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:
Bug fix & new tests
Description:
Fix another couple of issues with record removal and add some more tests to
make certain removals work correctly.
Platforms tested:
FreeBSD 4.11 (sleipnir)
Solaris 2.9 (shanti)
Description: This is the 2nd step of changing conversion test. This checkin is
only for conversion from integer to floating number. The source buffer of
integer is filled up in the same way described in the last checkin(revision
1.153).
Platforms tested: fuss, sol, and heping - copper was down. Couldn't do
h5committest.
Bug fix & new tests
Description:
Fix several errors in record removel routines, which should now be feature
complete for removing records at any location in the B-tree (further testing
will verify this)
Platforms tested:
FreeBSD 4.11 (sleipnir)
Solaris 2.9 (shanti)
New feature.
Description:
The nbit filter does not support situations where datatype of string,
time, enum, variable-length, bitfield, opaque, reference is a field of
a compound datatype or array datatype.
Solution:
Modify the nbit filter to hanle such situations. The filter is a no-op
to these dataytypes.
Platforms tested:
copper, heping.
Misc. update:
Forgot to run reconfigure
Description:
Forgot to run bin/reconfigure after removing my debug output from configure.in.
Shame on me!
Platforms tested:
mir
Bug fix
Description:
Corrected errors in record removal code to handle removing records in
higher nodes in the B-tree.
Platforms tested:
FreebSD 4.11 (sleipnir)
Solaris 2.9 (shanti)
Automake version upgrade
Description:
Upgraded automake version from 1.6.2 to 1.9.5.
Changed bin/reconfigure script to use automake 1.9.5.
Changed configure.in and Makefiles to use new FCFLAGS and FC variables
instead of FFLAGS and F9X.
Automake and configure should now do the lion's share of the work
supporting Fortran 9X; macros in acsite.m4 are now mostly unused (will
be cleaned later).
Altered how configure handles pmake; now root-level Makefile.in is
processed by bin/reconfigure to have a .MAKEFLAGS target, since
automake no longer allows us to define unused variables.
Configure now always checks for C++ compiler even if it is not
used, since automake thinks this is the Right Thing To Do and
will break otherwise.
Platforms tested:
Sol, copper, heping, mir, sleipnir, eirene, pommier, kelgia, modi4.
Bug fix & new feature
Description:
Allow B-tree's height to be reduced when removing records.
Platforms tested:
FreeBSD 4.11 (sleipnir)
Solaris 2.9 (shanti)
More tests
Description:
Add tests for removing records in the root of a level-1 B-tree and
promoting and merging leaf nodes.
Platforms tested:
FreeBSD 4.11 (sleipnir)
Solaris 2.9 (shanti)
Bug fix/new feature
Description:
Tweak the record promotion algorithm to get it working correctly when
promoting and redistributing records.
Added tests for that case.
Platforms tested:
FreeBSD 4.11 (sleipnir)
Solaris 2.9 (shanti)
New feature
Description:
Allow records in internal nodes to be removed, not just records in leaf
nodes.
Platforms tested:
FreeBSD 4.11 (sleipnir)
Solaris 2.9 (shanti)
Bug fix & new feature
Description:
Fix error in 3-node redistribution when nodes are only moving into the
middle node from the left & right nodes (which happens sometimes during
record removals).
Clean up internal insert & remove routines to remove lots of redundant
checking.
Added 3->2 node merge routine to handle more record removal cases.
Platforms tested:
FreeBSD 4.9 (sleipnir)
Solaris 2.11 (shanti)
Description: This is the first step of changing conversion test. This checkin is only for conversion
between integers. Instead of filling source buffer with randomly generated bit pattern, this new way
uses more sensible bit patterns. It's easier to see using "char" as source for example. The bit patterns
will be
00000001, 00000010, 00000100, 00001000, 00010000, 00100000, 01000000, 10000000
00000000, 00000011, 00000111, 00001111, 00011111, 00111111, 01111111, 11111111
11111111, 11111110, 11111100, 11111000, 11110000, 11100000, 11000000, 10000000
The main point of this way is to avoid casting and comparison between source and destination types by
compiler. The bit patterns will cover positive maximum and minimum, negative maximum and minimum if
the source is signed integer.
Platforms tested: h5committest and fuss.
Bug fix & new feature
Description:
Fix a couple of off-by-one errors in assertions (code was actually correct)
for 3 node redistributions.
Remove "old" node removal code that is unused now.
Add more tests that verify that 2-node and 3-node redistributions are
working correctly for removals.
Platforms tested:
FreeBSD 4.11 (sleipnir)
Solaris 2.9 (shanti)
New test
Description:
Add a couple more tests for removing records before I start modifying the
code further.
Platforms tested:
FreeBSD 4.11 (sleipnir)
Solaris 2.9 (shanti)
Bug fix
Description:
Filter callback routines were being called with memory "version" of disk
datatype (don't ask... :-), which could result in incorrect calculations in
the filter callbacks.
Solution:
Rearrange code to call the filter callbacks after the final disk "version"
of the disk datatype has been created.
Platforms tested:
FreeBSD 4.11 (sleipnir)
Too minor to require h5committest
Bug fix
Description:
Add some "safety net" sort of assertions and sanity checks contributed by
user.
Platforms tested:
FreeBSD 4.11 (sleipnir)
Too minor to require h5committest
Bug fix
Description:
The GASS VFL driver header file was bringing in the <string.h> header file,
which several other source code modules needed also, but weren't including
explicitly themselves.
Solution:
Add includes for <string.h> to files which actually need them.
Platforms tested:
FreeBSD 4.11 (sleipnir) w/C++ as CC
Configuration not tested by h5committest...
Bug fix.
Description:
MPI_LONG_LONG, which is not standard yet, was used to pass the
nubmer of differences found. This was needed because number of
differences is defined as type hsize_t which can be arbitary large
such that there is no MPI type that matches it. The value is passed
between processes as an array of bytes in order to be portable. But this
may not work in non-homogeneous MPI environments.
This fix was actually Leon's idea.
Platforms tested:
Tested in QSC in which this failed.
Bug fix & new feature
Description:
Fix problem with inserting existing keys into B-tree corrupting record
counts along the path to the failed insertion.
Add more support for removing records, it's now handling removing records
from leaves of level-1 B-trees.
Platforms tested:
FreeBSD 4.11 (sleipnir) w/parallel
Solaris 2.9 (shanti)
Bug fix
Description:
When C flags are passed in using $CC variable, this can confuse test
commands in configure (for instance, CC='pgcc -tp k8-32').
Solution:
Previous solution didn't entirely solve problem. Wrote a sed snippit
to strip flags from $CC for purposes of test in configure.
Platforms tested:
mir (just a configure bug)
Feature--to provide a standalone mode for t_mpi.c so that it can
be built outside of PHDF5 environment.
Description:
Move definitions that are common to all parallel test programs
to a new header file called testpar.h.
Leave only Parallel HDF5 tests related definitions in testphdf5.h.
Platforms tested:
heping (pp) and modi4(PP). Copper was down.
Misc. update:
Removed GASS configuration and software packages associated with
it. The following software are no longer configurable.
checking for main in -lcrypto... yes
checking for SSL_get_version in -lssl... yes
checking globus_common.h usability... yes
checking globus_common.h presence... yes
checking for globus_common.h... yes
checking for globus_module_activate in -lglobus_common... yes
checking for main in -lglobus_gass_cache... yes
checking for main in -lglobus_gaa... yes
checking for main in -lglobus_gss... yes
checking for main in -lglobus_gss_assist... yes
checking for main in -lglobus_io... yes
checking for main in -lglobus_gass_transfer_assist... yes
checking for main in -lglobus_gass_transfer... yes
checking for globus_gass_open in -lglobus_gass_file... yes
Platforms tested:
h5committested (but copper was down.) tested in modi4 too.
typo fix and small improvement.
Description:
t_coll_chunk.c:
ccdataset_vrfy() was using a wrong routine name to identify itself.
testphdf5.c:
Add a definition of NFILENAME to be the common dimension size of
FILENAME[] and filenames[][] since they must have the same first
dimension size.
Platforms tested:
h5committested.
Bug fix.
Description:
"testphdf -p" would with data verification errors. The reasons were
that the MPIPOSIX driver file open and close, especially the close
routine provide no "coordination" between processes. The testphdf5
tests reuse the same file for test data file by opening using H5Fcreate
with the HDF5_FCC_TRUNC option. The test routines do not provide any
code to ensure that all processes have finished one test before moving
to the next test. Some "faster" process would have finished verifying
its portion of data as correct and move to the next test which opens
the same file with TRUNCATOIN which truncates the previous data file.
But some "slower" processes are still verifying the "previous" data
file which all of a sudden got truncated by the "faster" process.
Solution:
Technically, the test program should be fixed to ensure all processes
have finished one test before any is allowed to move to the next test.
OTOH, the MPIO VFD has no problem with this test because MPI-IO requires
File open and close be called collectively correct and ensure it is
returned properly.
I choose to fix the MPIPOSIX close routine to provide some sort of
coordination between processes by requiring all processes to have
completed the close of a file before it is returned to user space.
This makes the MPIPOSIX close routine behaves more like the MPIO
close routine, thus provide more protection for user applications
that fail to code in the coordination. But having the barrier
in the MPIPOSIX close routine would penalize applications where
it is "okay" for some processes to close its file handle and race
ahead to do other things since it is not going to access this file,
therefore whether other processes are still using the file is immaterial.
Maybe this protective coordination should be optional and can be turned
off by confident users who need not this sort of protection.
Platforms tested:
"h5committested" and tested in modi4 and tesla.
Misc. update:
Bug fix
Description:
Libtool simply doesn't seem to support shared libraries for pgcc, but
tries to build them anyway.
Solution:
Edited configure to disable shared libraries in libtool script when
pgcc is being used.
This is a more drastic fix than the problem requires, but it is
difficult to detect when this problem will occur. There is
discussion online about shared libraries with pgcc, so hopefully
libtool will support pgcc and this fix will become unneccesary.
Platforms tested:
sol, copper, mir, sleipnir
Misc. update:
Description: Condition check of macro H5_ULLONG_TO_FP_CAST_WORKS was accidentally
taken out in last revision. This macro is defined for Windows because Windows
doesn't support conversion from "unsigned long long" to any floating-point type.
Solution: Added it back to H5T_conv_ullong_xxx() functions(xxx is any of "float",
"double", or "long double").
Platforms tested: tested on Windows.
Bug fix
Description:
A test in configure.in redirected output in a way that made
mir (at least) unhappy (test: too many arguments).
Solution:
Changed the semantics so that mir was happy. This didn't seem to make any
other platforms unhappy, either.
Platforms tested:
eirene, modi4, copper, mir, sleipnir (only tested configure)
bug fix
Description:
H5DSget_scale_name, H5DSget_lablel return wrong value when truncating to 1
character array.
Solution:
fix it.
Platforms tested:
verbena,shanti,copper
Misc. update:
Bug fix.
Description:
The nbit filter does not check error when calling some utility functions
that can return a failure.
Solution:
Add error handling for these utility functions.
Platforms tested:
heping, copper
Misc. update:
Bug fix
Description:
C++ tests would die with strange linking-type errors.
Solution:
C++ build needs to be static, too. Applied -static flags liberally in
c++/src and c++/test directories.
Platforms tested:
sleipnir, heping
Bug fix
Description:
Fortran builds were dying with unknown flags.
Solution:
Fortran builds should be static, not shared. Static flags had been specified
for libraries, but apparently were also needed for executables.
Solved problem by adding static flag to everything in fortran directories.
Platforms tested:
mir, eirene
Purpose:
Bug fix
Description:
Libtool 1.5.14 added some extra flags that needed to be passed from the compiler
to the linker. icc did not handle this correctly and complained when it saw
unknown flags.
Solution:
Set the -Wl flag correctly on freebsd so that icc knows to pass flags through
to the linker.
Platforms tested:
sleipnir (gcc, g++, icc, gcc40)
Bug fix
Description:
h5vers would spit out the output that reconfigure script produced.
Solution:
Sent output from reconfigure to /dev/null. h5vers is now quiet as before.
Platforms tested:
verbena, eirene, heping
