Reversed it to use gcc/g++ as the default C and CXX compilers for Darwin 10.X
systems.
Tested: duck (darwin 11), wren (darwin 12) and fred (darwin 10).
Use Clang as the compiler for the Mac platform.
Changed all Mac to use clang/clang++ as the default C and CXX compilers if
not provided by the user. Still use gfortran as the default fortran
compiler since Apple does not provide any fortran compiler.
Tested: passed all three macs OS, snow leapord (fred), lion (duck) and
mountain lion (wren). Also passed test using Apple gcc, Gnu gcc, Intel
as the C compilers and all corresponding CXX compilers fits in.
Did not run h5committest or other non-mac platform since this change
affect Apple platforms only.
Compile scripts will add paths for external libraries linked to executable files that they create.
Eliminates need for LD_LIBRARY_PATH for szip in locations not known to ld.
Teted on platypus and emu (64bit) in addition to h5committest.
Changes default value for USE_SHARED_LIB in compile scripts to yes when static is disabled.
Comments in scripts about the order of flag variables containing library paths was also corrected.
Tested with h5committest on duck, jam, koala, and ostrich in addition to specific testing for the
compile script change.
Description:
+ The C++ test failed with the new PGI compilers versions 12.4 and 12.5
+ An exception thrown by an internal function, which was called by
a constructor, was not propagating to the test program during the stack
unwinding, so it couldn't be caught by the test and the program terminated.
+ Various trials and errors indicated that the problem is where an STD string
converted to a char* being passed to the internal function, but confirmation
has not been found yet. It could be a compiler bug.
Solution:
+ Added a try/catch in the constructor around the internal function and
re-throw the exception when it is caught. This is a workaround.
+ Unrelated minor fixes: removed unused variables and MESSAGE's; commented
out tvlstr.cpp/test_read_vl_string_attribute because it may be redundant,
and commented out H5Tpkg.h inclusion because TEST_ALIGNMENT is not added
yet and probably not necessary in the C++ API.
Platforms tested:
Linux/32 2.6 (jam) with PGI compilers
Linux/32 2.6 (jam) with GNU compilers
Linux/64 2.6 (koala)
during variable-length type conversion. All variable-length type conversion
buffers are now initialized to contain 0s on allocation.
Tested on: jam(w/ fortran, C++, parallel, parallel fortran)
ostrich
This is a very small change (basially changing malloc to calloc) and
is unlikely to cause issues on other platforms.
This turned out being an issue with configure. The reporter submitted a patch which fixed the fact that we should not be
setting AM_FCFLAGS (an automake variable) with FFLAGS (a user variable).
I removed this, and we now only set FFLAGS if the environment variable is set, otherwise we don't.
Tested: jam (gnu)
HDFFV-8214 - h5repack failed converting small chunked dataset (size < 1K) to contiguous layout.
Description:
h5repack failed converting small chunked dataset (size < 1K) to contiguous layout.
The first case was when chunk dim is bigger than the dataset dim (at leat one), h5repack failed with displaying error stacks.
The other case is when chunk dim is smaller than the dataset dim, h5repack failed to change layout.
Tested:
jam (linux32-LE), koala (linux64-LE), ostrich (linuxppc64-BE), emu (solaris-BE),fred (mac64-LE), Windows (32-LE cmake), cmake (jam)
