that all HDF5 API functions are protected by a mutex lock. Basically,
serialized all API calls. To use it, use
configure --enable-threadsafe --with-pthread
** src/H5Tconv.c
** src/H5Tpkg.h
** src/H5Tpublic.h
The H5T_conv_struct_opt() function had a design flaw -- it
didn't keep information about the stride to use to step
through the temporary/background-value buffer and thus nested
invocations would clobber each other's temp buffers. This was
fixed by splitting the `stride' argument into `buf_stride' and
`bkg_stride' arguments for all the conversion functions. THIS
IS AN API CHANGE, but users will get a compiler warning when
they pass their conversion function pointer to H5Tregister().
** src/H5T.c
** src/H5Tprivate.h
Added a bkg_stride argument to the H5T_convert() definition in
order to fix a bug related to the optimized compound datatype
conversion function.
** src/H5T.c
** src/H5A.c
** src/H5D.c
** src/H5Ofill.c
** src/H5P.c
Added bkg_stride=0 argument to the H5T_convert() calls.
** test/dtypes.c
Added a test for the H5T_conv_struct_opt() bug fixed above.
** src/H5FL.c
The H5FL_term() function should return non-zero even when it
couldn't free all the free lists do to their being used by
some other package. When that other package terminates it
will return non-zero, causing H5FL_term() to be called
again. This fixes some of the `infinite loop closing library'
messages.
** tools/pdb2hdf
Uses print_version() instead of doing that itself.
** src/H5Ppublic.h
Renamed H5Pget_gc_reference() declaration to make it match the
definition.
** src/H5FDlog.c
Added API tracing macros.
Removed `const' qualifier from a `char*' member of a struct
which was allocated on the heap.
** src/H5TB.c
Added curly braces to a couple deeply-nested `if' statements
to make them clearer and to shut up the increadibly stupid and
just plain incorrect gcc warning about ambiguous `else'.
** test/titerate.c
Removed incomplete initialization in favor of memset() for one
auto variable to stop compiler warnings.
** tools/Depencencies
Regenerated to remove references to h5dumputil.c
was also a subtle bug in the previous code. A valid size of 16K would
have been rejected since it was checking for an invalid number. The
scheme that replaces it relies on the fact that powers of two and only
powers of two have 1 bit set in binary.
has turned off error reporting in the library.
Also, changed the way property lists are managed and closed so that they
can be automatically closed when the library terminates.
to double-check things. I've turned them back on again now. I also changed
the internal representation of a few struct fields to be float instead of
double, since the HP/UX 10.20 compiler was having problems with the alignment
of the doubles.
free-lists for the library. To turn off the free-lists, edit H5FL.c and
uncomment the "NO_FREE_LISTS" macro definition near the top of the file.
(Or define the macro during compile time)
Bug fix for parallel mode.
Description:
H5FD_mpio_tas_allsame was called for all cases, even when
MPIO is not used for access. That corrupted the internal
file handle structure.
Solution:
Define a macro, IS_H5FD_MPIO(f), for testing if f is opened with
MPIO access. Will call H5FD_mpio_tas_allsame only if this
condition is true.
Platform tested:
O2K, both -64 and -n32 modes.
Bug Fix.
Description:
Tests failed if drivers multi or split is used (split is implemented
as a special case of multi). When a file created via those two drivers
is opened again, the file sizes retrieved are incorrect. These seemed
to happen to non-little-endian machines only (sun, sgi failed, but
linux succeed.)
Solution:
The error was because the superblock encoding code was keeping track
in nseen the pairs of HADDR needed to be converted but did not tell the
H5Tconvert 2*nseen to convert. Thus only the first half of the addresses
were converted, the rest were left alone. Corrected the numbers of
addresses to convert.
Platforms tested:
IRIX64, Solaris 2.7, Linux.
