Close FAPL for external link test, to fix infinite loop closing the library.
Tested on:
MacOSX/64 10.10.5 (amazon) w/serial
(too minor to require h5committest)
Refactor property list code to "deep copy" properties in the correct way,
retraining the rest of the library to copy & release things correctly. This
cleans up another batch of memory leaks, etc. within the library.
Tested on:
MacOSX/64 10.10.5 (amazon) w/serial & parallel
Linux/32 2.6.x (jam) w/serial & parallel
(h5committest forthcoming)
compatible between 32 and 64 bit systems.
Expand cross platform property list encode/decode testing.
Fix warning in H5P__fill_value_enc.
Tested: jam, koala, ostrich (h5committest), ummon
NOTES:
- Developers will have to run autogen.sh before building with the autotools.
- autogen.sh takes the -p option to mimic the old bin/reconfigure behavior.
- The generated error, overflow and version headers have been left in place.
- The generated H5LT parser code has also been left in place.
- There are no changes for CMake users at this time.
Tested on: h5committest
Merge in part of Neil's property list fixes: Fix a minor memory leak,
add some sanity checks, clean up the code a little, and verify that encoding/
decoding default versions of the property lists works correctly.
Tested on:
MacOSX/64 10.10.5 (amazon) w/serial
(Too minor to require h5committest)
Complete revamp of package initialization/shutdown mechanism in the library.
Each package now has a single init/term routine.
This new way should avoid packages being re-initialized during library
shutdown and is also be _much_ more proactive about giving feedback for
resource leaks internal to the library.
Introduces a new "module" header file for packages in the library
(e.g src/H5Fmodule.h) which sets up some necessary package configuration macros
for the FUNC_ENTER/LEAVE macros. (The VFL drivers have their own slightly
modified version of this header, src/H5FDdrvr_module.h)
Also cleaned up a bunch of resources leaks all across the library and tests,
along with addressing many warnings, as I encountered them.
Tested on:
MacOSX/64 10.10.5 (amazon) w/serial & parallel
Linux/64 3.10.x (kituo) w/serial & parallel
Linux/64 2.6.x (ostrich) w/serial
Bring the "metadata rings" code from its branch (mdc_rings_v2) to the trunk.
(This change will support the page buffering feature)
Tested on:
MacOSX/64 10.10.5 (amazon) w/serial & parallel
(h5committest forthcoming)
Fix handling of layout version number.
Other minor fixes/cleanup.
NOTE: This commit changes the file format for VDS, all files must be
re-generated.
Tested: ummon
gcc 4.9.2 was used to create the warning list
- implicit casts
- shadowed variables
- various enum issues
- other minor fixes (comments, unused macros, etc.)
Tested on: h5committest
The only remaining code consists of a few floating-point tests
that rely on pre-generated and checked-in VMS files. These have
been left alone, even though they will not be possible to
recreate, since testing VMS float behavior is still important.
Tested on: h5committest
Tested w/ h5committest
NOTES: - The manifest may still be messed up.
- Cmake fails since the dual binary work needs to be merged with
this repo's CMake externals.
numbers of elements selected, incorrect "slice" through limited dimensions,
incorrect number of printf substitutions.
Patch source selection space extent with bounds of selection if extent is
unknown before returning via H5Pget_virtual_srcspace.
Write updated VDS dataspace to file when it changes due to unlimited selections.
Other minor fixes/cleanup.
Tested: ummon
Renamed a few "h5tri_t extended" variables to "h5tri_t was_extended".
Apparently, "extended" is a typedef name someplace in the headers used
on Solaris, which causes potentially confusing warnings to be emitted.
Tested on: h5committest
Solaris w/ Solaris Studio (emu)
Description:
XL compilers in ostrich (PowerPC64 linux) fail in test/dt_arith
because of the removal of the LLONG_TO_LDOUBLE_CORRECT (removed in
r26625) and LDOUBLE_TO_LLONG_ACCURATE (removed in r26623).
Solution:
Reverse revisions
r26623: bring back LDOUBLE_TO_LLONG_ACCURATE configure macro
r26625: bring back LLONG_TO_LDOUBLE_CORRECT configure macro
r26627: bring back WANT_DATA_ACCURACY configure macro which is
used together with the above two macros. This also brings
back the enable-dconv-accuracy configure option.
Tested:
h5committested.
Also tested in ostrich using the XL compilers.
Implement support for H5Pencode/decode with VDS properties.
Testing for H5Pencode/decode with VDS.
Relax assertion check for number of elements written to buffer as we do not
check for overlaps in mappings.
Other minor fixes/cleanup.
Tested: ummon
a different ID in the same type. Added a new skiplist routine,
H5SL_try_free_safe, which iterates over items, freeing some of them, and which
intercepts and defers attempts to remove from the list outside of the main
iteration. Changed H5I_clear_type to use this function.
Tested: jam, koala, ostrich (h5committest); ummon
- Updated the preprocessor logic that defines the H5_ATTR_* macros. The Solaris
Studio compiler understands __attribute__ but not the specific attributes
that we use. The H5_ATTR_* macros are now defined as blank with that
compiler which significantly cleans up the warnings, making debugging
easier.
- Replaced a few lingering __attribute__ macros with H5_ATTR_* markup.
Tested on: h5committest
Solaris w/ Solaris Studio (emu)
Implement support for H5Ocopy on VDS.
Testing for H5Ocopy on VDS.
Fix check-vfd failure.
Other minor fixes/cleanup.
Note there is a preexisting failure in h5repack testing.
Tested: ummon
Correct error with braces around macro (in production builds), also clean
up a few warnings.
Tested on:
MacOSX/64 10.10.3 (amazon) w/serial, parallel & production
Linux/32 2.6 (jam) w/serial & parallel
Clean up parameters to H5C__flush_single_entry: accept cache entry instead
of address, and make new flag for deleting entry from skiplist on destroy
instead of a separate parameter to the routine.
Tested on:
MacOSX/64 10.10.3 (amazon) w/serial & parallel
Linux/32 2.6.x (jam) w/serial & parallel
Split parallel metadata cache code into separate source code modules.
Tested on:
MacOSX/64 10.10.3 (amazon) w/serial & parallel
(too minor for h5committest)
Bring 'fail if unknown always' object header message flag from the
avoid_truncate branch, strengthen error checking for unknown messages, and try
to catch object header code configurations at compile time (instead of runtime).
Tested on:
MacOSX/64 10.10.3 (amazon) w/serial & parallel
(h5committest forthcoming)
to open source files and datasets, instead of using the default property lists.
Modify VDS to reject attempts at parallel I/O, fix parallel compile error.
Modify VDS to not write fill value to memory buffer if the fill value is
undefined.
Other minor fixes/cleanup.
Tested: ummon
extent.
Improve algorithm for calculation of VDS extent by removing the need for a
temporary copy of a dataspace.
Other minor fixes/cleanup.
Tested: ummon
Large code cleanups and optimizations, in preparation for the upcoming
v3 metadata cache merge.
Tested on:
MacOSX/64 10.10.3 (amazon) w/serial & parallel
Linux/32 2.6.x (jam) w/serial & parallel
library no longer iterates over all source datasets in the printf mapping, it
now only looks at the ones that could be involved in I/O (the ones whose bounds
overlap with the selection in the unlimited dimension).
Tested: ummon
H5_VDS_FIRST_MISSING)
Add test for this
Rework VDS code to not depend on unlimited selections having a "clipped" state
in preparation for removing the clipped state from unlimited selections.
Other bug fixes/cleanup
Tested: Kubuntu 64 (home computer)
test/Makefile.in was updated due to changes in the Makefile.am
file that missed a reconfigure. The changes just added a few
VDS test files to the deletion list.
Note that there are trace issues with a new VDS struct that
show up when running bin/reconfigure. I reverted these for
fixing at a future date.
Tested on: jam
