Bug Fix
Description:
uint64_t type was accidentally taken out and was causing tests on DEC
UNIX to fail.
Solution:
Brought back uint64_t type.
Also cleaned up some comments and moved the sec2 specific macros back
into the sec2 (and log) driver.
Platforms tested:
DEC UNIX (gondolin)
Bug fix/code improvement.
Description:
'all' selections were (ab)using the array reading code and required that
the internal data transfer buffer size be big enough to hold the an entire
slab of the data, which was confusing and limiting for users.
Solution:
Changed 'all' selections to use sequence reading code instead of array
reading code.
Platforms tested:
FreeBSD 4.3 (hawkwind)
Code cleanup
Description:
float32 & float64 typedefs are not used in HDF5 library and are causing
conflicts with Kent's h4toh5 library code.
Solution:
Take the float32 & float64 typedefs out
Platforms tested:
FreeBSD 4.3 (hawkwind)
Bug fix for Pablo integration
Description:
Several API functions were using FUNC_ENTER, without using FUNC_LEAVE,
HRETURN or HRETURN_ERROR to leave the function. (Using "plain" 'return'
statements instead).
Solution:
Changed return statements to FUNC_LEAVE, HRETURN or HRETURN_ERROR, as
appropriate.
Platforms tested:
FreeBSD 4.3 (hawkwind)
Code cleanups
Description:
Fixed a small number of warnings and also took out the H5private.h
header from H5Ppublic.h
Platforms tested:
FreeBSD 4.3 (hawkwind)
Small code addition
Description:
Addition logging of seeks during read operations (writes were already
logging seeks). Also added a few asserts to check memory allocations...
Platforms tested:
Solaris 2.6 (baldric)
Bug fix
Description:
This file is not C++ friendly/compliant because the protocols
are not bracketed by the #ifdef __cplusplus macro. This was
discovered by a user attempting to use C++ with enable-parallel.
Solution:
Though we are not supporting C++ in parallel mode yet, it is
simple to add the macro bracket. It would help if C++ is supported
in the future.
Platforms tested:
modi4 (64,n32) and eirene (mpich).
Code Simplification
Description:
H5Pget_class() was being called from the H5.c module. Changed it so
that it calls H5P_get_class() instead. Also, the H5Pget_class() and
H5P_get_class() functions were identical. Changed so that
H5Pget_class() calls H5P_get_class().
Platforms tested:
Linux
Replace Kludge with Another Kludge
Description:
The MPI_Get_count function isn't found on most systems. We need to
replace this with a configure call which checks and defines the macro
H5_HAVE_MPI_GET_COUNT or something.
Solution:
If-def'ed out the code for everyone (since H5_HAVE_MPI_GET_COUNT is
never defined anywhere).
Platforms tested:
SDSC HP N9000
Cosmetic adjustment
Description:
Changed Version information from just "HDF5 version ..."
to "HDF5 library version" since this is really about a
library implementation.
Platforms tested:
eirene (used "h5vers -s ..." to see the effect.)
Implemented the string version number stuff for Albert.
2001-05-09 16:24:18 Robb Matzke <matzke@llnl.gov>
*: Added definition for H5_VERS_INFO, which is a string
representation of the hdf5 version number.
Better Mouse Trap^W^WBug Fix
Description:
Instead of having the large file support stuff be in the
config/gnu-flags directory, I added it to the configure.in file so
that it first checks for the syscall ``getdents64()''. If it finds
it, then it removes the ``-malign-double'' flag if the GCC compiler
is 2.95.x since this flag doesn't play well with the LFS stuff.
Solution:
Added test for getdents64(), if found it removes the -malign-double
flag if the compiler is 2.95.x. This is done in the configure.in
file. The gnu-flags file is returned to normal. A new field is added
to the H5config.h.in file because of the getdents64 check.
Platforms tested:
Linux
Document bug fix
Description:
IMPORTANT! IMPORTANT! IMPORTANT!
A case where metadata in a file could get corrupted in certain unusual
sitations was detected and fixed.
In certain circumstances, metadata could get cached in the raw data cache,
and if that particular piece of metadata was updated on disk while
incorrectly cached, the new metadata would get overwritten with the stale
metadata from the raw data cache when it was flushed out.
Additionally, I've patched up the raw data cache to be smarter about how
much it caches and how much I/O it triggers, leading to some speedups.
Solution:
Changed the raw data I/O routines which perform caching to require a
parameter with the size of the dataset being accessed and limited the
cache to no more than that many bytes.
Platforms tested:
FreeBSD 4.3 (hawkwind)
Feature
Description:
The allocation by alignment (H5Pset_alignment) feature code somehow
got dropped in some 1.3.x version.
Solution:
Re-implemented it with "new and improved" algorithm. It keeps track
of "wasted" file-fragment in the free-list too.
Platforms tested:
modi4(parallel).
Update
Description:
On further discussion with Albert and Quncey, it was decided to keep
the original "unsignedness" of the nkeys variable.
Solution:
Changed the type from intn to uintn.
Platforms tested:
Linux
Bug Fix
Description:
The --enable-static-exec flag was failing on Linux systems. The
reason: a "long long" type was growing by some obscene amount if you
just add 1 to it (tis truth!). In one statement, it prints out the
correct value. Add one to it and print it out in the next statement
and it gives you a very large number. This would, in turn, cause the
resulting memcpy to fail...
Solution:
Changed the variable from "hsize_t" to "intn" which is what the field
it's getting is typed anyway.
Platforms tested:
Linux
Bug Fix.
Description:
Setting a non-zero userblock size was causing raw data caching code to
break.
Solution:
Changed from using absolute end-of-address-space offsets in cache size
calculations to relative offsets.
Platforms tested:
FreeBSD 4.3 (hawkwind)
Purpose:
a bug fix
Description:
forget considering daylight factor for mtime test on windows platform
Solution:
adding daylight factor in this file
Platforms tested:
windows 2000, confirmed at eirene
Update
Description:
Changed
#include <hdf_file.h>
construct to
#include "hdf_file.h"
so that the GNU compiler can more easily pick up the dependencies
which it places in the .depend and Dependencies files. Also
regenerated the Dependencies to go along with this.
Platforms tested:
Linux
Code cleanup
Description:
Defined snprintf macro for non-UNIX platform which needed tweak.
Solution:
Added extra ifdef around snprintf macro to accomodate platform.
Platforms tested:
FreeBSD 4.3 (hawkwind)
Code cleanup
Description:
ssize_t is not defined on certain (non-UNIX) platforms.
Solution:
Added steps to configure to generate SIZEOF_SSIZE_T macro and then used
the new macro to define ssize_t for platforms which don't have it.
Platforms tested:
FreeBSD 4.3 (hawkwind)
Improvment
Description:
H5config.h and H5pubconf.h are generated by configure. It does
not have the #ifndef _H5config_H ... #endif guard to prevent
duplicated definitions if it is included more than once.
It is messy to try make configure to put in those guards.
Solution:
HDF5 has set an internal rule that H5public.h includes H5pubconf.h
and H5private.h includes H5config.h. Source files should NOT include
H5config.h or H5pubconf.h directly but include it via H5public.h or
H5private.h respectively. The #ifndef ... #endif in the H5public.h
and H5private.h would prevent repeated definitions from repeated
include.
Adjusted H5FDstream.c and H5FDstream.h to follow this rule.
Platforms tested:
modi4, eirene.
Bug fix (sort of)
Description:
The RCSID string in H5public.h was causing the C++ code problem as it
was included multiple times and C++ did not like multiple definitions
of the same static variable.
Solution:
Since we don't really make use of the RCSID strings as we have not
installed it in all source files, we decided to remove it.
Platforms tested:
eirene (linux), modi4 (IRIX64-64) both serial and parallel modes.
Bug Fix
Description:
Partial hyperslabs which exactly fit the size of the lowest dimension of
a chunked dataset weren't being output correctly.
Solution:
Forgot to divide the offset (seq_len) by the size of the lower dimension
hyperslabs - fixed now.
Platforms tested:
FreeBSD 4.2 (hawkwind)
Bug Fix
Description:
For some reason, the H5FDstream.h file was defined not once, not
twice, but three whole times in the install part...
Solution:
Removed all but one of those.
C++ compatibility fix
Description:
'delete' is a reserved word in C++
Solution:
Change structure member named 'delete' into 'del'
Platforms tested:
FreeBSD 4.2 (hawkwind)
Bug fix.
Description:
Fencepost error in determining number of elements to include in a
partial hyperslab for a chunk.
Solution:
Changed "<down_size[i]"s into "<=down_size[i]"s in a few places, to allow
for exactly one slab being left to output/input.
Platforms tested:
FreeBSD 4.2 (hawkwind)
Code checkpoint
Description:
Checkpointing generic property code before removing the older [non-generic]
property list implementation and switching all the property lists to use
the generic code.
Generic properties are feature complete and as fully tested as I can
determine.
Platforms tested:
FreeBSD 4.2 (hawkwind)
Bug Fixes
Description:
Fixed a bug in H5Shyper.c where 'contiguous' hyperslabs (i.e. ones which
took up an entire dataset) were not being detected correctly and would
instead be read a part at a time instead of all at once.
Also fixed a bug in the handling of hyperslabs for chunked datasets where
hyperslabs from chunks which weren't aligned on exact dimension bounaries
were not reading/writing data correctly.
Solution:
H5Shyper.c was a single line change from a 'block' size to a 'count' size.
H5Fseq.c changes we much more significant and involved detecting when
non-chunk aligned sequences of data were being written and constructing
hyperslab blocks to pass down to the chunking I/O routine (which only
understand hyperslab I/O requests, not element sequence requests).
This was complicated by the need to align the hyperslabs requested on
dimension boundaries...
Platforms tested:
FreeBSD 4.2. (hawkwind)
