[svn-r17283] Description:
Bring r17276:17281 from trunk to revise_chunks branch
Tested on:
Mac OS X/32 10.5.7 (amazon) debug
h5committest not required on this branch
[svn-r17278] Description:
Bring r17272:17276 from trunk to revise_chunks branch.
tested on:
FreeBSD/32 6.3 (duty)
h5committest not required on this branch
[svn-r17273] Description:
Bring r17269:17272 from trunk to revise_chunks branch
Tested on:
FreeBSD/32 6.3 (duty) w/debug
(h5committest not required for this branch)
[svn-r17270] Description:
Bring r17265:17269 from trunk to revise_chunks branch
Tested on
FreeBSD/32 6.3 (duty)
(h5committest not required on this branch)
[svn-r17263] Description:
Bring r17244:17262 from trunk to revise_chunks branch
Tested on:
Mac OS X/32 10.5.7 (amazon) w/debug & production
FreeBSD/32 6.3 (duty) w/debug
h5committest not needed on this branch
[svn-r17236] Description:
Add full support for any [single] dimension being unlimited when using
extensible arrays as a chunk index.
Tested on:
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...
[svn-r17234] Improvement:
Added code to catch the exit code of the Generator.
Bug fixes:
Was doing "exit $nerrors" which is bad since Unix exit size
is defined as a small unsigned value. Changed it to exit 0
if no error, exit 1 if errors found.
Cosmetic change:
Rearrange some code locations so that it is cleaner and more
logical.
Tested:
Jam and Linew.
[svn-r17233] Description:
Bring r17209:17230 from trunk to revise_chunks branch.
Tested on:
Mac OS X/32 10.5.7 (amazon) debug
h5committest not required on this branch
[svn-r17232] Code improvement:
Added control variable to control the debugging output.
Changed default writer records to 200,000 since 1 million
took a long time to run in Linew. 200,000 still take 1.5minutes
to finish for linew to run.
Verified the error detection does work by making swrm_writer to
fail by setting the number of records to 10000.o
Tested:
Jam, amani and linew
[svn-r17225] Added code to detect the exit code of each of the writer and readers processes
and set nerrors accordingly. It still has a bunch of debug statements which
will be cleaned out in the next round.
Tested:
Jam.
[svn-r17209] Description:
Intermediate checkin for getting extensible array code working with
2+ dimension datasets that have only 1 unlimited dimension. This code is
working with n-dimension cases, but only when the slowest changing dimension
is unlimited. Generalization to n-dimension with any [single] dimension being
unlimited is forthcoming.
Tested on:
FreeBSD/32 6.3 (duty) in debug mode
FreeBSD/64 6.3 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (jam)...
[svn-r17184] Description:
Add "SWMR read" file access flag, for applications that are
concurrently reading from a file open with the "SWMR write" access flag.
This flag relaxes the internal checks for reading beyond the 'eoa'
for the file (since the eoa for the file could be out of date until the
SWMR write application flushes or closes the file)
Tested on:
FreeBSD/32 6.3 (duty) in debug mode
FreeBSD/64 6.3 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (jam...
[svn-r17157] Description:
Bring r17152 from trunk to revise_chunks branch & re-run
bin/reconfigure. (next merge from trunk to revise_chunks branch should start
with r17157)
Tested on:
FreeBSD/32 6.3 (duty)
(h5committested on trunk)
[svn-r17156] Description:
Hook fixed array data structure up to dataset code as a chunk index when
there are 0 unlimited dimensions.
Tested on:
Mac OS X/32 10.5.7 (amazon)
(h5committest not required on this branch)
[svn-r17154] Description:
Add fixed array data structure. (For initial use as a chunk index)
Tested on:
Mac OS X/32 10.5.7 (amazon)
(h5committest not required on this branch)
[svn-r17143] Description:
Bring r17084:17142 from trunk to revise_chunks brach
Tested on:
FreeBSD/32 6.3 (duty)
(h5committest not required on this branch)
[svn-r17141] Description:
Change name of public 'H5D_CHUNK_BTREE' symbol (indicating the type
of index used for locating chunks) to 'H5D_CHUNK_IDX_BTREE', putting old name
in "deprecated symbols" section. This will make adding new indices (like
extensible and fixed arrays) more nicely named.
Tested on:
FreeBSD/32 6.3 (duty)
(h5commitest not required on branch)
[svn-r17007] Description:
Final stage of connecting all the pieces of a chunked dataset with
an extensible array index together to fully enable SWMR-write access properly.
This change makes the dataset's object header depend on the extensible array
header, so that the extensible array must be fully synchronized to disk before
the object header (which contains possibly updated dimensions for the dataset)
is written to disk.
Tested on:
FreeBSD/32 6.3 (duty) in debug mode
FreeB...
[svn-r17005] Description:
Correct error in unpinning an entry from outside the cache to correctly
unpin the entry only when it is not being used for a flush dependency relation.
Tested on:
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,
...
[svn-r17003] Description:
Bring r16997:17002 from trunk into revise_chunks branch
Tested on:
Mac OS X/32 10.5.7 (amazon)
h5committest not required on this branch
[svn-r16998] Description:
Bring r16982:16997 from trunk into revise_chunks branch.
Tested on:
Mac OS X/32 10.5.7 (amazon)
h5committest not needed on this branch
[svn-r16988] Description:
Refactor layout initialization so that chunk index information can be
set for extensible array indices.
Tested on:
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, i...
[svn-r16985] Description:
Interim checkin of work toward closing race condition window which can
cause errors when reading a file that is used for SWMR-write access.
This change introduces a chunk proxy in the metadata cache, which
participates in the metadata cache's flush dependencies while representing a
raw data chunk in a dataset's chunk cache.
Also, the extensible array's SWMR behavior is only invoked when the
file is opened for SWMR-write access, allowing more flexibility in flush...
[svn-r16983] Description:
Bring r16977:16982 from trunk to revise_chunks branch.
Tested on:
FreeBSD/32 6.3 (duty)
(h5committest not required on this branch)
[svn-r16978] Description:
Bring r16976:16977 from trunk back to revise_chunks branch.
Tested on:
Mac OS X/32 10.5.7 (amazon)
(h5committest not required on this branch)
[svn-r16976] Description:
Merge r16960:16975 from trunk into revise_chunks branch.
Tested on:
Mac OS X/32 10.5.7 (amazon)
(h5committest not required on this branch)
[svn-r16975] Description:
Bring r16973 back from trunk to 1.8 branch:
Refactor chunk cache entry information to remove some [actually] unused
fields.
Tested on:
<tested on trunk>
[svn-r16960] Description:
Add a new H5F_ACC_SWMR_WRITE flag for H5Fcreate/H5Fopen, so that an
application can indicate that it is modifying the file in a SWMR scenario and
internal algorithms can make appropriate adjustments to order their I/O
operations to meet the constraints necessary for ensuring SWMR consistency.
Tested on:
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++ & FORT...
[svn-r16902] Description:
Bring r16875:16901 from trunk to revise_chunks branch.
Tested on:
FreebSD/32 6.3 (duty)
(h5committest not required for this branch)
[svn-r16877] Description:
Remove these files in preparation for merging them from the trunk.
Tested on:
(no test yet, this is just repository management)
[svn-r16850] Description:
Add some extra tests for when we are using extensible array as chunk
index.
Tested on:
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
So...
[svn-r16848] Description:
Add test (and bugfixes) for detecting if a filter makes a chunk size
larger than can be encoded in a 32-bit variable (in the file).
Tested on:
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,
...
[svn-r16846] Description:
Fix broken (for how long?) H5_ASSIGN_OVERFLOW macro to actually detect
overflows during assignments, along with several errors it [now] detected.
Cleaned up a fix minor warnings and/or pieces of code also.
Tested on:
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-amd6...
[svn-r16825] Description:
Bring revisions 16636:16821 from trunk to revise_chunks branch
Tested on:
FreeBSD/32 6.3 (duty)
Mac OS X/32 10.5.6 (amazon)
(h5committest not needed on this branch)
[svn-r16764] To fix a bug in extensible arrays as chunk index:
When the dataset is closed in H5D_close(), the pointer
to the extensible array struct in the layout message
is copied via H5D_flush_real() before H5D_chunk_dest().
This causes an abort from "Assertion `ea->hdr' failed"
later when the dataset is re-opened and read.
The bug was fixed by adding a flag to the reset function
for indexed storage to set/not set the address of the indexed
array. H5O_layout_copy() calls H5D_chunk_idx_rese...
