Change Summary for Revisions between Rmpich1_2_6 and current

List of revisions

• "Stefano TLC185" <stefano@tlc185.com> pointed out that we were not updatingfp_ind correctly in some cases.
• a better comment about our resize assumptions
• ADIOI_EXPLICIT_OFFSET -> ADIO_EXPLICIT_OFFSET
• ADIOI_INDIVIDUAL -> ADIO_INDIVIDUAL
• add another test for working aio that assumes two-argument aio_write (oldversions of AIX implement aio this way).
• add preprocesor magic to prevent multiple instances of MPIOI_ routines when thecompiler (e.g. intel compiler) does not support weak symbols
• Add support for catching additional signals
• added --disable-f90
• Added -f to trap statement for rm mpif.h
• Added -kb version of usersguide.pdf so it can be checked out correctly under windows.
• Added .libstamp* to .cvsignore to handle new method of ensuring that files are added to the libraries even when Make does not consistently compute dependencies based on time stamps (that fix handles the problem of make building an object file but then not including it in the files-built-because-of-dependency variable)
• added BEGIN/END ERROR CHECKING comments
• Added CXXFLAGS to C++ compile tests incase the user is using the flags to change the behavior of the compiler
• Added CategoryRatios.java
• Added DOCTEXT
• Added DOCTEXT targets
• added LineIDMap.toYCoordMap() method. Together with constructor LineIDMap( YCoordMap ), the method provides a forward and backward translation between LineIDMap and YCoordMap.
• Added MPICH_SCRIPT_ECHO check and a minor fix
• Added PAC_PROG_CHECK_INSTALL_WORKS after AC_PROG_INSTALL to make sure INSTALL actually work. If not, use install-sh, so libtool --mode=install works on solaris
• Added a VERSION_INFO string to identify the launcher's version and more informative messages about the change of values in the user's setup file.
• Added a check for null format in message
• Added a comment about reliable signal delivery
• Added a comment about the reason for EXEEXT
• Added a fix for node name leaders
• Added a footnode in userguide to explain why exclusion ratios are a tag less than what they should be. Similarly a note in FAQ as well.
• added a reminder to myself to be sure to run autoheader.
• added ac_prereq 2.59
• Added additional generated machinefile names
• Added an #ifdef protected definition for int64_t (but no corresponding configure test is available)
• Added an additional comment about int64_t
• Added breaks to make code more robust in the event that p4_error returns
• Added calls to recover SYSV IPC and other resources on error shutdown
• added changes to support calling MPI_Connect/Accept from F77
• Added check for local version of tarch (for stand-alongetestsuite) and for preexisting io directory (also for the standalone testsuite)
• added check of HAVE_STDIO_H and HAVE_STDLIB_H to the compile test in JAC_JNI_HEADERS
• added code to free registered datarep if any
• Added code to try and determine the autoconf include dir switch
• added code to verify that messages were matched properly
• added comment
• Added comment about atomiticy test - current runtest uses fewer processes than the test itself recommends
• added comment about using the --annotate option for used with Emacs and GdbMode
• Added configure cache to cvsignore
• Added configure test to check for leftover Makefiles in subpackages that is called by AC_CONFIG_SUBDIRS. If leftover Makefile is found, "make distclean" is invoked.
• Added configure.lineno as ignored file (cygwin autoconf produces this file)
• added constant color used by ShadowCategory
• Added debug information about killing processes
• Added display properties entry in Preference Window for Event drawables, and added InfoPanel support for Event drawables.
• Added display support for Event drawable.
• Added distclean target
• Added distclean, maintainer-clean and owner-clean targets
• Added export of F77 and FC to make sure that any set environment variables are seen by all configure steps
• Added extra message to clarify the reason to use JPanel instead of JFrame. Removed jar and class files that were checked in without -kb.
• Added formal definition of ShadowCategory which is a subset of ShadowCategory.
• added generalized request routines to the MPI to PMPI mapping
• Added ifdef for TIOCNOTTY, which Cygwin may not support
• Added logic so the OK button in ConvertorPanel is enabled only when logconversion process is returned successfully and the logconversion is not stopped by user manually and the button won't be presented or created when ConvertorPanel is used standalone, i.e. in a JFrame. In order to support this SwingProcessWorker is added a new function to tell the underlying java process's return status.
• Added logic to avoid redundant warning messages, "Null Pathname" and/or "Null input file", when user clicks on the Cancel button in the ConvertorDialog.
• Added max_port_name
• Added missing paren in coverage target for ROMIO
• Added missing prototypes and removed unused variables.
• Added missing semicolon
• Added more generated files to cvsignore
• Added new CategorySummary component member
• Added new CategoryWeightF which is similar to CategoryWeight except the number of real objects variable is double instead of long so that it can be used in dragging of statistics window which may grep partial drawables.
• Added new method summarizeCategories() which is intended for root treenode renormalizes all statistics in preview drawables to the CategoryMap.
• Added old/OLD to .cvsignore
• added prototypes for _FAKE functions
• added ranlib; changed ar crl to ar cr
• Added some support for DESTDIR
• Added some support for character in the choice module
• Added statement to mark -rsh argument as processed by the mpirun.ch_p4.args script
• added static IncreasingStarttimeComparator and IncreasingFinaltimeComparator private classes that make use of the corresponding classes in TimeBoundingBox.
• Added structured comments to wrap error handling for coverage purposes;minor cleanup.
• Added support for .exe files produced under cygwin
• added support for mpicxx
• Added support for shared libraries when weak symbols are not supported to ROMIO
• Added support for sleep under Windows
• Added support to allow "white space" in pathname to JVM and logfiles in invoking jumpshot_launcher.jar as well as in the GUI, especially the logfile convertor. In order to provide this support, Runtime.exec( String ) cannot be used. Instead a Runtime.exec( String[] ) is used. RuntimeExecCommand class is added to provide the necessary support to split the runtime execution command into appropriate string array without shell/runtime dependent delimiter, i.e. space.
• Added supported required by Slog2TOslog2, e.g. added InfoBox.releaseCategory() which undoes what InfoBox.releaseCategory() does.
• Added test for MPI_BUILD_PROFILING already defined
• Added test for XDR functions, not just the include files (cygwin may need an extra library)
• Added test that device setup is executable, and make it so if it is not
• Added tests that large file support works within the current compiler (check that the offset type is a scalar and not an array, as it is on Solaris in some cases)
• Added the support of the case when no command argument is supplied to various logfile convertors, i.e. print help message.
• Added updates for xml summary form
• Added various supports to allow slog2print to print (user) selectable feature of the slog2 file
• Added workaround for MPICH2 requirements in ROMIO so that ROMIO can be built with MPICH1. Error reporting, however, is now broken for ROMIO (errors print instead of returning a valid error code)
• alloc space for an entire PVFS_size type, not just a pointer
• Allow F77 as an alias for FC
• Allow findnosuccess2 to be used as a filter, reading from stdin
• Allow mpd to use secretword as an alias for password
• always build testfs.
• always have fd_direct, etc.
• and BufForShadows.
• Applied David's error code patch.
• at this point, but passes single-threaded MPI-IO tests.
• back out the changes suggested by stefano@tlc185.com for now
• big fat comment to explain how NFS_set_shared_fp differs from the GEN version
• bring error handling in line with current practices
• bugfix
• Build the mpich1 glue Makefile
• changed %Ld to %lld as format for long long printf because the Intel compiler doesn't like it
• changed BOOL to RLOG_BOOL
• Changed MPI_BYTE to use more general xdr routine in heterogeneous case to handle problems when the number of bytes is not a multiple of 4 (an xdr restriction)
• Changed the basic drawing of a Event and PreviewEvent.
• Changed the class hierarchy of CategoryWeight so that CategoryWeight is subclass of CategorySummary which extends CategoryRatios. With this change, CategorySummary can be used as a component in Category, so WriteObject() and ReadObject can be used as well.
• changes to support calling MPI_Connect/Accept from F77
• check all error codes
• Checkin corrected p4 configure
• Commit change to file system option for new Romio configure in mpich-1 configure
• Commited class and jar files as binary files.
• compiler (e.g. intel compiler) does not support weak symbols
• Complete the replacement MPICH and MPICH2 macro variables by MPI_IMPL. Replaced (cd .. ; ..) by (cd .. && .. ).
• Declared the datarep functions if necessary (needed for MPICH1)
• Defined TimeBoundBox.Order interface. Added 4 basic TimeBoundingBox.Order: INCRE_STARTTIME, DECRE_STARTTIME, INCRE_FINALTIME, DECRE_FINALTIME orders. Defined Drawable.Order based on the 4 TimeBoundingBox.Order's. Removed DrawOrderComparator class and DRAWING_ORDER is replaced by INCRE_STARTTIME_ORDER. These 4 Drawable.Order's form the foundation for the different iterators allowed by slog2.input.TreeTrunk.
• defined aiocb to aiocb64 in certain cases
• do not actually perform the seek. since we don't make a system call, don'tupdate fp_sys_posn.
• Do not override the choice of F90 compiler if one has been made; if FROM_MPICH2, use CC instead of mpicc (needed for LIBTOOL usage, and should work everywhere mpicc would)
• do not update fp_sys_posn. we only update the bookeeping in ROMIO (fp_ind) anddo not actually perform the seek. since we don't make a system call, don'tupdate fp_sys_posn.
• Enable the glue code for MPICH1
• Enhaced macro JAC_JNI_HEADERS so that it won't easily accept the include path that contains jni.h (through deduction of javah's location). Instead of using AC_TRY_CPP([#include <jni.h>...), a much stricter test that uses AC_TRY_COMPILE() is used. The test includes reference to jobject and JNIEnv as well as headers stdio.h/stdlib.h/jni.h. The goal is that GNU jni.h that is installed in /usr/include won't be accepted because /usr/include/linux contains conflicting headers that are in /usr/include.
• Enhaced the Shadow so that each CategoryWeight accounts for the number of drawables in each category.
• Ensure that old links are removed before creating new ones
• Ensure that shared library version is built
• error (don't warn) if pvfs2 support requested but cannot find pvfs2.h
• fix a small memory leak
• fix amazing thinko (writing instead of reading).
• fix an xfs compile error
• fix another case of full strings (instead of **whatever) in the error handler
• fix another small leak
• Fix default behavior for stdout-friendly output
• Fix for misspelling in #define for romioconf include
• Fix handling of the gnuplot options (when we switched to gnuplot by default, lost the option to select eps/ps/gif options unless selected cit first, e.g., -cit -gnuplotps)
• Fix misplaced generic error message and missing end-error-handling
• Fix misspelled header file name in test
• Fix test of logical flag output in Fortran I/O test
• Fix use of otherargs with with the configure in ch_p4 (was passing some of the arguments for the configure in ch_p4/p4)
• fixed ADIOI_Datatype_iscontig so that, for MPICH2, it returns the flag as noncontig when the datatype is otherwise contig but with a non-zero true_lb
• fixed Fortran compiles that didn't work on AIX
• fixed LIBNAME when called from within MPICH
• fixed ROMIO's profiling interface
• fixed alltoallv to not issue isend/irecv if it is 0 bytes
• fixed bug in alltoallv
• fixed bug in ifdef logic
• fixed bug in type_create_darray
• fixed bug reported by Jianwei, MPICH2 req #1174
• fixed bug reported in req 707
• fixed bugs reported by Tobias Eberle
• fixed docuementation
• fixed for builds from within MPICH2
• fixed misnamed variable
• Fixed missing character (rm was just m)
• Fixed missing dollar in test on MPI_AINT_SIZE
• Fixed prototypes
• Fixed prototypes for static functions, removed unused variables and initialized variables.
• Fixed prototypes, removed unused variables, fixed printf format for long variables.
• fixed race condition in test of persistent rsend
• fixed shlibname passed to ranlib
• Fixed the DESTDIR support so that the install objects do not know about DESTDIR and uninstall-package scripts in all sbin work correctly even after the installation is copied back from DESTDIR to $prefix. Using "dirname" in the install-package script, if trouble, use "sed+back reference" instead.
• Fixed the TreeTrunk.iteratorOfxxx() calls with new Drawable.Order arguments.
• Fixed the handling for alternate fortran compilers
• fixed things so that aio will compile for --enable-strict
• fixed typo
• fixed warnings. added #define aiocb aiocb64 where needed. fixed ifdef logic in ad_done.c for new IBM stuff
• Fixes for MPI_STATUS_IGNORE
• Fixes for filesystem with case-insensitive names
• fixes for out-of-tree builds.
• fixes for out-of-tree installs.
• Fixes for xdr tests
• Fixup directory for confdb in Fortran dir
• flag was an integer when it should have been a logical
• forgot to commit this on Friday; changed PAC_FUNC_STRERROR to AC_CHECK_FUNCS(strerror) so that MPICH2 and ROMIO define HAVE_STRERROR to the same value
• fp_ind correctly in some cases.
• Handle usernames with more than 9 characters and add tests for user names too long for the procgroup fields (instead of either overwriting other data or truncating the users name)
• If offset or aint kind are 0, use the default type size (e.g., 4 or 8)
• implement DISKSPACE, needed for MPI_File_preallocate
• Improve code to find the Fortran 90 kind for 8-byte integers to handle compilers that do not accept .f as the extension for Fortran 90 programs
• Improve shared library target
• Improve the code to select a Fortran 90 compiler
• Increased the loop count when trying to discover the size of Wtick
• initalize 'writebuf' to make valgrind happy
• Initial deposit of CLOG-2.1 code base. Sync the jar files with java source.
• initial stab at making MPI-IO routines threadsafe. Not very fine granularityat this point, but passes single-threaded MPI-IO tests.
• initialize shared_fp to zero, otherwise it will be garbage in the case whereshared fp is not null
• Let SOWING_CVSROOT set the source for the sowing CVS repository
• MPIOI_CHECK_INTEGRAL_ETYPE -> MPIO_CHECK...
• Made all logfile convertors consistent in terms of the definition of all ShadowCategories by using the formal definition of ShadowCategory from Category class.
• made changes to handle asynchronous I/O on AIX
• Made the verbose option of JARing process optional. The verbose option of JARing is available through setting JARFLAGS=v during make, i.e. make JARFLAGS=v.
• Make a copy of mpif.h with Fortran 90 comments instead of Fortran 77 comments
• Make sure that CFLAGS is substituted
• Make the cygwin autoconf do the right thing
• Make the tests for aio support more demanding to catch problems caused by different levels of runtime library support in the C compiler (e.g., aio may require POSIX levels to be defined)
• merged in RobR's hacking branch
• merged the wrong parts of HEAD and branch. all better now.
• merging of patch integration, etc.
• minor cleanup.
• minor fix to description of wtick
• Minor fix to man page description
• missed another spot where i needed to add a label
• Modified the info popup panel to show the Topology of the drawable.
• modify curr_index in the DUP case
• move ADIO_GEN_Close to its own file. should make NTFS happier.
• Moved end-error-handling
• now include pvfs2.h if available so we can autodetect pvfs2 files as well
• ok, for real this time: update fp_ind correctly. Fixes a b_eff_io errorreported by Stefano TLC185 <stefano@tlc185.com>
• on some platforms (Darwin) AIO headers are in sys/aio.h
• ROMIO no longer automatically aborts on errors. check retrun code of *all*MPI-IO routines. Needs to be done with every test in this directory...
• Rebuild some configures to use better detection of a change in the environment when using the configure cache
• remove excess comma
• Remove some warnings and fix some uninitialized data
• remove unused variable
• remove use of autoconf version 1 arguments passed to ROMIO for longlongsize
• removed double check for two-argument aio_suspend
• removed erroneous comments
• Removed extraneous file
• Removed mkdir of sysconfdir (rarely used?)
• removed mpich2 include that isn't necessary.
• removed ranlib shlib
• Removed stray tabs from Fortran test files
• removed the autom4te.cache dirs
• removed unused variables
• Removed usersguide.pdf that cannot be checked out under windows.
• replace BOOL with RLOG_BOOL
• Replace MPICH and MPICH2 macro variables by MPI_IMPL so src/Makefile.in and sbin/mpeinstall.in can use MPI_IMPL. Replaced (cd .. ; ..) by (cd .. && .. ).
• replace duplicated code with a common (naive) implmentation of preallocate
• Replaced DrawOrderComparator class by Drawable.DRAWING_ORDER.
• Replaced ForeItrOfxxx and BackItrOfxxx by the generalized ItrOfxxx that makes use of the just added Drawable.Orders. Added consistence check if the TreeTrunk(i.e. TreeFloorList) is created with different Start/Final Time order than requested Drawable.Order in the iteratorOfXXXX methods. Added reorderDrawables() method so TreeTrunk.iteratorOfxxx() can be called with INCRE_FINALTIME_ORDER and DECRE_FINALTIME_ORDER.
• Replaced IteratorOfBackDrawablesOfAll and IteratorOfForeDrawablesOfAll by the generalized IteratorOfAllDrawables that makes use of the newly added Drawable.Order class. Defined TreeNodeID.Order and BufForObjects.Order based on the correspinding Drawable.Order. Elimated the Comparable interface and explict uses those classes by used with the appropriate Comparator, i.e. Drawable.Order and BufForObjects.Order. Added reoderDrawables() method in TreeNode, BufForDrawablesand BufForShadows.
• Replaced JFLAGS by JSFLAGS and then used JSFLAGS and JFLAGS in .java.class implicit rule so make can take JFLAGS, e.g. "make JFLAGS=-g" works as expected.
• Replaced base.drawable.CategoryWeight by base.statistics.CategoryWeightF so that they account for the fractional number of real drawables.
• Replaced base.statistics.CategoryWeight by base.statistics.CategoryWeightF
• Replaced by the system.err.println() by runtime exception.
• Replaced the Return button in the Convertor by OK and Cancel buttons. OK means OK to display the last converted slog2 file. Cancel means Cancel any display and exit the convertor.
• Replaced the old CategoryWeight by a new class which extends CategoryRatios to include an extra variable that accounts for the number of real drawables in each category
• reported by Stefano TLC185 <stefano@tlc185.com>
• Reset execute bit on tarch if the file is present but not executable
• RobL's subarray fix, slightly improved by me.
• set atime/mtime/ctime on PVFS2 files.
• Set firstcall to zero after first call
• set fp_ind to 'off', not 'offset'. seek converts offset (units of etyperelative to filetype) into 'off' (offset into the actual file).
• shared lib updates; untested because of lack of sgi platform.
• Switch to using just the formatting name (not including the %, which complicates using configure since it uses sed with % as a separator) for formmating MPI_Offset values
• Sync prebuilt jar files with SLOG-2 2.0.6.
• Sync with source code change.
• Sync with source modification that change MPE_MAX_STATES to support logging of MPI-RMA calls.
• Sync with the source code update. Added slog2filter.jar.
• Sync with the source of added Event drawable support
• the MPIO_CHECK_* macros goto fn_exit, so add the label
• try to commit a meaningless change to romio/confdb/install-sh to see if that might change its permissions back to executable
• turned off top aware MPI_Alltoall and resorted back to default; also fixed it so that TCP is no longer polled with MPI_SOURCE_ANY on comm that _only_ speaks vMPI
• typo
• Update install notes and README that includes the new slog2filter utility program. Updated configure.in to build slog2filter.
• update makefiles for shared library support
• Update the Makefiles to use $PMPILIBNAME instead of p$MPILIBNAME for the profiling library name and add symbols needed to support libtool
• Updated SLOG2 version from 2.0.5 to 2.0.6
• Updated clog2 version from 2.1 to 2.2. Added support for CLOG2 related tools to enter a compatible mode for older versions.
• Updated comment to reflect exact limitation on buggy autoconf 2.13
• updated comments
• Updated cvsignore files for inplace builds
• updated error case for gridftp not installed
• updated for lack of std directory containing correct output
• Updated jar files with the addition of the updated html files of the userguide. So the Help button will show a synchronized version of html files as the running GUI.
• Updated some ROMIO Makefile.in files for building shared libraries
• Updated test configure
• Updated the launcher message and added a VERBOSE parameter in setup file.
• Updated the userguide's lyx, tex, pdf and html files with changes in logfile convertor GUI.
• Updated version number
• updates to remove warnings on IA-32 and IA-64
• Use -p when copying the libraries for the Fortran step to preserve modification times
• Use rm instead of /bin/rm for consistency with other commands; also be consistent with AC_MSG_CHECKING and use for/whether instead of if/that/nothing
• use the striping_factor hint to set the number of pvfs2 datafiles used
• Used the formal definition of ShadowCategory from Category class.
• we need to use CPPFLAGS in the shared library path as well

Files for each revision