Change Summary for MPICH2 for week of Nov 06, 2005

This list was automatically generated on Mon Nov 7 05:12:21 2005 .

Information on all weekly change summaries is available.

List of revisions

Files for each revision

add rc to spawn_result src/pm/mpd/mpdman.py
added HAVE_INET_PTON case winconfigure.wsf
added a test to test concurrent spawn calls test/mpi/spawn/Makefile.sm test/mpi/spawn/concurrent_spawns.c test/mpi/spawn/testlist
Added another error test test/mpi/configure.in test/mpi/errors/Makefile.sm test/mpi/errors/testlist test/mpi/errors/coll/noalias.c test/mpi/errors/pt2pt/proberank.c
Added ch3_pre_init option src/mpid/ch3/src/mpid_init.c
added error checking src/mpi/romio/adio/ad_ntfs/ad_ntfs_done.c src/mpi/romio/adio/ad_ntfs/ad_ntfs_fcntl.c src/mpi/romio/adio/ad_ntfs/ad_ntfs_iread.c src/mpi/romio/adio/ad_ntfs/ad_ntfs_iwrite.c src/mpi/romio/adio/ad_ntfs/ad_ntfs_open.c src/mpi/romio/adio/ad_ntfs/ad_ntfs_read.c src/mpi/romio/adio/ad_ntfs/ad_ntfs_write.c
added error checking and removed unused locking macros src/mpid/ch3/src/mpidi_kvs.c
Added explanation of MPICH name doc/faq/faq.txt
Added finalize callback for datatypes to free the pairtypes allocated atinit time. This removes all the reachable memory reported by valgrindin a init/finalize program. There are still some blocks of unreachablememory that were allocated in CH3. This corresponds to MPICH2 Req#1497. src/mpi/datatype/typeutil.c
added flush statements test/mpi/cxx/util/mtest.cxx
Added g95 for 64-bit fix doc/faq/faq.txt
added getNodeRootRank to determine which process should initialize the bootstrap queue src/mpid/ch3/util/shm/ch3u_init_sshm.c
Added installcheck test to checkbuilds maint/checkbuilds.in
Added local testing target test/mpi/attr/Makefile.sm
Added missing default f90 branch and update version number configure.in maint/Version
Added missing stdarg.h declaration src/mpi/romio/adio/common/strfns.c
Added missing test for ctype.h src/pm/smpd/configure.in
added more detail to --ncpus option in Q on where processes run doc/faq/faq.txt
Added more old code to archive src/mpid/ch3/channels/ssm/src/archive.txt
added new MPE functions. src/mpe2/mpe.def
Added new test that creates 1024 indexed types with 1024 blocks each, toverify that we can indeed allocate a lot of types without error. Wethen send/recv/get_count/get_elements using each type (short sends tokeep runtime short). In response to MPICH2 Req #657. test/mpi/datatype/Makefile.sm test/mpi/datatype/lots-of-types.c test/mpi/datatype/testlist
Added new test that performs send/recv with a struct with a zero-blklenleading element and tests get_count/get_elements results. This testwould cause a FPE with versions of MPICH2 prior to yesterday andcorresponds with MPICH2 Req #1780. test/mpi/datatype/Makefile.sm test/mpi/datatype/struct-empty-el.c test/mpi/datatype/testlist
added pmi error names src/mpi/errhan/errnames.txt
Added special handling for CPPFLAGS in immediate subdir configure configure.in
Added specific warning message about Fortran90 and FOrtran 77 compilers incompatibility configure.in
Added support for disabling spawn tests as a first step in cleaning up the tests for the devices and channels that do not support the dynamic process routines test/mpi/configure.in test/mpi/testlist.in
Added test for alignment error for get_attr functions to give better error messages when the user uses an int instead of the address of an int in those routines src/mpi/attr/attr_get.c src/mpi/attr/comm_get_attr.c src/mpi/attr/type_get_attr.c src/mpi/attr/win_get_attr.c src/mpi/errhan/errnames.txt
Added test that alias errors are detected test/mpi/errors/coll/.cvsignore test/mpi/errors/coll/Makefile.sm test/mpi/errors/coll/noalias.c test/mpi/errors/coll/testlist
Added tests for headers and inet_pton function src/mpid/ch3/configure.in src/mpid/ch3/util/sock/ch3u_connect_sock.c
added windows spawn question doc/faq/faq.txt
Began adding support for singleton init to gforker mpiexec and pm/util . Still need documentation on what singleton init expects in terms of ports and wire protocol before this can be finished src/pm/gforker/mpiexec.c src/pm/util/cmnargs.c src/pm/util/pmiport.c src/pm/util/pmiserv.c src/pm/util/pmiserv.h src/pm/util/pmutil.h src/pm/util/process.c
Better handling of mpdallexit, esp for large numbers of nodes (e.g. 50+) src/pm/mpd/mpd.py
Change handling of optional directory installs to avoid getting a warning message about a successfully ignored error from some implementations of make maint/simplemake.in
clean up casts src/mpid/common/sock/poll/sock_post.i
Cleaned up one commend and added a fixme about invalid hostname in singleton init args src/pmi/simple/simple_pmi.c
Commented out the trid call that suppressed warnings about memory leaks occuring from memory allocated during MPI_Init src/mpi/init/initthread.c
corrected local memory allocations src/mpid/ch3/util/shm/ch3u_init_sshm.c
corrected logic src/mpid/ch3/util/shm/ch3u_init_sshm.c
Corrected names on func exit macros src/mpid/ch3/channels/ssm/src/ch3_shm_memory.c
data is contiguous, but type specifies that data be processed in anoncontiguous ordering. Bug in struct was reported in MPICH2 Req #2040;I guessed (correctly) that the same problem would exist in vector codeand fixed that too. src/mpid/common/datatype/mpid_type_vector.c
detect extraneous % in continutation line doc/faq/faqbuild
Ensure filename available for error message maint/getcoverage.in
Fix newly created problem regarding MPICH_INTERFACE_HOSTNAME src/pm/mpd/mpd.py src/pm/mpd/mpiexec.py
Fixed bug in debug printing of contents of indexed types. Oops! src/mpid/common/datatype/mpid_type_debug.c
fixed order of params to MPIU_Strncpy src/mpid/ch3/src/mpid_init.c
fixed test for integer vs pointer alignment in get_attr routines src/mpi/attr/attr_get.c src/mpi/attr/comm_get_attr.c src/mpi/attr/type_get_attr.c src/mpi/attr/win_get_attr.c
Fixed things that IA-64's gcc does not like. src/mpe2/src/wrappers/test/iotest.c
Fixes to deal with inappropriate use of "is_contig" optimizations whendata is contiguous, but type specifies that data be processed in anoncontiguous ordering. Bug in struct was reported in MPICH2 Req #2040;I guessed (correctly) that the same problem would exist in vector codeand fixed that too. src/mpid/common/datatype/mpid_type_struct.c src/mpid/common/datatype/mpid_type_vector.c
Fixes to deal with zero block length elements in structs. Includes alittle improvement to the type debug output functions. src/mpid/common/datatype/gen_type_struct.c src/mpid/common/datatype/mpid_type_debug.c src/mpid/common/datatype/mpid_type_struct.c
handle concurrent spawn commands using a delayed spawn queue src/pm/smpd/smpd.h src/pm/smpd/smpd_connect.c src/pm/smpd/smpd_handle_command.c src/pm/smpd/smpd_handle_spawn.c
Improved test to more quickly identify how we've messed up pairtypes. test/mpi/datatype/pairtype-size-extent.c
Improved the PAC macro that determined alignment of items within structures confdb/aclocal_cc.m4
initialized pmi_errcodes to 0 src/mpid/ch3/src/ch3u_comm_spawn_multiple.c
leading element and tests get_count/get_elements results. This testwould cause a FPE with versions of MPICH2 prior to yesterday andcorresponds with MPICH2 Req #1780. test/mpi/datatype/struct-empty-el.c test/mpi/datatype/testlist
little improvement to the type debug output functions. src/mpid/common/datatype/mpid_type_debug.c src/mpid/common/datatype/mpid_type_struct.c
Make sure AC_OUTPUT_COMMANDS are always executed configure.in
Make sure that we get largestorword for alignment on Power confdb/aclocal_cc.m4
misc src/mpid/ch3/channels/sock/src/ch3_progress.c src/mpid/ch3/util/sock/ch3u_connect_sock.c
missing variable declaration src/mpid/ch3/util/shm/ch3u_init_sshm.c
Modified USERBUFFER macro to allow NULL buffer pointers when thecorresponding datatype is of zero size. Response to MPICH2 Req #1347. src/include/mpierrs.h
NULL pointer in those cases. Also verifies that we're dealing withhindexed types with zero blks in the middle, including receivingpartials and verifying both data correctness and get_count/elementsreturn values. test/mpi/datatype/testlist
noncontiguously. Versions are written for struct, vector, and indexedtypes. Prior to today, we would fail the struct and vector versions ofthe test. This is in response to MPICH2 Req #2040. test/mpi/datatype/testlist test/mpi/datatype/unusual-noncontigs.c
ran update_windows_version for the 1.0.3 release maint/mpich2i.vdproj
removed bug, ";". src/mpe2/src/wrappers/test/fcomm1_isr.f
renamed isnumber to smpd_isnumber src/pm/smpd/mp_parse_command_line.c
set initialize_queue to 1 to fix hanging bug src/mpid/ch3/util/shm/ch3u_init_sshm.c
Setup default ifhn for MPICH src/pm/mpd/mpiexec.py
Simplified the output of the installcheck target. src/mpe2/src/collchk/test/Makefile.in src/mpe2/src/graphics/contrib/test/Makefile.in src/mpe2/src/wrappers/test/Makefile.in
state name typo src/mpid/ch3/src/mpidi_kvs.c
Switch to AC_COMPILE_IFELSE from AC_TRY_COMPILE to work around an outstanding and apparently never-to-be fixed bug in the autconf definitions for AC_LANG_FORTRAN77 configure.in
Switch to AC_{LINK,COMPILER}_IFELSE to work around bugs in teh Fortran support for AC_TRY_{LINK,COMPILE} confdb/aclocal_f77.m4 confdb/fortran90.m4
Take steps toward supporting singleton init with gforker src/pm/gforker/mpiexec.c src/pm/util/cmnargs.c src/pm/util/process.c src/pm/util/process.h
Test creates types that have contiguous data but process that datanoncontiguously. Versions are written for struct, vector, and indexedtypes. Prior to today, we would fail the struct and vector versions ofthe test. This is in response to MPICH2 Req #2040. test/mpi/datatype/Makefile.sm test/mpi/datatype/testlist test/mpi/datatype/unusual-noncontigs.c
Test designed to help verify that we deal appropriately with datatypesthat are of zero length and don't return an error if the user passes aNULL pointer in those cases. Also verifies that we're dealing withhindexed types with zero blks in the middle, including receivingpartials and verifying both data correctness and get_count/elementsreturn values. test/mpi/datatype/Makefile.sm test/mpi/datatype/hindexed-zeros.c test/mpi/datatype/testlist
that are of zero length and don't return an error if the user passes aNULL pointer in those cases. Also verifies that we're dealing withhindexed types with zero blks in the middle, including receivingpartials and verifying both data correctness and get_count/elementsreturn values. test/mpi/datatype/hindexed-zeros.c test/mpi/datatype/testlist
then send/recv/get_count/get_elements using each type (short sends tokeep runtime short). In response to MPICH2 Req #657. test/mpi/datatype/testlist
types. Prior to today, we would fail the struct and vector versions ofthe test. This is in response to MPICH2 Req #2040. test/mpi/datatype/unusual-noncontigs.c
typo src/pm/smpd/smpd_handle_spawn.c
Updated for fixes to intercomm_create/merget and to sock/ssm RELEASE_NOTES
updated install manual for 1.0.3 release doc/installguide/install.tex
Updated release number README
updated release number doc/userguide/user.tex
updated stdin redirection for Macs using pthreads src/pm/smpd/mpiexec.c src/pm/smpd/smpd.h src/pm/smpd/smpd_connect.c src/pm/smpd/smpd_do_console.c src/pm/smpd/smpd_handle_command.c src/pm/smpd/smpd_state_machine.c
Updates to design doc doc/design/design.tex
Various little fixes related to coverage testing. src/mpi/datatype/get_count.c src/mpi/datatype/get_elements.c src/mpi/romio/adio/ad_ufs/ad_ufs_open.c
verify that we can indeed allocate a lot of types without error. Wethen send/recv/get_count/get_elements using each type (short sends tokeep runtime short). In response to MPICH2 Req #657. test/mpi/datatype/lots-of-types.c test/mpi/datatype/testlist
When supporting 'root' mode, mac OSX does not like setreuid, so nowwe try: setuid as well src/pm/mpd/mpd.py
would cause a FPE with versions of MPICH2 prior to yesterday andcorresponds with MPICH2 Req #1780. test/mpi/datatype/testlist