structs containing those arrays. Encapsulating the arrays in this way
makes it easier to write and think about pointers to these types, casts
to/from these types, etc.
An interesting side-effect that we probably should *not* rely on is
that the struct-encapsulation changes the alignment so that some GCC
warnings about casts that increase the alignment requirement of the
operand go away. Warnings like that have to be taken seriously: I will
add -Werror=cast-align to the default compiler flags so that they stop
the build quickly.
GCC warnings led me to some surprising casts in test/trefer.c. I found
that it was possible to make many simplifications after introducing the
struct-encapsulation that I described, above.
In test objcopy_ref `same_file` is assigned but never used. Delete it.
* commit '0f1be317c37ca04db263680e6f1fe4d97b937b39':
More fixes for the PR: (1) Set the version for reference datatype message to H5O_DTYPE_VERSION_4. (2) The tests for the new reference types should work for V112 and beyond.
Verify the decoded version for hyperslab selection.
* commit '75f2ed6167b094431fce2b4b700b3cf9da564c51':
Fixed a bug (HDFFV-10899) where the behavior of the deprecated H5Gget_objinfo() call had changed and passing in a non-existing soft link with a statbuf of NULL passed instead of failed (as it should as per the RM). The HDF5 1.8 and 1.10 behavior was restored and H5Gget_objinfo() will now fail and return -1 as before.
Fixed the java assert issue with H5O cache corking calls.
Corrected missing parallel #ifdefs.
Converted H5O MD cache cork calls to use the VOL.
Marked up H5Fget/set_mpi_atomicity() to use the VOL.
H5Gget_objinfo() call had changed and passing in a non-existing
soft link with a statbuf of NULL passed instead of failed (as
it should as per the RM). The HDF5 1.8 and 1.10 behavior was
restored and H5Gget_objinfo() will now fail and return -1 as
before.
* commit '5d2545ee26d4b7013ed363545705f16a67087549':
Fix func enter macro in H5T_ref_reclaim()
Update RELEASE.txt for reference changes
Fix reference type comparison in h5dump
Make wrappers, tests and tools use H5Treclaim() instead of H5Dvlen_reclaim()
Add new H5R API that abstracts object, region and attribute reference types
Remove ability to loc by ref from H5VL layer
Add support for retrieving object name by token
Add H5VL_OBJECT_GET_TYPE to get object type
Add H5VL_MAX_TOKEN_SIZE and H5VL_token_t
Adapt Jerome's "file info" H5VL 'get' query to retrieve container token info.
Fix H5VL_blob_get to return size of blob
Add 'blob' callbacks to VOL, along with a native implementation to store them in the global heap, and changed the VL datatype conversion code to use blobs.
* commit 'bffa18b684de83359f4ad25e11f4ad0b15f34040':
Merge pull request #1957 in HDFFV/hdf5 from ~LRKNOX/hdf5_lrk:develop to develop
Merge pull request #1934 in HDFFV/hdf5 from ~VCHOI/my_third_fork:bugfix/HDFFV-10585-investigate-slowness-of-regular to develop
Also support references to external files
Add new H5T_REF type and type conversion routines
Support conversion from H5T_REF_OBJ/DSET_REG to H5T_REF
Add H5Treclaim() API to reclaim memory of vlen/reference types
Deprecate H5Dvlen_reclaim()
Fix H5T_vlen_reclaim() and H5T_reclaim() to use private callback
Add H5T_ref_reclaim()
Move previous H5R APIs to H5Rdeprec.c
Clean up H5Ocopy
Separate H5O_copy_expand_ref() to H5Ocopy_ref()
Add support for copying new reference types
Clean up deprecated routines to go through VOL and same code path
Fix return codes in existing trefer.c test
Rename trefer.c to trefer_deprec.c
trefer.c is for new references
Add performance test for trefer
Add additional obj_copy_ref test
Make use of tokens and blobs to store references
Skip blob encoding for object references
Start adding new reference examples
* commit '7924eee0e5ee0745b784c635042b8633886fb799':
(1) Address the feedback from the PR review (2) Add release notes
Fix for HDFFV-10585 investigate hyperslab slowness: 1) Improve hyperslab performance when doing I/O from 1-d disjoint file dataspace to 1-d contiguous memory dataspace. 2) Move coding in H5D__chunk_io_init() that is constructing the chunk mappings to a separate routine.
Description:
Fixed tests to handle the case when a requested compression filter
is not available.
Platform tested
Linux/64 (jelly)
Darwin (osx1011test)
- Added new chunk query functions: H5Dget_num_chunks(), H5Dget_chunk_info(), and H5Dget_chunk_info_by_coord()
- Fixed iterator issue in H5EA_iterate(), H5FA_iterate(), and H5D__none_idx_iterate()
* commit '96dab622e3bbe75d85b0cef3ca2bd5288cf4ac32':
Fixed error code and error messages Tested on Jelly (very minor)
Changed H5VL_NATIVE_DATASET_GET_CHUNK_INFO_BY_COOR to H5VL_NATIVE_DATASET_GET_CHUNK_INFO_BY_COORD per a review comment.
Updated comments
Updated based on reviews and refactored test code
Code cleanup and refactor
Re-ordered items w.r.t. date.
Fixed typo
Added notes about HDFFV-10677 and HDFFV-10661
HDFFV-10677 and HDFFV-10661
Description:
- Added functions to query chunk information:
H5Dget_num_chunks(dset_id, fspace_id, *nchunks)
Gets the number of written chunks that intersect with the given
dataspace. However, in this version, the intersection is not
yet completed. Thus, the number of all written chunks will be
returned.
H5Dget_chunk_info_by_coord(dset_id, *offset, *filter_mask, *addr, *size)
Given a chunk's logical coordinates, returns the chunk's filter,
address, and size.
H5Dget_chunk_info(dset_id, fspace_id, index, *offset, *filter_mask, *addr, *size)
Given a chunk's index, returns the chunk's logical coordinates, filter,
address, and size. The chunk belongs to a set of chunks that have
nonempty intersection with the specified dataspace. However, in
this version, the intersection is not yet completed, and the index
is of all the written chunks.
These functions comply with VOL.
- Fixed some oversights found in the library for the tests in chunk_info.c
to work correctly. The returned value from a callback function was not
checked in H5EA_iterate(), H5FA_iterate(), and H5D__none_idx_iterate().
This oversight caused a callback function to continue iterating even though
it's supposed to stop.
Platforms tested:
Linux/64 (jelly)
Linux/64 (platypus)
Darwin (osx1011test)
