Code cleanup
Description:
Clean up compiler warnings, especially the 'FUNC' variable not used which
comes out in production mode.
Solution:
Had to add a new FUNC_ENTER_NOAPI_NOINIT_NOFUNC macro for those non-API
functions which don't need the 'FUNC' variable defined. (This will be _so_
much easier when C99 is standard on all our supposed platforms, since it has a
__FUNC__ macro... )
Platforms tested:
FreeBSD 4.9 (sleipnir)
too minor for h5committest (although there were lots of files changed, the
change was minor in each one)
Bug fix
Description:
Make a copy of the index value for H5Giterate and use that instead of
dereferencing the index pointer.
Platforms tested:
FreeBSD 4.9 (sleipnir)
too minor to need h5committest
Omnibus floating-point bug fix changes
Description:
There are a number of problems in the floating-point conversion code that
were exposed by Ray's recent int<->float checkin:
- The 'my_isnan' code in test/dtypes.c was broken and would always return
true. The meant that the actual values in the float<->float conversion
tests were _never_ checked, hiding the other bugs included in this
checkin.
- A recent change I made to the type conversion code used "FLT_MIN" instead
of "-FLT_MAX" for the most negative 'float' value for the double->float
conversion, which meant that any the negative number that was converted
from a double to a float would have been mapped to zero, essentially.
- A change that Robb appeared to have made ~2.5 years ago to the "generic"
float->float conversion routine appears to be incorrect and I've backed
it out.
- Floating-point conversions on SGI's which converted denormalized values
would be mapped to zero instead of being propertly preserved in the new
type. This was addressed by an SGI-specific system call to prevent the
behavior.
Solution:
Described above, generally.
Platforms tested:
FreeBSD 4.9 (sleipnir)
h5committest
Misc. update:
release_docs/RELEASE update forthcoming...
Code cleanup
Description:
Remove an unused typedef, fix a LONG->LLONG typo and use LLONG_MIN where
appropriate.
Platforms tested:
FreeBSD 4.9 (sleipnir)
too minor to require h5committest
Code cleanup
Description:
Make some functions static and remove unused variables, etc.
Platforms tested:
FreeBSD 4.9 (sleipnir)
too minor to require h5committest
Description: data type conversion between integers and float numbers.
(Cover your ears. It's going to explode.:)
Solution: covers all native type conversion. Mainly uses hardware
conversion but handles overflow more gracefully.
Platforms tested: h5committest
Bug fix
Description:
Our previous "optimization" of metadata writing which only wrote metadata
from one process was abusing MPI-I/O and after some consultation with Rob Ross
and Rajeev Thakur, Albert & I have come up with a solution...
Solution:
Instead of only writing from one process, issue a collective write
operation with all processes, for metadata writes.
Platforms tested:
FreeBSD 4.9 (sleipnir)
h5committest
Bug Fix
Description:
The SAP was sending back replies to the client but the client wasn't
picking them up (this was after a dump from the server.
Solution:
Read the extra replies from the server.
Platforms tested:
AIX (w/ FPHDF5)
Linux (w/ FPHDF5)
Solaris (w/ Fortran & C++)
Misc. update:
Bug fix.
Description:
The new routines added for H5MM_[m|c]alloc were causing the function
stack code to infinitely recurse while allocating room for the thread-specific
information (when thread-safety was turned on).
Solution:
Call HDmalloc directly instead of H5MM_malloc
Platforms tested:
FreeBSD 4.9 (sleipnir)
Changed are not tested in h5committest
Bug fix
Description:
Variable length strings and sequences with NULL pointers were not handled
by library, causing problems access the data. This also affected fill values
for variable-length datatypes.
Solution:
Address the issues in the library by detecting NULL sequences/strings
and avoid trying to convert them.
Patched up dumper to display NULL sequences/strings.
Platforms tested:
FreeBSD 4.9 (sleipnir)
h5committest
Bug fix and File Addition
Description:
- Fixed when reading from the file with FPHDF5. It wasn't recording
how many bytes it read.
- Added Arithmetic Transformation modules. These haven't been
included into the HDF5 build. I just added them here for future
porposes...
Solution:
Added a call to "MPI_Get_count" to get the number of bytes read.
Platforms tested:
Linux (w/ FPHDF5)
AIX (w/ and w/o FPHDF5)
FPHDF5-specific fixes...No need for H5committest
Misc. update:
Bug fix & code cleanup
Description:
Allowing the library to call malloc with a size of 0 bytes causes problems
for some users, so we check for allocations of 0 bytes and disallow them now.
Cleaned up some code which could call malloc with 0 size.
Changed some code calling HDmalloc directly to call H5MM_malloc(), which
allows us to check for 0 sized allocations.
Platforms tested:
FreeBSD 4.9 (sleipnir)
too minor to require h5committest
Bug fix
Description:
Make --disable-hsizet work properly again by fixing a couple of tests and
correcting the definition of SSIZET_MAX.
Platforms tested:
FreeBSD 4.9 (sleipnir)
h5committest not used, as it doesn't test this configuration.
Code cleanup
Description:
Cleanup compiler warning by breaking apart the cast checking.
Platforms tested:
FreeBSD 4.9 (sleipnir)
too minor to need h5committest
Code optimization
Description:
Improved integer & floating-point datatype conversions by removing some
corner cases. Got rid of algorithm which walked the buffer of elements to
convert backwards, switching to algorithm which computes the non-overlapping
space in the buffer and repeatedly converts it in the forward direction.
Platforms tested:
FreeBSD 4.9 (sleipnir)
h5committest
Cray SV1 & T3E (T90 is not accessible)
Bug fix
Description:
Don't check the number of szip parameters set during the "can apply" and
"set local" callbacks, be safe about setting the parameters instead.
Platforms tested:
FreeBSD 4.9 (sleipnir)
too minor to require h5committest
Warning cleanup
Description:
Make a parameter 'const' to clean up a warning during compiles.
Platforms tested:
FreeBSD 4.9 (sleipnir)
too minor to require h5committest
Bug Fix
Description:
The End of Address information needed to be kept by the SAP. Some
processes were trying to get the EOA information in collective mode,
but the action wasn't collective at the time.
Solution:
Keep the EOA information for the file on the SAP. Clients query the
SAP to get/set it when needed.
Platforms tested:
Linux (w/ FPHDF5)
Copper (w/o FPHDF5)
(FPHDF5 specific, so no need for full testing)
Misc. update:
Minor code cleanup
Description:
Fixed up comments, made the H5I_get_file_id() routine static, removed
H5Gprivate from including itself...
Platforms tested:
FreeBSD 4.9 (sleipnir)
too minor to require h5committest
Bug fix
Description:
MPE color for new routine H5Iget_file_id is missing.
Solution:
Added it.
Platforms tested:
Only tested in copper since it is the only one that can do MPE.
Misc. update:
Code cleanup & simplification
Description:
Replace [non-working] code in routine to build up an MPI derived type for
a hyperslab selection that was supposed to "flatten" out contigous blocks with
code that uses a selection iterator (which _does_ have working "flattening"
code).
Remove "contiguous hyperslab" code for MPI selections as it is unnessary
now.
Platforms tested:
FreeBSD 4.9 (sleipnir) w & w/o parallel
Linux 2.4 (verbena) w/FORTRAN & C++
Solaris 2.7 (arabica) w/64-bit
Bug fix & code cleanups
Description:
Change our use of MPI derived datatypes to not create datatypes with
"0-sized" lengths, which causes the LANL Q machine to hang.
Also, get rid of "prefer MPI derived datatypes" environment variable since
it has no advantage.
Platforms tested:
FreeBSD 4.9 (sleipnir) w & w/o parallel
h5committest
Code cleanup & add a feature
Description:
Added support for querying the file ID of named datatypes to new
H5Iget_file_id function.
Platforms tested:
FreeBSD 4.9 (sleipnir)
too minor to require h5committest
Code cleanup
Description:
Untwist the last few sections of code before starting on fixing the
problems with reading off the end of the file in earnest.
Platforms tested:
FreeBSD 4.9 (sleipnir)
FreeBSD 4.9 (sleipnir) w/parallel
h5committest not necessary - parallel only changes
Bug fix
Description:
Code was using an internal DXPL with collective I/O turned on during
independent I/O operations to fill dataset on disk.
Solution:
Switched to internal DXPL with independent I/O
Platforms tested:
FreeBSD 4.9 (sleipnir)
FreeBSD 4.9 (sleipnir) w/parallel
Linux 2.4 (verbena) w/fortran & C++
Solaris 2.7 (arabica) w/64-bit extensions enabled
Fix serial build
Description:
I accidentally put the "use_par_opt_io" variable in an #ifdef PARALLEL
section.
Solution:
Hoist it out of parallel section
Platforms tested:
Eyeballed it - very trivial
Code cleanup
Description:
Straighten out more goofiness in the MPI code dealing with collective I/O
transfers - mostly make certain that a view is set if-and-only-if collective
I/O is occurring on raw data (and vice versa for views and independent I/O)
Platforms tested:
FreeBSD 4.9 (sleipnir) w/parallel & FPHDF5
too minor to repquire h5committest
Code cleanup
Description:
Remove collective parallel I/O checking code, now that it's impossible to
get into this code with collective access enabled.
Platforms tested:
FreeBSD 4.9 (sleipnir)
too minor to require h5committest
Code cleanup
Description:
More refactoring on setup of MPI-IO parameters: hoisted the code to change
the transfer mode from collective to independent up into the H5D_read and
H5D_write routines, instead of duplicating it inside each of the chunk/contig
read/write routines.
Platforms tested:
FreeBSD 4.9 (sleipnir)
too minor to require h5committest
Code cleanup
Description:
Clean up setup code for collective I/O transfers to make the logic more
obvious.
Platforms tested:
FreeBSD 4.9 (sleipnir)
too minor for h5committest
Bug fix
Description:
Single hyperslab selections (which were set with only one call to
H5Sselect_hyperslab) that had dimensions that could be "flattened" but were
interspersed with dimensions that could not be flattened were not correctly
handled, causing core dumps.
Solution:
Re-work "flattening" code to handle this case properly.
Platforms tested:
FreeBSD 4.9 (sleipnir) w/parallel
h5committest
Code cleanup
Description:
Refactored handlier of VFL drivers in file access and data transfer property
lists in order to simplify and unify the code dealing with them.
Platforms tested:
FreeBSD 4.9 (sleipnir)
too minor to require h5committest
Bug fix
Description:
Change "H5_HAVE_PARALLEL" to "H5_HAVE_FPHDF5" around VFL ID, to prevent
daily builds from failing.
Platforms tested:
FreeBSD 4.9 (sleipnir) w/parallel
too specialized to require h5committest
Bug Fix
Description:
The owners and reference counts of locks weren't being handled
correctly.
Solution:
Allocate an array the size of the COMM FPHDF5 is invoked with for
each lock. Then reference count within this array. Keep an extra
"counter" to see how many different processes have this particular
lock. Free the lock when that counter reaches zero.
Platforms tested:
Linux & Copper (FPHDF5 specific).
Misc. update:
Bug Fix
Description:
Was using HGOTO_ERROR within a "done:" block.
Solution:
Changed HGOTO_ERROR macro to HDONE_ERROR inside of a done: block.
Platforms tested:
Linux (Small change).
Misc. update:
Message Correction
Description:
Uncapitalized the beginning of the error messages.
Solution:
Platforms tested:
Linux (Very small change).
Misc. update:
Bug fix
Description:
Remove the [duplicated] native C9x datatype initializatin code from H5T.c
and let the automatically generated code in H5Tinit.c handle initializing them,
if they are allowed by the compiler.
Platforms tested:
FreeBSD 4.9 (sleipnir)
Linux 2.4 (verbena, eirene)
probably should be h5committested, but it's late and I'm fairly certain
this will fix the problem... :-/
Bug fix
Description:
Avoid "OPAQUE" symbol in template macro invocation, since it's already
defined as a macro on Windows.
Solution:
OPAQUE -> OPAQ
Platforms tested:
FreeBSD 4.9 (sleipnir)
Windows ? (Pedro's machine :-)
Refactor code
Description:
Use "template macro" technique to eliminate a huge amount of repeated code
in type initialization code (almost 1000 lines, between this change and the
changes to the H5Tconv.c file earlier).
This centralizes the common parts of the code and makes the different parts
of the code more obvious.
Platforms tested:
FreeBSD 4.9 (sleipnir)
too minor to require h5committest
