Bug 2089 - GMQS: h5diff segfault on a compound dataset with fixed length +
vlen string type order
Description:
This is fix for the segfault when h5diff compares a compound dataset with
combination of fixed length string types and vlen string types in certain
orders. Optimized vlariable length string handling codes. The fix is
referred from h5dump handling vlen strings.
For testing, several compound datasets were added with various combinations.
Previous failed cases:
- Vlen string, Fixed length string, Vlen string, Fixed length string
- Fixed length string, Fixed length string, Vlen string, Vlen string
- Fixed length string, Vlen string, Fixed length string, Vlen string
Tested:
jam (linux32-LE), amani (linux64-LE), heiwa (linuxppc64-BE), tejeda (mac32-LE), linew (solaris-BE)
"*buffer_size = j + 1;" was mistakenly taken out by someone. It is necessary to update the new size. I
put it back and made 2 test cases for integer and float to verify the correct dataset size.
I'm bringing the fix from 1.8 branch. The changes to configure.in, tools/misc, config, Makefile.am are only property changes.
Tested on jam. But I tested 1.8 on jam, heiwa, and amani.
Checked in fix for failure in shape same tests that appeared after
Quincy's recent massage of the test code. The problem was a race
condition created when Quincey re-worked the code selecting either
collective or independant I/O.
Previously, when independant I/O was selected in the test, I had
used H5Pset_dxpl_mpio() and H5Pset_dxpl_mpio_collective_opt() to
select collective semantics with independant I/O going on under
the hood. Quincey modified this to call H5Pset_dxpl_mpio() when
collective I/O was selected, and do nothing in the independant I/O
case. As a result, processes were able to race ahead and
modify the initial values of the data set before some processes
had verified that the initialization was correct.
Solved the problem by adding barriers, and making all barriers
dependant on independant I/O being selected.
Tested parallel on amani and phoenix. h5committested.
Note that parallel on amani and h5committest on heiwa failed
several times before I got a clean pass without code changes.
The failures on amani seemed to be time outs caused by contention
for the machine -- worryingly, they occurred in the shape same
tests. However, given subsequent passes and passes on jam and
phoenix, I am going ahead with the commit.
The failure on heiwa was in the fheap test. I don't see how
this can be related to changes in testpar, and in any case, it
went away on the second try.
Changed the test to run Huge Dataset (tens of GB), Xtra large dataset(4GB big),
Large dataset (2GB big), or merely 1GB big, depending on if the file system
supports sparse file or if it supports larger than 32bigs I/O.
Tested: h5committest, jam (serial), Windows (Bangan which is Windows 7, 32bits).
file name listed for DISTCLEANFILES should be testlinks_env.sh
2. Add comments to test_4() in test/external.c about the link name "/ link".
3. Fix memory leak as reported by valgrind in src/H5Lexternal.c:
free memory pointed to by tmp_env_prefix for HDF5_EXT_PREFIX case.
Add "--enable-unsupported" configure flag.
Description
The "--enable-unsupported" configure flag allows a user to
prevent configure from failing due to the use of incompatible
options, such as c++ with parallel.
Specifying --enable-unsupported will bypass all of configure's
checks for incompatible and unsupported combinations of flags.
There are no guarantees that the library will be configured
in any sort of working condition, but that's the risk of
using the --enable-unsupported flag.
I've changed all default error messages related to unsupported
option combinations to indicate that using --enable-unsupported
will allow configure to complete without error.
Tested:
by hand on jam, tested all unsupported configure option
combinations with and without the new flag, making sure
the flag allows configure to finish without error.
(h5committest wouldn't do any good here; it won't test
the new option, and since we're enabling unsupported
combinations, failures are likely to occur in build or
tests with --enable-unsupported turned on anyways.
That's why they're unsupported!)
Add additional error checking to catch erroneous user input.
Description:
Attempting to retrieve a links's name by index in the case
where the link is external and the file that the object is
located in doesn't exist was causing a segmentation fault
(in production) and an assertion failure (in debug).
The segfault wasn't occuring until the metadata accumulator
attempted a write, so I've added error checking higher in
the pipeline in H5O_protect (where there was previously just
an assert) to catch this. I've also added additional asserts
in the H5F layer where there were none.
Additionally, I added another case to the links.c test to
test that this fails gracefully instead of segfaulting or
asserting out.
Tested:
h5committest and gandalf (mac os x)
larger than 32bits I/O depends on the support of fseeko. Windows does not
use that. Instead, it uses _fseeki64 to support larger than 32bits I/O.
Tested: jam (linus) and bangan (windows)
Also updated root CMakeLists.txt to output a message when unsopported options are configured with the PARALLEL option. CMake will still generate files.
Tested: windows and local linux
1) Move the test for H5D_EXT_PREFIX in links.c to a separate file: links_env.c
2) links_env.c will be used by testlinks_env.sh to test for the environmental
variable H5D_EXT_PREFIX in searching for the external linked file.
