changes.
Last round of check in fixed the problem when an attribute was opened twice and data was
written with one of the handles, the file didn't have the data. It happened because each
handle had its own object structure, and the empty one overwrote the data with fill value.
This is fixed by making some attribute information like the data be shared in the
attribute structure.
Tested on smirom, kagiso, and linew.
Bring back r15367 from the trunk:
Add check to avoid mounting a file on a group twice, when the mounts
are done on the same HDF5 file, but opened with separate H5Fopen queries.
Also add new 'mounted' flag to the H5G_info_t struct, queried with the
H5Gget_info() API call, to allow applications to detect and avoid this
situation.
Tested on:
Mac OS X/32 10.5.4 (amazon)
Linux/64 10.5.4 (chicago)
Description:
This bug was previously fixed, but the fixed was modified slightly from the 1.8 branch and the trunk. The problem is that after we write a particular byte to manipulate the fill values, we need to write it back before reading and comparing to the expected value. The previous change makes the second write after the hyperslab is manipulated, so the initial hyperslab is still incorrect.
Tested:
VS2005 on WinXP (32- and 64-bit)
h5committest
the file didn't have the data. It happened because each handle had its own
object structure, and the empty one overwrote the data with fill value. This is
fixed by making some attribute information like the data be shared in the
attribute structure.
Tested on smirom, kagiso, and linew.
compound datatype. The size of compound type's member type wasn't updated. It's
fixed in this commit.
Tested on smirom. The same change was tested on three systems for v1.9.
Description:
On Windows, certain users were having trouble with the "ohdr" test, which does some processing on object header messages. The errors were hard to reproduce on our machines, and we eventually determined that the errors were timezone-specific.
The bug is triggered on Windows when processing timestamps very near the "Epoch" (midnight on 1/1/1970)-- the mktime() function does some automatic adjustment on the time to correct for timezones. In the USA, the correction adds a few hours; in Europe, it subtracts, thus giving us times pre-Epoch.
This only affects Windows because the Windows mktime() function cannot handle times before 1970-- other systems seemingly can.
The fix is to simply create timestamps only as early as 01/02/1970. This way, any timezone adjustment will still be post-Epoch.
This bug only affects the ohdr test, and shouldn't be a problem in the library. The earliest timestamps that will actually be read will be around the time HDF5 was created (~1996-7, per Quincey).
Tested:
(tested in trunk)
VS2005 on WinXP
h5committest (kagiso, linew, smirom)
Description:
As part of our Windows cleanup, we try to remove windows-specific tweaks in the source code. There are many instances where Windows code is introduces via ifdef's. We re-evaluate whether they are still required, and found that many of them are not. Others we change to "feature"-specific code, rather than Windows-specific.
Tested:
VS2005 on WinXP
VS.NET on WinXP
h5committest (kagisopp, smirom, linew)
Description:
On Windows, the pthread_self function cannot be used to print the returned thread ID for debugging. Instead, we need a separate function, GetCurrentThreadId. To eliminate some Windows ifdef's in the code, we create two new function macros which can be used by all platforms. It is conditionally defined in H5win32defs.h, and globally in H5private.h.
Tested:
VS2005 w/ pthreads on WinXP
kagiso w/ pthreads
Description:
In library code, we try not to use system calls directly, but instead use the HD{function} macro instead. This way, we can map special versions of the call on particular systems. Previously, it was all done in H5private.h. However, in an effort to clean up platform-specific definitions, we move all of the Windows macros into a separate file, win32defs.h. This way, we can use the non-Posix versions that Visual Studio sends warnings about.
Some macros are set specifically in the platform-specific header files. Then, any macros left unset will be set by the "default" implementation in H5private.h.
This checkin also cleans up various source files to use the HD* macros when possible.
Tested:
VS2005 on WinXP
VS.NET on WinXP
h5committest (kagiso, linew, smirom)
Bring back revision 15131 from trunk:
Finish omnibus chunked dataset I/O refactoring, to separate general
actions on chunked datasets from actions that are specific to using the v1
B-tree index.
Cleaned up a few bugs and added some additional tests also.
Tested on:
Mac OS X/32 10.5.3 (amazon)
Linux/32 2.4 (chicago)
Description:
On Linux-like systems, we can get the ID of the current thread through a pthread_self. However on Windows, the return cannot be cast as a threadID, so we simply couldn't get the ID. Previously we simply gave up and printed a message that we couldn't get an ID. Instead, though, we can use the Windows-specific call to GetCurrentThreadId(), which achieves the same goal. This way we can provide better debug output with threadsafe features.
Tested:
VS2005 on WinXP
VS.NET on WinXP
(other platforms not tested because change is within _WIN32 ifdef)
Description:
The fortran Makefile.am used HDF_FORTRAN to indicate it is part of the
Fortran API source so that conclude.am will give fortran api prefix in the
test output. The symbox HDF_FORTRAN is also used in configure for a different
purpose (indicated --enable-fortran). They conflicted.
Similar problem for the symbol HDF_CXX.
Solution:
Changed all the involved Makefile.am to use "FORTRAN_API" instead. It is
a more appropriate name. Same for CXX_API.
Along the way, discovered that the Makefile.am of hl/fortran/test and
hl/cxx/test did not have those symbols at all. Added them in.
Platform tested:
Kagiso serial and h5committested (kagisopp, smirom, linew).
Backport revision 15034 from trunk:
Correct another case of using 'size_t' for chunk sizes where 'uint32_t'
was necessary.
Tested on:
FreeBSD/64 6.2 (liberty) w/C++ & FORTRAN
Improvement.
Description:
src/libhdf5.settings was the initial configure summary and is installed.
Then configure is changed to dump a summary of the configure settings to
the output and also append it to src/libhdf5.settings. That created
two different output formats and duplicated information. This is the
initial attempt to clean up this confusion and unify the output format.
It is decided to use the src/libhdf5.settings template as the unified means.
This requires more macros symbols be defined. The following symbols are
all related to generating the src/libhdf5.settings file.
AC_SUBST(EXTERNAL_FILTERS)
AC_SUBST(MPE) MPE=no
AC_SUBST(STATIC_EXEC) STATIC_EXEC=no
AC_SUBST(HDF_FORTRAN) HDF_FORTRAN=no
AC_SUBST(FC) HDF_FORTRAN=no
AC_SUBST(HDF_CXX) HDF_CXX=no
AC_SUBST(CXX) HDF_CXX=no
AC_SUBST(HDF5_HL) HDF5_HL=yes
AC_SUBST(GPFS) GPFS=no
AC_SUBST(LINUX_LFS) LINUX_LFS=no
AC_SUBST(INSTRUMENT) INSTRUMENT=no
AC_SUBST(CODESTACK) CODESTACK=no
AC_SUBST(HAVE_DMALLOC) HAVE_DMALLOC=no
AC_SUBST(DIRECT_VFD) DIRECT_VFD=no
AC_SUBST(THREADSAFE) THREADSAFE=no
AC_SUBST(STATIC_SHARED)
AC_SUBST(enable_shared)
AC_SUBST(enable_static)
AC_SUBST(UNAME_INFO) UNAME_INFO=`uname -a`
The src/libhdf5.settings.in has CONDITIONAL's added to it too. The
untrue conditions turned into a "#" and these lines are cleaned by the
post processing script.
Platform tested:
h5committest on kagiso, smirom and linew. (smirom had failures not due to
my changes.)
Purpose: Configure Fix
Description: The test program h52gifgentst was getting installed in the bin
directory during 'make install', and it shouldn't. Make now builds
the program for use in testing but doesn't install it during
'make install'.
Tested: kagiso
Port revision 15015 back to 1.8 branch:
> Detect chunks that are >4GB before dataset gets created and return error
> to application.
>
> Tweak lots of internal variables that hold the chunk size/dimensions to
> use an 'uint32_t', instead of a 'size_t', so that the integer size is constant.
>
> Correct a number of our tests which were creating datasets with chunks
> that were >4GB and add some specific tests for >4GB chunk size detection.
>
> Minor whitespace & other code cleanups.
Tested on:
Mac OS X/32 10.5.2 (amazon)
Forthcoming testing on other platforms...
Bring revision 14860 back to 1.8 branch, change log for rev 14860 is:
Omnibus raw data I/O revisions, with wide-ranging changes and
refactoring, in order to prepare for implementing "fast append" feature.
These changes remove the majority of the code duplication for raw data
I/O which has crept in over the last ten years and introduces a more object-
oriented design for operating on different types of dataset storage.
Description:
Omnibus raw data I/O revisions, with wide-ranging changes and
refactoring, in order to prepare for implementing "fast append" feature.
These changes remove the majority of the code duplication for raw data
I/O which has crept in over the last ten years and introduces a more object-
oriented design for operating on different types of dataset storage.
Chunked storage no longer has it's own I/O routines, it is now handled
as either contiguous (if chunk is not pulled into the cache) or compact (if the
chunk is cached in memory).
No bug or feature changes, at least intentionally... :-)
Tested on:
Mac OS X/32 10.5.2 (amazon) w/production
Description: Updated bin/reconfigure script to reflect the new versions of
libtool and automake in the /home1/packages/ directory.
Rearranged configure.in script. For some reason, when using
libtool 2.2, the libtool script doesn't generate until
later in the configuration process, so I had to move
a test that parsed through the libtool script to a point
after where it was actually being generated.
Ran libtoolize on the project, and ran bin/reconfigure to
regenerate configure and Makefile.in's throughout.
Tested: kagiso, smirom, linew, tg-login
Fix bug which would incorrectly encode the member offsets for compound
datatypes whose size was between 256 & 511 bytes, when the "use the latest
format" feature was enabled.
Tested on:
Mac OS X/32 10.5.2 (amazon) w/debug
FreeBSD/32 6.2 (duty) w/production
Minor bug fix to H5Aget_num_attrs() to return error when an invalid
location ID is passed in.
Tested on:
Mac OS X/32 (amazon)
Too minor to require h5committest
callback functions modifies the skip list or LRU out from under a partial
or complete flush of the cache. This is a test for a situation that
should not occur, but we test anyway.
Also added code to skip longer tests in cache_api in express tests.
Tested serial and parallel on phoenix (debian --x86-32), and commit test.
Elena commit tested as well, and ran a manual test under MacOS X.
large entry, or to a large increases in the size of an existing entry.
This required some additions to the cache configuration structure, and
thus will require changes in the metadata cache documentation.
The basic idea is to monitor the size of entries as they are loaded,
inserted, or increased in size. If the size of the entry (or increase)
exceeds some user selected fraction of the size of the cache, increase
the size of the cache.
Note that this fix was designed quickly -- while it deals with the
use case that exposed the problem, we may have to revisit the issue
later.
Tested serial and parallel on Phoenix, and h5committest.
Correct the prototype for H5Sselect_elements() to take an 'hsize_t *' for
the coordinates, instead of 'hsize_t **'.
Tested on:
Mac OS X/32 10.5.1 (amazon)
Add work-around to allow reading files that were produced with a buggy
earlier version of the library, which could create objects with the wrong
object header message count. There is now a configure flag
"--enable-strict-format-checks" which triggers a failure on reading a file
with this sort of corruption (when enabled) and allows the object to be read
(when disabled). The default value for the "strict-format-checks" flag is
yes when the "debug" flag is enabled and no when the "debug" flag is disabled.
Note that if strict format checks are disabled (allowing objects with
this particular kind of corruption to be read) and the file is opened with
write access, the library will re-write the object header for the corrupt
object with the correct # of object header messages.
This closes bugzilla bug #1010.
Tested on:
FreeBSD/32 6.2 (duty) in debug mode
FreeBSD/64 6.2 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (kagiso) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64-amd64 2.6 (smirom) w/default API=1.6.x, w/C++ & FORTRAN,
in production mode
Linux/64-ia64 2.6 (cobalt) w/Intel compilers, w/C++ & FORTRAN,
in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, in production mode
Mac OS X/32 10.4.10 (amazon) in debug mode
Linux/64-ia64 2.4 (tg-login3) w/parallel, w/FORTRAN, in production mode
Tweak the constants for the shared message flags to be equal to the
other flags useds for the underlying messages.
Tested on:
FreeBSD/32 6.2 (duty) in debug mode
FreeBSD/64 6.2 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (kagiso) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64-amd64 2.6 (smirom) w/default API=1.6.x, w/C++ & FORTRAN,
in production mode
Linux/64-ia64 2.6 (cobalt) w/Intel compilers, w/C++ & FORTRAN,
in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, in production mode
Mac OS X/32 10.4.10 (amazon) in debug mode
Linux/64-ia64 2.4 (tg-login3) w/parallel, w/FORTRAN, in production mode
for it, too. This round of checkin isn't well tarnished yet. I'll come back to work on Jan. 8. I'll revise it then.
Tested it on kagiso, smirom, linew.
- Add hash value for skip list string types, to reduce # of string
comparisons.
- Fixed bug with metadata/small data block aggregator adding size == 0
block into file free space list.
- Refactored metadata/small data block aggregator code into single set of
common routines.
- Changed block aggregator code to be smarter about releasing space in the
'other' block when the 'other' block has aggregated enough data.
Tested on:
FreeBSD/32 6.2 (duty) in debug mode
FreeBSD/64 6.2 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (kagiso) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64-amd64 2.6 (smirom) w/default API=1.6.x, w/C++ & FORTRAN,
in production mode
Linux/64-ia64 2.6 (cobalt) w/Intel compilers, w/C++ & FORTRAN,
in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, in production mode
Mac OS X/32 10.4.10 (amazon) in debug mode
Linux/64-ia64 2.4 (tg-login3) w/parallel, w/FORTRAN, in production mode
Skip one of the tests in flush2.c since it looks like VMS system
flsuhes the file anyway (like on Winodws with DLLs)
Renamed files family*.h5 that are used in backward compatibility
testing to have only one '.' in a file name, otherwise VMS system
becomes confused and test fails.
Cleaned up the only test (Fortran multi-file test) that is failing on
VMS (C counterpart passesa).
Added new script to facilitate the build process on VMS
Updated MANIFEST
PLatforms tested: VMS server, kagiso and linew (minor changes)
- Avoid trying to update 1-D dataset's chunk indices (they can't change)
- Cache a copy of a dataspace describing a chunk, when doing single
element I/O
- Keep a 'chunk info' struct around, for single element I/O
- Avoid creating a skip list for chunk infos when performing I/O on single
element
Also, minor formatting cleanups to testing code
Tested on:
FreeBSD/32 6.2 (duty) in debug mode
FreeBSD/64 6.2 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (kagiso) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64-amd64 2.6 (smirom) w/default API=1.6.x, w/C++ & FORTRAN,
in production mode
Linux/64-ia64 2.6 (cobalt) w/Intel compilers, w/C++ & FORTRAN,
in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, in production mode
Mac OS X/32 10.4.10 (amazon) in debug mode
Linux/64-ia64 2.4 (tg-login3) w/parallel, w/FORTRAN, in production mode
- Extracted common code from H5D_select_read/H5D_select_write into single
routine (H5D_select_io) and made H5D_select_read/H5D_select_write
"gateway" routines, passing I/O buffer to operate on in "op" struct.
- Create optimized pathway in H5D_select_io for I/O operations on single
elements.
- Make "get linear offset of first element" callback for each type of
selection (used in "optimized pathway", above)
Tested on:
FreeBSD/32 6.2 (duty) in debug mode
FreeBSD/64 6.2 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (kagiso) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64-amd64 2.6 (smirom) w/default API=1.6.x, w/C++ & FORTRAN,
in production mode
Linux/64-ia64 2.6 (cobalt) w/Intel compilers, w/C++ & FORTRAN,
in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, in production mode
Mac OS X/32 10.4.10 (amazon) in debug mode
Linux/64-ia64 2.4 (tg-login3) w/parallel, w/FORTRAN, in production mode
