Finish deprecating last H5G symbol (H5G_obj_t) - yay!
Lots of misc. library fixes to remove confusion between links and
objects. The tools could still use another pass, to remove h5trav_type_t type
and make the correct distinction between links & objects.
Tested on:
FreeBSD/32 6.2 (duty) in debug mode
FreeBSD/64 6.2 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (kagiso) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64 2.6 (smirom) w/default API=1.6.x, w/C++ & FORTRAN,
in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, in production mode
AIX/32 5.3 (copper) w/FORTRAN, w/parallel, in production mode
Mac OS X/32 10.4.10 (amazon) in debug mode
Tested platform:
Kagiso only since it is only a comment block change. If it works in one
machine, it should work in all, I hope. Still need to check the parallel
build on copper.
Bug fix
Description:
ph5diff had been hanging in Tflops. Found out that vsnprintf ph5diff
uses was a local coded that did not limit printing according to the size
argument. That resulted in buffer overflow and other problems.
Solution:
Added some sort of size checking in the home-grown vsnprintf and
had ph5diff checked for error return of vsnprintf. Leon also
revamped the ph5diff manager's way of handling communications with
the workers. That eliminated all but the last case of hanging.
Platforms tested:
Tflops.
Misc. update:
Code cleanup
Description:
Trim trailing whitespace, which is making 'diff'ing the two branches
difficult.
Solution:
Ran this script in each directory:
foreach f (*.[ch] *.cpp)
sed 's/[[:blank:]]*$//' $f > sed.out && mv sed.out $f
end
Platforms tested:
FreeBSD 4.11 (sleipnir)
Too minor to require h5committest
Bug fix for parallel case for new "Some objects were not comparable" patch to h5diff.
Description:
The parallel h5diff wouldn't print out "Some objects were not comparable" because the worker
tasks were not communicating the not_cmp flag of the diff_opt_t struct back to the manager, who ultimately
prints everything.
Also, some miscellaneous fixes for error printing. Some errors were printed out with printf instead of fprintf(stderr,...).
In parallel environments, this can result in output getting lost.
Solution:
Had the worker tasks pass along the not_cmp flag to the manager when they sent along
the number of differences they found.
Platforms tested:
heping (pp), sol (pp)
Misc. update:
Purpose:
Bug fix.
Description:
ph5diff fails on modi4 due to the way snprintf works on IRIX.
Solution:
The C99 standard says that, if there isn't enough room in the string,
snprintf should return the number of characters that
would have been written to the output string if there were enough room.
The snprintf on modi4 would return the number of characters that is was able to write
succesfully to the string if space ran out. The ph5diff logic that checks if
the output buffer was full did not handle this sort of return value correctly.
Used VSNPRINTF_WORKS from configure test to check how snprintf works and do
the logic accordingly.
Platforms tested:
modi4
Misc. update:
Bug Fixes
Description:
Fixes for several bugs, including dumping of excess output to a temporary file, fix for printing
hsize_t datatype, and the long awaited fix for intermixed output.
Solution:
Fix 1: Overflow file
Previously, any output that a worker task made was buffered locally in memory, up to a point. Any
output beyond the size of the buffer (used to be 10k) was discarded. Now, the memory buffer size has been
changed to 1k and any output beyond this amount is sent a temporary file. This way, no output is lost
and memory usage is kept under control. The temporary file is deleted as soon as a worker task finishes
sending its contents to the manager.
Fix 2: hsize_t printing
Printing of the hsize_t datatype used to be handled by %Hu passed to HDfprintf. However, there is no corresponding HDvsnprintf that
is able to print hsize_t types. These are now printed with the aid of H5_PRINTF_LL_WIDTH.
Fix 3: Intermixed output fix
Intermixed output would occur on some machines (although I haven't seen it happen for a while) due to the unpredictability of the underlying network
and the speed at which various message would travel. This has been fixed by having all output send to the manager
for printing. The worker tasks no longer print the output themselves upon receipt of a token, but instead
send that data to the manager.
Platforms tested:
heping, eirene, tg-login (the only place that seems to still experience intermixed output every now and then)
Misc. update:
Fixed numerous ph5diff bugs.
Description:
Fixed manager output printing
Fixed out of order output printing
Fixed test script execution problem
Temporary fix for large amounts of output overflowing buffer.
Solution:
The manager task buffers its output. However, since the manager task
never gets a print token, this output was lost. Solution: new function called
print_manager_output that prints buffered output is called in places where the
manager buffers its output.
printf was apparently buffering output. This means that a task would sometimes
print even after it had given up its print token. Added fflush() call after
printf() calls, which seems to have fixed the problem.
calling rsh multiple times in succession seems to overwhelm something in Linux,
as it begins to refuse new connections until the old ones reset. Since each
call to mpirun in the test script starts up 4 rsh sessions, the test script
eventually is unable to run any further tests. Solution: Added a short delay in the
testscript between successive calls to mpirun to allow old connections to reset.
The 10k output buffer was of insufficient size to hold the large amounts of
output generated by some of the tests. Since code to buffer to a file has not
been implemented yet, a temporary fix was to increase the size of the output
buffer to 50k.
Platforms tested:
heping
Misc. update:
Bug Fix
Description:
Permanent fix for the incompatibilities between h5diff and h5repack.
Solution:
h5diff now contains the code to run both parallel and serial diffs.
Depending on how the binary is called, it will run either the serial or
the parallel versions respectively.
Platforms tested:
heping(serial + parallel), copper.
Misc. update:
fix bugs.
Description:
The three variables, g_nTasks, outBuff, and outBuffOffset, were
declared in ph5diff.h. That means they are declared in various
*.c files that include it, thus having the appearance of multiple
defined. Most C compilers are forgiving of this error but
the g++ rejects this practice.
Solution:
Make it only an extern declaration in ph5diff.h but really
defined them in h5diff_util.c
Platforms tested:
sleipnir using g++ since this only failed in Daily test.
Also testes in heping, serial and parallel modes.
Temporary daily tests fix.
Description:
The newly added ph5diff is not quite compatible with the h5repack tool.
Disabled the ph5diff build (reverted back to building serial h5diff)
to allow daily tests to work while ph5diff is fixed.
Platforms tested:
Tested in heping both serial and parallel modes.
Misc. update:
Bug fix: Temporary fix for h5repack failures in all parallel builds.
Description:
The parallel additions to h5diff interfered with h5repack.
Solution:
Added a second set of "parallel" functions to h5diff.c. h5repack uses the serial versions, whereas h5diff will use the parallel versions.
Also, h5diff will now be smart about when to enter parallel mode. If is run with mpirun with more than 1 task, it will enter parallel mode. Otherwise, it will stay in serial mode as before.
Platforms tested:
heping (serial and parallel)
Misc. update:
New feature.
Description:
Added new tool ph5diff. (Code done by Leon Arber.)
Code is changed but test is not working yet. For now,
it skipped all tests.
Platforms tested:
Tested in heping, serial and parallel modes.
