changelog

A C++ class library for scientific computing

	local copy of the iterator before it was incremented.  Some
	compilers such as the MS VS.NET 2003 C++ compiler would crash if
	you tried to use the const reference that was returned previously.
2005-05-18  Julian Cummings  <[e-mail]>

	* blitz/mathfunc.h: Replaced broken test "#ifdef isnan" with a
	test for the preprocessor symbol defined by the new
	AC_CXX_ISNAN_IN_NAMESPACE_STD autoconf macro.  If the isnan
	function is provided in namespace std, we invoke it as std::isnan
	in order to avoid problems with using the C99 isnan macro from
	within the blitz namespace.
	* blitz/funcs.h: Replaced broken test "#ifdef isnan" with a test
	for the preprocessor symbol defined by the new
	AC_CXX_ISNAN_IN_NAMESPACE_STD autoconf macro.  If the isnan
	function is provided in namespace std, we invoke it as std::isnan
	in order to avoid problems with using the C99 isnan macro from
	within the blitz namespace.
	* blitz/blitz.h: Only include <math.h> header explicitly if the
	test for standard math functions in the std namespace failed.
	Otherwise, it will normally be included implicitly by the <cmath>
	header.
	* m4/ac_cxx_math_fn_in_namespace_std.m4: Added an AC_REQUIRE of
	AC_CXX_NAMESPACES here, since C++ namespaces are a prerequisite
	for this test.

	* m4/ac_cxx_standard_library.m4: Added invocation of new macro
	AC_CXX_ISNAN_IN_NAMESPACE_STD that checks if the isnan function is
	provided in the std namespace when you include the <cmath> header.
	If it is, we will use std::isnan when calling this function from
	within the blitz namespace to avoid problems with the C99 isnan
	macro.
2005-05-13  Julian Cummings  <[e-mail]>

	* testsuite/promote.cpp: Used cpp directives to skip some code
	that will not work under the old type promotion system.

	* m4/ac_cxx_flags_preset.m4: Just discovered that the Intel C++
	compiler is unhappy with the change from unnamed enums to static
	const bool.  It gives an error on one particular initialization in
	<blitz/promote.h>.  I have reported this to Intel as a bug in the
	compiler, since other compilers seem to accept it.  As a temporary
	solution, I am adding -DBZ_DISABLE_NEW_PROMOTE to the preset flags
	for icpc.  This will force the usage of the older code in
	promote-old.h.  This is just to allow the library and testsuite
	codes to compile under icpc again.  I will undo this change as
	soon as we determine what's going wrong here.

	* blitz/promote.h: Some additional cleanup after migrating from
	anonymous enum types to static const types.  Changed promoteToT1
	from int to bool and simplified expression for promoteToT1.
2005-05-06  Julian Cummings  <[e-mail]>

	* blitz/array-impl.h: Converted most unnamed enums into static const 
	int or bool types where appropriate, in order to resolve compilation 
	problems under gcc-4.0.0.  This new compiler complains whenever it 
	sees an unnamed enum being used in an expression where there has been
	a templated version of the operator in the expression already declared.
	It tries to instantiate the templated operator with the unnamed enum,
	which is illegal and results in a compile error.  Very annoying!  In
	any case, I also removed the usages of BZ_ENUM_CAST that were no longer
	necessary, since we are now using real types.  This is probably a safer
	way to write the code anyway...
2005-04-20  Julian Cummings  <[e-mail]>

	* m4/ac_fortran_flags_preset.m4: Added the proper flag for Fortran
	external symbol translation when using the Intel Fortran compiler.
	Moved the AM_CONDITIONAL for F90_COMPILER into configure.ac so
	that it is always defined.
	* m4/ac_cxx_flags_preset.m4: Use -ansi flag with Intel compiler
	for now.  The -strict_ansi flag does not work with the newer gcc
	header files, but -ansi is better than nothing.  Also fixed the
	profiling flag for the Intel compiler.

	* configure.ac: Added option --disable-fortran that allows user to
	skip the Fortran compiler configuration.  With this option, the
	banchmarks directory will not be configured.  The default is
	--enable-fortran.  I also moved the C++ compiler configuration
	ahead of the Fortran configuration, in case Fortran is skipped.
2005-04-14  Julian Cummings  <[e-mail]>

	* m4/ac_cxx_flags_preset.m4: A few modidfications to the GNU and
	IBM XL C++ compiler optimization flags.

	* m4/ac_fortran_flags_preset.m4: Added cases to specifically
	handle the darwin build target.  Added cases for F77=g95, which
	can be useful on Darwin platform.  Generalized all Fortran
	compiler names.  A few modifications to the GNU and IBM XL Fortran
	compiler optimization options.

	* configure.ac: Added g95 compiler to the Fortran 77 compiler
	search list.  This is primarily for the benefit of users running
	on the Darwin (Mac OS X) platform.  These systems often do not
	come with the g77 compiler installed, and it can be difficult to
	install g77 and g++ at the same version level.  Mixing different
	versions of g77 and g++ can lead to link-time errors in the blitz
	benchmark codes.  Using g95 as both the Fortran 77 and Fortran 90
	compiler offers another solution.  Note that use of the g95
	compiler under Mac OS X requires cctools-528.  Info on how to
	install this is at www.g95.org.
2005-04-06  Julian Cummings  <[e-mail]>

	* m4/ac_cxx_flags_preset.m4: Added case for solaris target with CC
	compiler (SunPRO).  The preset flags have been verified to work
	with Sun Studio 10, C++ version 5.7.  Also added handling of c++
	compiler name as pseudonym for g++ and reordered the cases
	somewhat.  I am also removing the -strict_ansi flag with the Intel
	C++ compiler for now, since this has been reported to cause
	trouble with newer versions of icc and gcc.
	* m4/ac_compiler_specific_header.m4: Added case for SunPRO CC
	compiler (CC on solaris target) and handling of c++ compiler name.
	Generalized all compiler names.
	* configure.ac: Inserted c++ into C++ compiler search list, so
	that Darwin systems (which are case-insensitive) will pick up c++
	(a symbolic link to g++) rather than CC (which is really cc, a 
	symbolic link to gcc).  We need c++ in front of CC in the search list.
	* blitz/bzconfig.h: Added case to check for SunPRO CC compiler and
	defer to sun/bzconfig.h.

2005-04-01  Julian Cummings  <[e-mail]>

	* testsuite/reduce.cpp: Added test of count() function.
	* blitz/array/expr.h: Use typedef T_expr consistently inside
	ArrayExpr class definition.

2005-02-23  Patrick Guio  <[e-mail]>

	* blitz/numinquire.h: Added config header file <blitz/blitz.h> so 
	that numerical inquiries stuff can be used by just including
	<blitz/numinquire.h>.


2005-02-08  Julian Cummings  <[e-mail]>

	* configure.ac: Added g95 to the list of FC compilers that are
	tried and reordered the list to prefer g95 over gfortran.  Also,
	reordered the compiler macros to do our compiler flag presets
	first, before any other macros that may add more flags.
	* m4/ac_fortran_flags_preset.m4: On many Linux systems, f77 points
	to the g77 compiler.  So I've added a test for this possibility,
	instead of just assuming that f77 on Linux systems is the Absoft
	compiler.  If $G77 equals yes, use the g77 flag presets.  Added
	flag presets for the PGI F77 and F90 compilers, and for the g95
	compiler.  Reverted to the practice of overwriting the default
	values of FFLAGS and FCFLAGS because the default values are not
	appropriate for benchmark codes.
	* doc/stencils/Makefile.am: Removed explicit -lm flag from LDADD.
	This is now inserted into the LIBS variable automatically if
	needed for standard math functions like sin().
	* benchmarks/Makefile.am: Simplified settings for F90 compilation,
	taking advantage of the improved F90 support in automake 1.9 using
	the FC variables.
	* m4/ac_lib_blas.m4: Replaced FORTLIBS with standard autoconf
	variable FLIBS.
	* m4/ac_env.m4: Print out settings of FLIBS and FCLIBS rather than
	FORTLIBS, which is now obsolete.
	* m4/ac_fortran_flags_preset.m4: Removed settings for FORTLIBS
	variable.  Fortran library loader flags are now handled using
	standard autoconf macros.  Separated settings for F77 and FC
	variables and only set FC variables if FC has been set.  Ensure
	that we add to FFLAGS and FCFLAGS rather than overwriting them.
	Added gfortran case for F77 and FC on Linux platform.  Added g77
	case for all Unix platforms.
	* configure.ac: Advanced required automake version number to 1.9,
	which has improved support for F90 compilers via the FC variables.
	Added gfortran to the list of supported F77 and F90 compilers.
	Use standard autoconf macros to determine the Fortran library
	flags to pass to the loader, rather than our ad-hoc variable
	FORTLIBS.  Simplified test for whether FC has been set.  Improved
	test for standard math functions like sin() so that -lm is only
	explicitly included on the link line if necessary.  Removed test
	for existence of lib directory, since this is now guaranteed.
2005-02-07  Julian Cummings  <[e-mail]>

	* m4/ac_compilers_64bits.m4: Generalized names of a few other
	compilers, notably the SGI and IBM compilers.

	* m4/ac_compiler_specific_header.m4: Generalized names of some C++
	compilers.

	* m4/ac_compilers_64bits.m4: Generalized name of g++ compiler in
	case statements.

	* m4/ac_cxx_flags_preset.m4: Generalized name of g++ compiler in
	case statements.

2005-01-28  Julian Cummings  <[e-mail]>

	* doc/stencils/Makefile.am: Modified compilation rules to provide
	separate -D flag for defining stencilargs symbol.  Breaking this
	into two separate preprocessor symbols eliminates confusion that
	some compilers were having with parentheses embedded in a -D flag.
	Should fix problems with pgCC and xlC.
	* doc/stencils/dump-stencil.cpp: Introduce stencilargs symbol to
	represent argument(s) of stenciltoapply.

	* blitz/mathfunc.h: Regenerated using genmathfunc.cpp program to
	include PGI compiler hacks.

	* blitz/generate/genmathfunc.cpp: Added some lines to skip
	pow(float,float) and all transcendental functions acting on
	complex<long double> if we are using PGI compiler.  PGI does not
	support complex<long double> math functions because of issues with
	how some Linux systems define the long double type.  The
	pow(float,float) function fails because of the presence of
	overloaded versions of pow() taking the combination of a float and
	a complex<float>.  PGI does not provide an explicit
	pow(float,float) overload to call powf().
2005-01-06  Julian Cummings  <[e-mail]>

	* benchmarks/stencilf2.f: Added explicit declaration for local
	loop counter variable.

	* benchmarks/stencilf.f: Added explicit declaration for local loop
	counter variables.

	* Makefile.am: Removed unneeded compile rules for suffix ".f90.f"
	which is not used.


2005-01-07  Patrick Guio  <[e-mail]>

	* blitz/array/ops.h: Completed ET(array,scalar) for scalar 
	of all integral types.

2004-11-17  Patrick Guio  <[e-mail]>

	* doc/doxygen/Makefile.am: Fixed install related targets to take into
	account a possible DESTDIR.


2004-11-04  Julian Cummings  <[e-mail]>

        Tagging repository contents with Blitz_0_8 tag.
2004-11-04  Patrick Guio  <[e-mail]>

	* m4/ac_prog_doxygen.m4: Added test on doxygen version. In order 
	to be used doxygen must be at least 1.3.x.


2004-11-03  Patrick Guio  <[e-mail]>

	* m4/ac_prog_doxygen.m4 doc/doxygen/Makefile.am: Renamed automake
	conditional tag 'DOC' into more explicit 'DOXYGEN_DOC' as well as
	references.

	* doc/doxygen/Doxyfile.in: Upgraded config file to doxygen 1.3.6.
	Replaced ../.. by autoconf variables in INPUT tag. 
	Removed *.cc from the FILE_PATTERNS tag since they are just
	implementation. Limited the class to document by explicitly exclude 
	concerned files using the EXCLUDE tag. Fine tuned some doxygen and
	dot behaviour tags.

	* doc/doxygen/Makefile.am: Set docdir to be 
	$(datadir)/doc/$(PACKAGE)-$(VERSION)/doxygen, indeed a 
	subdirectory of the blitz root document directory. 
	Use more systematically available variables.
	Cosmetic Changes.

	* TODO: new item: make available RPM as a dist medium.


2004-10-29  Julian Cummings  <[e-mail]>

	* doc/doxygen/Makefile.am: Set docdir to
	$(datadir)/doc/$(PACKAGE)-doc-$(VERSION) for the doxygen generated
	documentation, to distinguish from the texinfo generated docs.
	Removed $(VERSION) from names of PS and PDF manuals.  There is
	still something wrong with the ps and pdf targets.  How do you get
	doxygen to generate the manuals?  Patrick, please fix!
	* doc/Makefile.am: Changed docdir to standard value
	$(datadir)/doc/$(PACKAGE)-$(VERSION).

	* doc/doxygen/Doxyfile.in: Changed FILE_PATTERNS from *.cpp to
	*.cc, since that is the extension used for blitz template
	implementation header files.