Description: H5Tset_order will now properly reject H5T_ORDER_NONE for most
datatypes. Previously this could cause major problems as the file could not be
flushed.
Tested: jam, smirom (h5committest)
Bring r16353 back from revise_chunks branch:
Refactor internal layout information, making it easier to add another
type of chunk index.
Tested on:
FreeBSD/32 6.3 (duty)
(other platforms tested with original patch)
Description: H5Tinsert will now detect when a compound type that was previously
not packed becomes packed due to out of offset order insertion of a member.
H5Tinsert will now attempt to keep members sorted by offset order. This should
improve performance of H5Tinsert in all cases due to the fact that it no longer
needs to check every other member for overlapping, and should improve
performance of H5Tpack and possibly type conversion when compounds are packed
out of order.
Tested: jam, smirom (h5committest)
Bug Fix
Description:
Removing some unnecessary system headers from test/cache.c which snuck
their way into one of my previous check-ins ... thanks to Allen for
catching this while testing on Windows.
Tested:
Windows, Linux
Code Cleanup
Description:
- Pulled out repetitive cache stats code into its own new macro.
- Converted HDasserts in cache stats macros to if / HGOTO_ERROR statments
in order to keep code consistent.
Tested:
jam, smirom, liberty
Improvement
Description:
Modified the warning messages from the Version mismatch checking to suggest
the users to try recompiling or checking the shared lib setting.
Platforms tested:
Tested in Jam only since it was just a simple text string changes.
Description:
If a compound type was packed except for some extra space at the end, H5Tpack
would not modify the type and the extra space would remain. Changed
H5T_is_packed to fix this behaviour.
Tested: jam, smirom (h5committest - linew down)
Bring r16305 back from revise_chunks branch:
Add detection of C99 "designated initializers" to configure script and
use new H5_HAVE_C99_DESIGNATED_INITIALIZER macro to conditionally compile
default layout variables in src/H5Pdcpl.c
Also, minor code cleanups, etc.
Tested on:
FreeBSD/32 6.3 (duty) in debug mode
(Other platforms tested on branch)
-N, --nan Avoid NaNs detection
Note: there is no shell script run for datasets with NaN because the output is non portable (different results and NaN strings for different systems)
Tested: windows, linux
Storage: information not available
When displaying storage information for VL and dataset region types
Added 2 shell runs that display this information
#818
Tested: windows, linux
Bug Fix
Descriotion:
Removing problematic debugging code, and switching a leftover TRUE verbose
statement to FALSE in cache_common.c
Tested:
jam, liberty
Adding code to maintain a min_clean_fraction in the cache in serial mode.
Description:
The metadata cache now has the ability to maintain a min_clean_fraction
when in serial mode. The default initial cache size has been changed
from 1MB to 2MB, and the default min_clean_fraction has been set at 30%.
This check-in includes modifications to H5C.c to support maintaining a
min_clean_size, including the addition of clean_index_size and
dirty_index_size trackers, modifications to the H5C_make_space_in_cache
algorithm, as well as associated test code and additional statistics
tracking variables.
Maintaining the min_clean_fraction addresses the possibility of
experiencing a "metadata blizzard" when the cache gets completely
full with dirty entries. Upon having to make space, the cache would
previously need to flush every single entry in the cache before coming
across a clean entry which could be evicted. This resulted in unnecessary
flushing of oftentimes hot entries in the cache. Maintaining the
min_clean_fraction ensures that, when space is needed, clean entries
are more readily available to evict.
Tested:
jam, smirom, linew (h5committest)
- h5repack: When user doesn't specify a chunk size, h5repack now defines a default
chunk size as the same size of the size of the hyperslab used to read the chunks.
The size of the hyperslabs are defined as the size of each dimension or a
predefined constant, whatever is smaller. This assures that the chunk
read fits in the chunk cache. (PVN - 2008/11/21)
- H5Dset_extent: when shrinking dimensions, some chunks were not deleted.
(PVN - 2009/01/8)
Description: Added H5Pset/get_elink_cb to allow the user to specify a callback
function to be called whenever an external link is traversed. Added
H5Pset/get_elink_acc_flags to allow the user to specify the file access flags
to use to open the target file of an external link. All these properties are set on a LAPL.
Tested: jam, linew, smirom (h5committest)
Bring revision 16278 back from revise_chunks branch:
Update layout information in DCPL to unify all information in one
underlying property and switch to using H5O_layout_t for storing it, which
simplifies things considerably.
Also, fix many compiler warnings.
Tested on:
FreeBSD/32 6.3 (duty) in debug mode
(Original patch tested on many machines)
The Tail command in jam (a newer linux) does not accept the +2l option.
It ended up wiping most of the release_doc/RELEASE.txt file contents.
Replaced the "tail +2l" by "sed -e 1d".
Tested:
Tested in Jam to verify it functions properly again.
new test program: It adds tests for several ranks, use of fill value or not, compression, different fill value allocation times, use of different storage layouts, and external files
tested: windows, linux
PG compiler complains about array out of bounds (a rank of zero was not checked)
Adding a scalar dataset to the test generator program. this case is run on a previous existing run, the case was added to 2 existing files
Tested: windows, linux
to the beginning of the file. Otherwise, the file might be re-extended later on Open VMS.
Also updated the return value for the HDlseek to be more appropriate.
Tested 1.8 on Open VMS.
Bug fix (#1357)
Description:
Three filters have not assigned correct value to one value-result argument, "buf_size". N-bit, szip, and scale offset filter have had this problem.
However, I don't think this problem has been making buffer overrun because those filters were informing the caller that the "buf", another value-result argument, is smaller than it actually is. If there was actual buffer overrun, I believe another problem exists although I don't know.
Tested:
jam, smirom, linew
Although all test were passed, I'm concerned about valgrind memcheck error. There can be another miscommunication between filter and the caller.
Moved v1 B-tree debugging routines into separate module and thinned
out header files a bit.
Tested on:
Mac OS X/32 10.5.5 (amazon) in debug mode
Mac OS X/32 10.5.5 (amazon) w/C++ & FORTRAN, w/threadsafe,
in production mode
FreeBSD/32 6.3 (duty) in debug mode
FreeBSD/64 6.3 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (jam) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64-amd64 2.6 (smirom) w/Intel compilers w/default API=1.6.x,
w/C++ & FORTRAN, in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, in production mode
Linux/64-ia64 2.6 (cobalt) w/Intel compilers, w/C++ & FORTRAN,
in production mode
Linux/64-ia64 2.4 (tg-login3) w/parallel, w/FORTRAN, in production mode
