0000 (23:04 +0200)]
Include crc32.h only if needed.

This shuts up a compiler warning about unused static function crc32.

9 years agoRemove unused variable.
Richard Kreckel [Mon, 21 Sep 2015 20:49:44 +0000 (22:49 +0200)]
Remove unused variable.

This shuts up a compiler warning.

9 years agoFix parser example in tutorial.
Richard Kreckel [Fri, 31 Jul 2015 22:52:11 +0000 (00:52 +0200)]
Fix parser example in tutorial.

Thanks to Jean-Michaël Celerier for pointing this out.

9 years agoFix some apparent typos and misindentations.
Richard Kreckel [Fri, 31 Jul 2015 22:41:30 +0000 (00:41 +0200)]
Fix some apparent typos and misindentations.

9 years agoFinalizing release 1.6.5. ginac_1-6-5
Richard Kreckel [Mon, 27 Jul 2015 19:28:44 +0000 (21:28 +0200)]
Finalizing release 1.6.5.

9 years agoFix bug in PolyLogs of rational numbers.
Stefan Weinzierl [Fri, 24 Jul 2015 18:50:57 +0000 (20:50 +0200)]
Fix bug in PolyLogs of rational numbers.

Thanks to Alexander Hasselhuhn for reporting this bug.

9 years agoSpeed up pow(+(...),n).expand() where n>2.
Richard Kreckel [Sat, 18 Jul 2015 22:04:49 +0000 (00:04 +0200)]
Speed up pow(+(...),n).expand() where n>2.

This new algorithm is faster than the old one because it re-uses
multinomial coefficients as much as possible.

9 years agoFix pow(+(...),2).expand().
Richard Kreckel [Sat, 18 Jul 2015 21:56:55 +0000 (23:56 +0200)]
Fix pow(+(...),2).expand().

Due to a failure to expand result terms, expand((sqrt(1+x)+y*sqrt(1+x))^2)
returned 1+y^2+x+x*y^2+2*y*(1+x). Note that 2*y*(1+x) was not expanded
to 2*y+2*y*x.

9 years agoFinalized 1.6.4 release. ginac_1-6-4
Richard Kreckel [Fri, 8 May 2015 07:35:24 +0000 (09:35 +0200)]
Finalized 1.6.4 release.

9 years agoFix pow(+(...),2).expand().
Richard Kreckel [Thu, 7 May 2015 20:33:13 +0000 (22:33 +0200)]
Fix pow(+(...),2).expand().

Due to a failure to recombine coeffs and rests to expairs,
expand((x+sqrt(2)*x)^2) returned x^2+2*x^2+2*sqrt(2)*x^2. The
2*x^2 term was not combined with the x^2 term to 3*x^2 because it
was not the canonical expair [[x^2,2]] but rather [[2*x^2,1]].

Thanks to Isuru Fernando for the bugreport.

9 years agoAdd test cases for improved abs(arg).
Richard Kreckel [Thu, 23 Apr 2015 20:33:52 +0000 (22:33 +0200)]
Add test cases for improved abs(arg).

Exam the evaluations abs(±sqrt(p)) -> sqrt(p) for p > 0.

9 years agoImprove abs(arg).
Vladimir Kisil [Wed, 22 Apr 2015 21:21:53 +0000 (23:21 +0200)]
Improve abs(arg).

Make abs(arg) return -arg if arg is negative or if -arg is known to be
nonnegative.

9 years agoFix pow::info(info_flags::nonnegative).
Vladimir Kisil [Wed, 22 Apr 2015 21:06:56 +0000 (23:06 +0200)]
Fix pow::info(info_flags::nonnegative).

This function call was missing the case where a positive base is raised
to a real power (where pow::info(info_flags::positive) correctly returned
true).

9 years agoFix mul::info(info_flags::nonnegative).
Vladimir Kisil [Wed, 22 Apr 2015 19:52:15 +0000 (21:52 +0200)]
Fix mul::info(info_flags::nonnegative).

Due to an obvious Boolean inversion, this function call always
returned the wrong answer.

9 years agoMake power::expand() (x*p)^c -> x^c * p^c, if p>0.
Vladimir Kisil [Tue, 14 Apr 2015 21:14:09 +0000 (23:14 +0200)]
Make power::expand() (x*p)^c -> x^c * p^c, if p>0.

This expansion seems to be helpful in many cases.

9 years agoMake compile_ex() work with CMake.
Martin Vala [Tue, 14 Apr 2015 20:50:31 +0000 (22:50 +0200)]
Make compile_ex() work with CMake.

Infrastructure fixes to make CMake find libdl and make ginac-excompiler
listen to the $CXXFLAGS environment variable.

9 years agoImprove normalization with nested functions.
Ladislav Zejda [Sat, 14 Mar 2015 10:12:11 +0000 (11:12 +0100)]
Improve normalization with nested functions.

normal() fails to fully normalize expressions where nested functions occur
multiple times because replace_with_symbol() searches repl for an
original expression but inserts into repl the substitued expression. Then
the repeated search does not work and a new symbol is introduced.

This patch fixes that by searching for the expression with the repl
substituted instead of for the original expression.

9 years agoFixing index in tutorial.
Vladimir V. Kisil [Sun, 8 Feb 2015 19:53:55 +0000 (20:53 +0100)]
Fixing index in tutorial.

The index entry for clifford::get_metric() did not work, change it to get_metric().

Signed-off-by: Vladimir V. Kisil <kisilv@maths.leeds.ac.uk>
9 years agoExplicit derivation of functions.
Vladimir V. Kisil [Sun, 8 Feb 2015 19:50:51 +0000 (20:50 +0100)]
Explicit derivation of functions.

Some function cannot be cleanly differentiated through the chain rule.
For example, it is natural to define derivative of the absolute value as

(abs(f))'=(f'*f.conjugate()+f*f'.conjugate())/2/abs(f)

This patch adds a possibility to define derivatives of functions in this way.
In particular the derivative of abs(), Order(), real_part(), imag_part() and
conjugate() are defined.

For example, conjugate of a derivative with respect of a real symbol
If x is real then U.diff(x)-I*V.diff(x) represents both
conjugate(U+I*V).diff(x) and conjugate((U+I*V).diff(x))
Thus in this patch we use the rule

conjugate(f)'=conjugate(f')

for a derivative with respect to the real symbol.

Signed-off-by: Vladimir V. Kisil <kisilv@maths.leeds.ac.uk>
9 years agoPut a reference on eta function into tutorial.
Vladimir V. Kisil [Sat, 31 Jan 2015 12:04:13 +0000 (13:04 +0100)]
Put a reference on eta function into tutorial.

9 years agoHappy new year!
Richard Kreckel [Sun, 25 Jan 2015 10:50:43 +0000 (11:50 +0100)]
Happy new year!

9 years agoSmall optimization in simplify_indexed.
Vladimir V. Kisil [Sun, 25 Jan 2015 10:39:04 +0000 (11:39 +0100)]
Small optimization in simplify_indexed.

If the new expression is product again then we call the same function return
simplify_indexed_product() rather than the its mother implify_indexed().

9 years agoEvaluate abs(z)^(2n) -> z^n * conjugate(z)^n.
Vladimir V. Kisil [Wed, 31 Dec 2014 18:07:26 +0000 (18:07 +0000)]
Evaluate abs(z)^(2n) -> z^n * conjugate(z)^n.

9 years agoFinalized 1.6.3 release. ginac_1-6-3
Richard Kreckel [Thu, 27 Nov 2014 21:29:39 +0000 (21:29 +0000)]
Finalized 1.6.3 release.

Updated NEWS file, moved some comments so Doxygen won't be confused.

10 years agoPrepare for 1.6.3 release.
Alexei Sheplyakov [Sun, 19 Jan 2014 17:50:23 +0000 (19:50 +0200)]
Prepare for 1.6.3 release.

10 years agoUpdate copyright statements.
Richard Kreckel [Tue, 14 Oct 2014 19:41:20 +0000 (19:41 +0000)]
Update copyright statements.

10 years agoThe FSF has moved (long ago).
Richard Kreckel [Tue, 14 Oct 2014 19:27:34 +0000 (19:27 +0000)]
The FSF has moved (long ago).

Update some its postal address in some remaining places.

10 years agoFix log() error in multiple polylog G.
Stefan Weinzierl [Sun, 12 Oct 2014 18:30:41 +0000 (18:30 +0000)]
Fix log() error in multiple polylog G.

Tobias Huber reported the following bug:
> G({-0.18784426860496276*I,-0.1579006477353911*I,+0.18784426860496276*I,-0.18784426860496276*I},1);
log_eval(): log(0)

The problem is the sorting operation in G_do_trafo. This operation should
1) put the elements in increasing order of |x|
2) put equal elements next to each other (which then will avoid the log(0) problem).

The current code fails for goal 2) if we have complex numbers of equal absolute
value, one number occuring more than once, in an initial order like in the example
above:
Entries 1,3,4 of the list all have the same absolute value, entries 1 and 4 are
equal. In the sorting operation 1 and 4 should be put next to each other.
Previously, the sorting operation would give 2,1,3,4. What we would like to have
is either 2,1,4,3 or 2,3,1,4. It is clear that the key for the sorting operation
cannot be |x| alone. If |x1|=|x2| we have to use the phase as well.

This patch takes the phase into account in the sorting.

10 years agoFix evaluation of some G-polylogs.
Stefan Weinzierl [Sun, 12 Oct 2014 18:24:07 +0000 (18:24 +0000)]
Fix evaluation of some G-polylogs.

This patch corrects and supersedes the patches
97ef604e "G_numeric: fix numeric evaluation with trailing zeros and y != 1."
and
9e80b0d3 "G_numeric: fix evaluation with y == 1".

The original motivation for 97ef604e was to make sure that Hoelder convolution
is only used if there no trailing zeros.

Patch 97ef604e delegated the case of trailing zeros to G_do_trafo, which correctly
removes the trailing zeros, but in addition also performed the transformation
described in sect. 5.3 of hep-ph/0410259 ("convergence transformation").
The inappropriate call of the convergence transformation is the cause of the new bug
reported at http://www.cebix.net/pipermail/ginac-list/2014-September/002011.html

The patch 9e80b0d3 cures the symptons mentioned in the above bug report,
but fails for other cases, like evalf( G({-2,2,-2,0},2) ).

What should be done is the following: If trailing zeros are detected in G_numeric,
these should be removed and the result should be returned to G_numeric. The
routine G_numeric decides then what to do next: either calling the convergence
transformation, or Hoelder convolution or direct summation. What is needed is a
subroutine, which just removes trailing zeros, but does not perform the convergence
transformation.

With the present code the minimal modification to achieve this goal is to add an
additional boolean parameter flag_trailing_zeros_only to G_do_trafo (and its
dependent sub-routines), so G_do_trafo can be called for the removal of trailing
zeros only.

This patch implements this and uses G_do_trafo to remove trailing zeros only for
the case at hand.

10 years agoG_numeric: fix evaluation with y == 1
Alexei Sheplyakov [Fri, 12 Sep 2014 13:50:51 +0000 (17:50 +0400)]
G_numeric: fix evaluation with y == 1

Do apply Hoelder convolution if y == 1 even if there are trailing zeros.
Fixes the regression introduced by commit 97ef604e ("G_numeric: fix
numeric evaluation with trailing zeros and y != 1.")

10 years agoAdd more ex::is_polynomial() tests
Alexei Sheplyakov [Sun, 7 Sep 2014 18:56:02 +0000 (22:56 +0400)]
Add more ex::is_polynomial() tests

Improves commit 92da4c16 ('[bugfix] ex::is_polynomial(x): fix miscomputations')
a little bit.

10 years agoProviding default (un)archiving for cliffordunit class.
Vladimir V. Kisil [Mon, 18 Aug 2014 20:29:01 +0000 (21:29 +0100)]
Providing default (un)archiving for cliffordunit class.

Signed-off-by: Vladimir V. Kisil <kisilv@maths.leeds.ac.uk>
10 years ago[bugfix] ex::is_polynomial(x): fix miscomputations
Richard Kreckel [Fri, 25 Jul 2014 23:11:11 +0000 (23:11 +0000)]
[bugfix] ex::is_polynomial(x): fix miscomputations

The function power::is_polynomial(x) sometimes returned false,
even if x did not occur in the power object.

Thanks to Rubel Zilibowitz for reporting this bug.

Similarly, mul::is_polynomial(x) sometimes returned true, even
if x occurred only in the denominator of a rational function.

The function mul::is_polynomial(x) does a simpler version of
power::is_polynomial(x) on its factors. (It's simpler because the
exponents cannot be functions of x.)

10 years agoFix example of symmetrization in tutorial.
Richard Kreckel [Tue, 22 Jul 2014 07:48:25 +0000 (07:48 +0000)]
Fix example of symmetrization in tutorial.

The old example was illegal C++. Thanks to Rubel Zilibowitz for
reporting this error.

10 years ago[bugfix] log_series: avoid infinite recursion
Alexei Sheplyakov [Tue, 1 Jul 2014 18:46:11 +0000 (21:46 +0300)]
[bugfix] log_series: avoid infinite recursion

Check if the argument is on a branch cut check if it actually depends
on the expansion variable before trying to expand around a different
point in order to avoid the infinite recursion (or rather a segfault
due to a stack overflow).

Thanks to Mario Prausa for a bugreport.

10 years agoFix an infinite loop in factor_multivariate.
Alexei Sheplyakov [Sun, 18 May 2014 10:24:40 +0000 (13:24 +0300)]
Fix an infinite loop in factor_multivariate.

Make sure the leading coefficient is really factorized (take care of
the integer content).

10 years agoG_do_hoelder: fix the transformation of the imaginary part.
Stefan Weinzierl [Sat, 12 Apr 2014 11:32:20 +0000 (14:32 +0300)]
G_do_hoelder: fix the transformation of the imaginary part.

The G_do_hoelder function transforms the arguments as z -> 1-z.
If z has an imaginary part, the sign of the imagniary part transforms
as s -> -s.
In the special case where z is real and > 1, the transformed value is
real and < 0. For (transformed) real negative values of z the sign of
the imaginary part is not relevant (there are no branch cuts). The code
requires that in this case the sign takes the default value +1.
This is now fixed.

10 years agoG_numeric: fix numeric evaluation with trailing zeros and y != 1.
Stefan Weinzierl [Wed, 12 Feb 2014 20:07:40 +0000 (22:07 +0200)]
G_numeric: fix numeric evaluation with trailing zeros and y != 1.

Make sure G_numeric makes use of scaling relation (G_do_hoelder) only if
there are no traling zeros. Use G_do_trafo otherwise (which can handle
trailing zeros).

10 years ago[docs] remind users tho send EOF marker when doing input from cin.
Richard Kreckel [Mon, 3 Feb 2014 08:02:53 +0000 (08:02 +0000)]
[docs] remind users tho send EOF marker when doing input from cin.

The example program waiting infinitely for EOF has confused at least one
user: <http://www.ginac.de/pipermail/ginac-list/2014-January/001983.html>

10 years agoBinary compatibilty has been broken, update the library version.
Alexei Sheplyakov [Sun, 2 Feb 2014 17:55:22 +0000 (19:55 +0200)]
Binary compatibilty has been broken, update the library version.

commit 798d53ebb4da4e8e3865ed7bd7f31412fe2be3a7 (`Make it possible to override
info() for functions.') overrides the function::info() virtual method. This
breaks binary compatibility (adds a new entry in the middle of the vtable of
the `function' class).

The experimental proof of the above statement: ginsh linked with the previous
release of libginac segfaults if libginac.so is upgraded to the current
version.

Thereore bump the library version.

10 years ago[build] Document that Python is required for building GiNaC.
Alexei Sheplyakov [Sun, 2 Feb 2014 10:56:46 +0000 (12:56 +0200)]
[build] Document that Python is required for building GiNaC.

10 years ago[build] Fix `make pdf' for out of the source (auto* tools) build.
Alexei Sheplyakov [Sun, 2 Feb 2014 15:49:18 +0000 (17:49 +0200)]
[build] Fix `make pdf' for out of the source (auto* tools) build.

10 years ago[build] Include the generated sources (function.{h,cpp}) into the tarball.
Alexei Sheplyakov [Sun, 2 Feb 2014 15:47:38 +0000 (17:47 +0200)]
[build] Include the generated sources (function.{h,cpp}) into the tarball.

10 years ago[build] Include CMake related files into the tarball.
Alexei Sheplyakov [Sat, 1 Feb 2014 16:56:29 +0000 (18:56 +0200)]
[build] Include CMake related files into the tarball.

10 years ago[build] Include the helper scripts into the tarball.
Alexei Sheplyakov [Sat, 1 Feb 2014 16:50:37 +0000 (18:50 +0200)]
[build] Include the helper scripts into the tarball.

10 years agocheck: auto* tools build: compile and run check_mul_info.
Alexei Sheplyakov [Fri, 31 Jan 2014 19:46:26 +0000 (21:46 +0200)]
check: auto* tools build: compile and run check_mul_info.

10 years ago[build] ginsh: fix `make distcheck' failure.
Alexei Sheplyakov [Tue, 28 Jan 2014 06:59:48 +0000 (08:59 +0200)]
[build] ginsh: fix `make distcheck' failure.

10 years agolog_eval: don't apply the log(x^p) -> p*log(x) rule.
Alexei Sheplyakov [Sat, 18 Jan 2014 12:13:46 +0000 (14:13 +0200)]
log_eval: don't apply the log(x^p) -> p*log(x) rule.

Automatically transforming log(p^2) into 2 log(p) might be quite annoying.
Perhaps this transformation should be moved to expand() and applied only
if expand_options::expand_transcendental is specified.

10 years agoLi_eval(): avoid the "numeric::operator>(): complex inequality" exception.
Stefan Weinzierl [Tue, 28 Jan 2014 06:58:06 +0000 (08:58 +0200)]
Li_eval(): avoid the "numeric::operator>(): complex inequality" exception.

Basically the problem is that 1 + 0.0*I is equal to 1, but checking if
the former is positive results in the the "numeric::operator>(): complex
inequality" exception. This behavior is definitely inconsistent and should
be fixed.

As a work around teach Li_eval and convert_parameter_Li_to_H to explicitly
convert 1 + 0.0*I to 1 (and -1 + 0.0*I to -1, respectively).

[Alexei Sheplyakov: comment the code a little bit, otherwise it looks weird]

10 years agoFix S_num for arguments close to the sixth root of unity or its conjugate.
Stefan Weinzierl [Sun, 12 Jan 2014 22:31:22 +0000 (22:31 +0000)]
Fix S_num for arguments close to the sixth root of unity or its conjugate.

The method S_num within the Nielsen polylogs used to map the region
abs(x)<=1 && abs(x)>0.95 && abs(1-x)<=1 && abs(1-x)>0.95 infinitely many times
onto itself. This infinite recursion is now avoided.

This however reveals the next problem: The numerical convergence in this
region is very slow. Within the Nielsen polylogs there is no transformation
available to improve the convergence. However we can use the (1-x)/(1+x)
transformation within the harmonic polylogs. In order to avoid another
infinite recursion I have inserted a few hold()'s in the method H_evalf,
otherwise we would fall back immediately again to Nielsen polylogs. The hold()s
should have been there anyway.

10 years agoFix G3_evalf and G3_eval for real and negative x.
Stefan Weinzierl [Sun, 12 Jan 2014 22:19:56 +0000 (22:19 +0000)]
Fix G3_evalf and G3_eval for real and negative x.

The user supplied signs of the imaginary parts are only relevant for x real
and positive. A negative sign in the case of x real and negative is irrelevant
(there is no branch cut) and led to wrong results. This is now avoided.

10 years agoWithin mLi_numeric, set signs of the imaginary parts correctly for G_numeric.
Stefan Weinzierl [Sun, 12 Jan 2014 22:07:28 +0000 (22:07 +0000)]
Within mLi_numeric, set signs of the imaginary parts correctly for G_numeric.

10 years agoAvoid calling log(1-x) for x=1 in Li_projection.
Stefan Weinzierl [Sun, 12 Jan 2014 21:58:12 +0000 (21:58 +0000)]
Avoid calling log(1-x) for x=1 in Li_projection.

10 years agoFix evaluation of log(p^a) -> a*log(p), if p>0 and a is real.
Vladimir V. Kisil [Sat, 28 Dec 2013 23:58:40 +0000 (23:58 +0000)]
Fix evaluation of log(p^a) -> a*log(p), if p>0 and a is real.

This evaluation was broken with commit e5c76f659e2e882d.

10 years agoexpairseq::match(): no side effects if match failed. Fixes spurious match failures.
Matthias Dellweg [Tue, 10 Dec 2013 08:55:28 +0000 (10:55 +0200)]
expairseq::match(): no side effects if match failed. Fixes spurious match failures.

> match(sin(y)*exp(b)+sin(x)*exp(a), sin($0)*exp(a)+exp(b)*sin($1))
FAIL

The reason is that expairseq::match() might assign a wildcard even if
the match fails. The first attempted submatch is sin(y)*exp(b) with
sin($0)*exp(a). It fails (as it should) but $0 == y gets assigned as
a side effect (which is wrong). Next submatch is sin(x)*exp(a) with
sin($0)*exp(a) (the same pattern as in the first submatch). This one
fails because of spurious $0 == y assignment.

Due to the unpredicatable term ordering the sequence of submatches might
be different and the match might succeed (as it should). This makes
debugging a bit more funny.

Signed-off-by: Matthias Dellweg <dellweg@tp1.uni-duesseldorf.de>
[Alexei Sheplyakov: figure out the cause of the problem, make a test case]