ncmul::expand() doesn't allocate a new object unless necessary

[ginac.git] / NEWS

This file records noteworthy changes.

1.2.2 (<date>)
* Added const_preorder_iterator and const_postorder_iterator classes (and
  associated methods ex::pre/postorder_begin/end()) providing tree traversal
  with iterators.
* Fixed the LaTeX output of the varidx class.
* Symbolic functions without any eval(), evalf() etc. functions now work
  properly.

1.2.1 (23 April 2004)
* Fixed infinite recursion in atan2_evalf() and improved atan2_eval().
* Added automatic evaluations for trigonometric functions with negative
  arguments (e.g. sin(-2) -> -sin(2)).
* Fixed a static initialization order goof-up.
* Fixed various bugs in series expansion.

1.2.0 (19 March 2004)
* Added a structure<T> template class for the easy creation of user-defined
  algebraic classes.
* Added support for (acyclic) visitors, to allow cleaner implementations of
  algebraic algorithms.
* Added a const_iterator class that can be used instead of op()/nops().
* Completely revamped the implementation of expression output. It is now
  possible to add new output formats, to change the behavior of predefined
  formats at run-time, and to have different output styles for algebraic
  functions.
* Symbols can be made non-commutative.
* Added a method ex::conjugate() and a function conjugate() for complex
  conjugation. Symbols can be declared as real or complex-valued.
* Improved the speed of subs(), normal(), to_rational() and to_polynomial()
  by the use of maps instead of lists. The old forms
    subs(const lst & ls, const lst & lr, unsigned options)
    to_rational/to_polynomial(lst & repl)
  are still available for compatibility, but using the new forms
    subs(const exmap & m, unsigned options)
    to_rational/to_polynomial(exmap & repl)
  is more efficient, especially when the number of replacements is large.
* quo(), rem(), prem(), sprem(), decomp_rational(), unit(), content(),
  primpart() and matrix::charpoly() now take a "const ex &" instead of a
  "const symbol &".
* Redundant expressions (two ex pointing to different objects are found to be
  equal in compare()) are now actively deleted/fused to conserve memory and
  speed up subsequent comparisons. This behavior can be suppressed on a
  per-object level with status_flags::not_shareable. Lists and matrices are
  not shareable by default.
* Lots of internal streamlining and optimizations.
* Caveats for class implementors:
   - basic::copy() and basic::destroy() are gone; classes derived from
     basic can use the defaults for the assignment operator and copy
     constructor.
   - basic::subs(), basic::normal(), basic::to_rational() and
     basic::to_polynomial() take 'exmap' objects instead of lists.
   - basic::subs() now descends into subexpressions (if accessible via
     nops()/op()/let_op()). If you have a custom implementation of subs()
     that calls basic::subs() after substituting subexpressions, this needs
     to be changed to a call to subs_one_level().
   - lst::thislst() and exprseq::thisexprseq() renamed to thiscontainer().
   - thiscontainer() and associated constructors now take a std::auto_ptr.
   - Overloading basic::print() is now deprecated. You should use
     print_func<>() class options instead.

1.1.7 (11 March 2004)
* Fixed a bug in canonicalize_clifford().
* Series expansion now works predictably. All terms with the exponent of the
  expansion variable smaller than the given order are calculated exactly. If
  the series is not terminating, the Order function is (at least) of the given
  order.

1.1.6 (22 January 2004)
* Added a function option "dummy()" which means "no options". This simplifies
  the implementation of symbolic functions which are not to be further
  evaluated.
* Removed a bug in the numerical evaluation of Li() that caused the system
  to hang for certain parameter combinations.
* Fixed a bug in the calculation of hash values for indices that could lead
  to wrong results or bogus error messages from simplify_indexed().
* Fixed a bug in the evaluation of harmonic polylogarithms for complex
  arguments with positive imaginary part.

1.1.5 (5 November 2003)
* Harmonic polylogarithms now numerically evaluate for arbitrary arguments
  (parameter must still be positive integers).
* The zeta function now can also be given a lst as a parameter in which case
  it becomes a multiple zeta value. The use of mZeta is deprecated.
* The order of parameters for the multiple polylogarithm has been corrected.
* Documentation for the nested sums functions zeta, harmonic polylog, multiple
  polylog, etc. has been added.

1.1.4 (17 October 2003)
* Lists and matrices can now be initialized from comma-separated lists of
  expressions, like this:

    lst l;
    l = x, 2, y, x+y;

    matrix M(3, 3);
    M = x, y, 0,
       -y, x, 0,
        0, 0, 1;

  This is both faster and produces much smaller code than the old constructors
  lst(ex, ex, ...) and matrix(unsigned, unsigned, lst), especially in the case
  of matrices, and is now the recommended way to create these objects.
* The function mZeta now evaluates much faster for arbitrary parameters. The
  harmonic and multiple polylogarithms evaluate considerably faster and check
  for convergence. The order of parameters for the harmonic polylogarithm
  has been corrected.

1.1.3 (22 August 2003)
* Added new symbolic functions for better integration with nestedsums:
  (multiple) polylogarithm Li(), Nielsen's generalized polylogarithm S(),
  harmonic polylogarithm H(), and multiple zeta value mZeta().
* New exhashmap<T> template intended as a drop-in replacement for
  std::map<ex, T, ex_is_less> using GiNaC's hashing algorithms.

1.1.2 (11 August 2003)
* Fixed a bug in the unarchiving of sums and products: terms were not
  reordered in a canonical way.
* Fixed a bug in normal()/numer_denom(): denominator was not made unit
  normal if it was a simple number.
* Improved the speed of subs() in some cases.

1.1.1 (18 June 2003)
* lst (and exprseq) provide iterators for read-only element access. For
  sequential access this is one order faster than using op().
* Implemented relational::subs() (this was done in 1.0.9 but inadvertently
  omitted from the 1.1 branch).
* pole_error and do_taylor are available to library users.
* Added on-line help and Tab-completion for print(), iprint(), print_latex()
  and print_csrc() in ginsh.

1.1.0 (3 April 2003)
* Removed deprecated macros is_ex_a, is_ex_exactly_a and friends for good.
* The scalar_products mechanism allows the specification of an index dimension.
* Removed dirac_gamma6/7().
* Added ex::to_polynomial().
* subs() accepts an optional "options" argument. The option
  subs_option::subs_algebraic enables "smart" substitutions in products and
  powers.
* Added stream manipulators "dflt", "latex", "python", "python_repr", "tree",
  "csrc", "csrc_float", "csrc_double", "csrc_cl_N", "index_dimensions" and
  "no_index_dimensions" to control the output format. Calling basic::print()
  directly is now deprecated.
* Made the hashing more simple and efficient.
* Caveats for class implementors:
   - basic::subs(): third argument changed from "bool" to "unsigned"
   - unarchiving constructor and basic::unarchive(): "const" removed from
     second argument
   - basic::let_op() should only be implemented if write access to
     subexpressions is desired
   - simplify_ncmul() renamed to eval_ncmul()
   - simplified_ncmul() renamed to hold_ncmul()
   - nonsimplified_ncmul() renamed to reeval_ncmul()

1.0.14 (1 March 2003)
* Improved the C-source output: complex numbers are printed correctly (using
  the STL complex<> template or cln::complex()), rational numbers use cl_RA()
  in the CLN output, and small integers are printed in a more compact format
  (e.g. "2.0" instead of "2.0000000e+00").
* function_options::set_return_type() and function_options::do_not_evalf_params()
  now actually work.

1.0.13 (27 January 2003)
* Contracting epsilon tensors with Euclidean indices now works.
* Improved dummy index symmetrization in sums.
* Added dirac_gammaL/R(), which can be used instead of dirac_gamma6/7()
  but are single objects, to allow for a more compact notation of Dirac
  strings.
* Powers with negative numeric exponents are printed as fractions in the
  LaTeX output.
* Added symbolic_matrix() for the convenient creation of matrices filled
  with symbols.
* Added collect_common_factors() which collects common factors from the
  terms of sums.
* simplify_indexed() converts "gamma~mu*p.mu" to "p\".

1.0.12 (30 October 2002)
* Fixed a bug in power::expand() that could produce invalid expressions.
* The input parser no longer ignores extra data following accepted input.
* Improved the CLN C-source output (integers are printed as integers, and
  floating point numbers include the precision).
* Fixed a problem in the LaTeX-output of negative fractions.
* Added print_latex() and print_csrc() to ginsh.
* The sprem() function is now public.

1.0.11 (18 September 2002)
* Fixed a possible memory corruption in contractions of indexed objects with
  delta or metric tensors.
* Computing the derivative of a power series object with respect to a symbol
  that is not the expansion variable now works correctly.
* Several bugfixes in code generation.

1.0.10 (24 July 2002)
* Powers of indexed objects are now parenthesized correctly in LaTeX output.
* Input parser handles indices (they have to be specified in the same list
  as the symbols).
* Added some limited support for subspaces in the idx and tensor classes.
* Fixed a bug in canonicalize() (antisymmetric canonicalization of an already
  sorted list containing two or more equal objects failed to return 0).

1.0.9 (11 June 2002)
* simplify_indexed() now raises/lowers dummy indices to canonicalize the index
  variance. This allows some simplifications that weren't possible before,
  like eps~a.b~c~d*X.a*X~b -> 0 and X.a~a-X~a.a -> 0.
* Implemented relational::subs().
* Fixed bug in simplify_ncmul() for clifford objects.

1.0.8 (31 March 2002)
* Improvements in memory usage of the expand() methods.

1.0.7 (18 March 2002)
* Fixed LaTeX output of indexed and matrix objects.
* Fixed matrix::pow(n) for n==0 and added helper functions to create unit
  matrices "ex unit_matrix(unsigned, unsigned)".

1.0.6 (4 March 2002)
* "(x+1).subs(x==x-1)" now returns the correct result "x" instead of "x-1".

1.0.5 (27 January 2002)
* (l)degree(s), coeff(s, n) and collect(s) were extended to accept expressions
  of any class (except add/mul/ncmul/numeric) for "s". They should even work
  if "s" is a "power" object, as long as the exponent is non-integer, but with
  some limitations. For example, you can "collect(a*2^x+b*2^x, 2^x)" to get
  "(a+b)*2^x", but "degree(2^(3*x), 2^x)" yields 0 instead of 3).
* Fixed a small output bug.

1.0.4 (24 January 2002)
* Speedup in expand().
* Faster Bernoulli numbers (Markus Nullmeier).
* Some minor bugfixes and documentation updates.

1.0.3 (21 December 2001)
* Fixed a bug where quo() would call vector::reserve() with a negative
  argument.
* Fix several bugs in code generation.

1.0.2 (19 December 2001)
* Input parser recognizes "sqrt()", which is also used in the output.
* divide(a,b,q) only modifies q if the division succeeds; also, divide(a,b,a)
  works now.
* Fixed small bug in dummy index renaming which could cause it to not
  recognize renamable indices in some cases.
* power::degree() and power::ldegree() throw an exception when encountering
  a non-integer exponent.
* Add output-support for Python bindings.

1.0.1 (22 November 2001)
* Function sqrfree() handles a few more cases now.
* Class relational has real canonical ordering now.
* Handle obscene libreadline version numbers when building ginsh.

1.0.0 (6 November 2001)
* Some internal reorganization resulting in a general speed-up.
* The last 3 evaluated expressions in ginsh are now referred to with the
  tokens '%', '%%' and '%%%'. The old '"', '""' and '"""' remain for
  compatibility but may be removed in a future version of GiNaC.

0.9.4 (20 September 2001)
* Functions have better support for external scripting languages.
* Interface cleanups and bugfixes.
* Fix silly bug in evalf() that prevented things like 2^Pi being computed.

0.9.3 (16 August 2001)
* series expansion now much more consistent for small order expansion.
* lsolve() accepts algorithmic hint as parameter.

0.9.2 (31 July 2001)
* Epsilon tensor is more functional.
* simplify_indexed() is better at detecting expressions that vanish for
  symmetry reasons.
* Several little bugfixes and consistency enhancements.

0.9.1 (27 June 2001)
* Ctors of class numeric are not explicit any more.  All built-in callers for
  pseudofunctions are now templated and default to ex arguments which relaxes
  the need for explicit ctors.
* New functions/methods:
   - find()
   - remove_first(), remove_last(), sort() and unique() for lists
   - symmetrize_cyclic()
   - decomp_rational()
* Instead of just totally symmetric or antisymmetric, complex symmetries
  can now be defined for indexed objects. Symmetries are described by a
  tree of "symmetry" objects that is constructed with the sy_none(),
  sy_symm(), sy_anti() and sy_cycl() functions. The symmetry of a function
  with respect to its arguments can also be defined (this is currently
  only used for the Beta function).
* Generalized map() to take a function object instead of a function pointer.
  This allows passing an arbitrary number of additional state to the
  function being called.
* color_trace(), dirac_trace(), diff(), expand(), evalf() and normal() work
  better with container classes, e.g. using color_trace() on a relation will
  take the trace on both sides, using diff() on a matrix differentiates every
  element etc.
* diff() works properly with non-commutative products and indexed objects.
* New option flag "expand_function_args" for expand().
* Supplement some (now deprecated) macros by inlined template functions:
  - is_of_type(foo, type) -> is_a<type>(foo)
  - is_ex_of_type(foo, type) -> is_a<type>(foo)
  - is_exactly_of_type(foo, type) -> is_exactly_a<type>(foo)
  - is_ex_exactly_of_type(foo, type) -> is_exactly_a<type>(foo)
  - ex_to_foobar(baz)  ->  ex_to<foobar>(baz)
* rem(c, p[x], x) (c: numeric, p[x]: polynomial) erroneously returned p[x]
  instead of c.
* Small bugfixes in pattern matching.
* Updated libtool to version 1.4.

0.9.0 (7 June 2001)
* In the output and in ginsh, lists are now delimited by { } braces, and
  matrices are delimited by single [ ] brackets.
* simplify_indexed() renames dummy indices so, e.g., "a.i*a.i+a.j*a.j" gets
  simplified to "2*a.i*a.i" (or "2*a.j*a.j").
* New functions/methods:
   - canonicalize_clifford() (helpful when comparing expressions containing
     Dirac matrices)
   - symmetrize() and antisymmetrize()
   - numer_denom() (return numerator and denominator in one call)
   - map() (apply function to subexpressions)
   - evalm() (evaluate sums, products and integer powers of matrices)
* Added a new function match() for performing pattern matching. subs() and
  has() also accept patterns as arguments. A pattern can be any expression,
  optionally containing wildcard objects. These are constructed with the
  call "wild(<unsigned>)" and are denoted as "$0", "$1" etc. in the output
  and in ginsh.
* Positive integer powers of non-commutative expressions (except matrices)
  are automatically expanded.
* Removed cint subdirectory, ginaccint is a separate package now due to 
  packaging considerations.
* Several little bugfixes.

0.8.3 (11 May 2001)
* color and clifford classes are functional and documented.
* New "spinidx" class for dotted/undotted indices.
* Predefined spinor metric tensor (created by spinor_metric()).
* Symbols can have a LaTeX name, e.g. symbol s("s", "\\sigma");
* LaTeX output of indexed objects is much nicer.
* Fixed some build problems (with recent libreadline).
* Semantics of arithmetic operators now follows the C++ rules more strictly.

0.8.2 (24 April 2001)
* degree(), ldegree(), coeff(), lcoeff(), tcoeff() and collect() work with
  non-symbols as the second argument in ginsh.
* the argument to collect() can be a list of objects in which case the
  result is either a recursively collected polynomial, or a polynomial in
  a distributed form with terms like coeff*x1^e1*...*xn^en, as specified by
  the second argument to collect().
* Several bugfixes (including a nasty memory leak in .normal()).
* class matrix: solve() doesn't call algorithms redundantly any more and
  inverse() falls back to solve() which works in more general cases.

0.8.1 (16 April 2001)
* degree(), ldegree(), coeff(), lcoeff(), tcoeff() and collect() can now
  be used with constants, functions and indexed expressions as well, so you
  can use it to collect by powers of Pi or sin(x), or to find the coefficient
  of gamma~0.
  Limitations:
   - it only works with symbols, constants, functions and indexed expressions,
     trying to find the coefficient of, e.g., "x^2" or "x+y" won't work;
   - it does not know about dummy index summations; the coefficient of
     gamma~0 in p.mu*gamma~mu should be p.0 but is returned as 0;
   - using coeff(), tcoeff(), lcoeff() or collect() on elements of
     noncommutative products might return wrong or surprising results.
* subs() no longer only substitutes symbols and indices but performs a more
  general "syntactic substitution", i.e. it substitutes whole objects in sub-
  expressions. You can subs((a+b)^2,a+b==3) and get 9, but subs(a+b+c,a+b==3)
  doesn't do anything.
  Limitations:
   - substituting numerics (subs(expr, 2==4)) will not replace then in all
     occurences; in general, you shouldn't substitute numerics, though.
* Added preliminary (re)implementations of color and clifford classes.
* simplify_indexed(): contraction of symmetric and antisymmetric tensors
  is zero.
* Replaced the various print*() member functions by a single print() that
  takes a print_context object that determines the output formatting. This
  should make it easier to add more output types, such as LaTeX output,
  which is based on work by Stefan Weinzierl.
* Added functions to retrieve the properties stored in archive objects
  outside of unarchive() (for printing or debugging purposes).
* Some bugfixes (indexed objects, archive writing).
* .collect() on non-polynomials is now algebraically correct.

0.8.0 (24 March 2001)
* Complete revamp of indexed objects. Instead of multiple classes for
  indexed things and their indices there is now only one "indexed" class
  and two types of indices: "idx" for simple indices and "varidx" for
  indices with variance. There are predefined delta, epsilon and metric
  tensors, and a function simplify_indexed() that performs canonicalization
  and dummy index summations. Matrix objects can be indexed for doing simple
  linear algebra.
* Added an option "expand_indexed" to expand() to perform expansion of
  indexed objects like (a+b).i -> a.i + b.i
* Renamed get_indices() to get_free_indices(), which no longer returns
  dummy indices and checks the consistency of indices in sums.
* sqrfree() factorization fixed and improved syntactically.
* subs() works on matrices.
* Matrices can be constructed from flat list of elements; diagonal matrices
  can be constructed from list of diagonal elements with diag_matrix().
* Fixed memory leak in expand().
* Operator% for objects of class ncmul has gone.  Use operator* now for that
  case too, which is much more natural.

0.7.3 (28 February 2001)
* Several bugfixes and minor performance tunings.
* Added a section to the tutorial about adding new algebraic classes to GiNaC.
* Closed many in-source documentation gaps.

0.7.2 (17 February 2001)
* Several bugfixes in power series expansion, one of them critical.

0.7.1 (7 February 2001)
* Fix problems with Cint that were caused by CLN's overloaded operator new.
* Fix compilation errors with GCC3.
* normal() handles large sums of fractions better and normalizes the exponent
  of power expressions.
* expand() always expands the exponent and transforms x^(a+b) -> x^a*x^b.
* Some bugfixes of series expansion around branch cuts of special functions.

0.7.0 (15 December 2000)
* Requires CLN 1.1 now.  Class numeric doesn't use an indirect pointer to the
  actual representation any more.  This is a speedup.
* mul::expand() was reengineered to not allocate excess temporary memory.
* Non-integer powers of a symbol are treated as constants by (l)degree() and
  coeff().  Using these functions on an expression containing such powers used
  to fail with an internal error message.  The side-effect is that collect()
  can be used on expressions which are not polynomials.
* Added a man page for the ginac-config script.
* Ctor of numeric from char* honors Digits.

0.6.4 (10 August 2000)
* Complete revamp of methods in class matrix.  Some redundant (and poor)
  implementations of elimination schemes were thrown out.  The code is now
  highly orthogonal, more flexible and much more efficient.
* Some long standing and quite nasty bugs were discovered and fixed in the
  following functions: add::normal(), heur_gcd(), sr_gcd() and Order_eval().

0.6.3 (25 July 2000)
* Derivatives are now assembled in a slightly different manner (i.e. they
  might 'look' different on first sight).  Under certain circumstances this
  can result in a dramatic speedup because it gives hashing a better chance,
  especially when computing higher derivatives.
* Some series expansions of built-in functions have been reengineered.
* The algorithm for computing determinants can be chosen by the user.  See
  ginac/flags.h and ginac/matrix.h.
* The Dilogarithm (Li2) now has floating point evaluation, derivative and a
  proper series expansion.
* Namespace 'std' cleanly disentangled, as demanded by ISO/EIC 14882-1998(E).
* Some minor bugfixes, one major lsolve()-bugfix and documentation updates.

0.6.2 (21 June 2000)
* ginaccint.bin is now launched by a binary program instead of by a scripts.
  This allows us to write #!-scripts.  A small test suite for GiNaC-cint was
  added.
* Several minor bugfixes.

0.6.1 (18 May 2000)
* Cleanup in the interface to Cint.  The required version is now Cint 5.14.38.
* Several bugfixes in target install.

0.6.0 (11 May 2000)
* IMPORTANT: Several interface changes make programs written with GiNaC 
  much clearer but break compatibility with older versions:
  - f(x).series(x,p[,o]) -> f(x).series(x==p,o)
  - series(f(x),x,p[,o]) -> series(f(x),x==p,o)
  - gamma() -> tgamma()  (The true Gamma function, there is now also
    log(tgamma()), called lgamma(), in accord with ISO/IEC 9899:1999.)
  - EulerGamma -> Euler
* #include'ing ginac.h defines the preprocessor symbols GINACLIB_MAJOR_VERSION,
  GINACLIB_MINOR_VERSION, and GINACLIB_MICRO_VERSION with the respective GiNaC
  library version numbers.
* Expressions can be constructed from strings like this:
    ex e("2*x+y", lst(x, y));
* ex::to_rational() provides a way to extend the domain of functions like
  gcd() and divide() that only work on polynomials or rational functions (the
  good old ex::subs() method reverses this process)
* Calling diff() on a function that has no derivative defined returns the
  inert derivative function "Derivative".
* Several new timings in the check target.  Some of them may be rather rude 
  at your machine, feel free to interrupt them.

0.5.4 (15 March 2000)
* Some algorithms in class matrix (notably determinant) were replaced by
  less brain-dead ones and should now have much better performance.
* Checks were completely reorganized and split up into three parts:
  a) exams (small regression tests with predefined input)
  b) checks (lenghty coherence checks with random input)
  c) timings (for coherence and crude benchmarking)
* Behaviour of .evalf() was changed: it doesn't .evalf() any exponents.
* Expanded expressions now remember they are expanded to prevent
  superfluous expansions.
* Small bugfixes and improvements in the series expansion.

0.5.3 (23 February 2000)
* A more flexible scheme for registering functions was implemented,
  allowing for remembering, too.
* Some Bugfixes.

0.5.2 (16 February 2000)
* Mainly fixes a bug in the packaging of release 0.5.1.

0.5.1 (14 February 2000)
* Fixes a small number of bugs.

0.5.0 (7 February 2000)
* Expressions can be written ("archived") to files and read therefrom.
* Addition of GiNaC-cint, which lets you write complete programs in
  an interactive shell-like manner in your favoured programming
  language (i.e. C++).

0.4.1 (13 December 1999)
* Series Expansion of Gamma function and some other trigonometric
  functions at their poles works now.
* Many more evaluations of special functions at points where
  exact results exist.
* info_flags::rational doesn't return true for complex extensions
  any more---use info_flags::crational for the old behaviour.
  info_flags::integer and -::cinteger work similarly, the same
  holds for types like info_flags::rational_polynomial.

0.4.0 (26 November 1999)
* First public release.