April 2013 - commits - mailman3.cels.anl.gov

[mpich] MPICH primary repository branch, master, updated. v3.0.4
by noreply＠mpich.org 24 Apr '13

24 Apr '13

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via 85acf6c54540d3e5e63deb4ef3a003dc78803659 (commit) from 0f1067da5ffe74f6288f2767fe701d82a85ce99c (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/85acf6c54540d3e5e63deb4ef3a003dc7… commit 85acf6c54540d3e5e63deb4ef3a003dc78803659 Author: Pavan Balaji <balaji(a)mcs.anl.gov> Date: Wed Apr 24 10:18:11 2013 -0500 Use VPATH builds for docs in release tarball creation. We were making an additional copy of the code to create documents in-path because VPATH builds didn't work correctly for docs. At some point (possibly when we moved to the new build system), this got fixed and this round-about method is no longer needed. Reviewed by dinan. diff --git a/maint/release.pl b/maint/release.pl index a4d2916..701494b 100755 --- a/maint/release.pl +++ b/maint/release.pl @@ -284,23 +284,13 @@ print("done\n"); # Get docs print("===> Creating secondary codebase for the docs... "); -run_cmd("cp -a ${expdir} ${expdir}-tmp"); -print("done\n"); - -print("===> Configuring and making the secondary codebase... "); -chdir("${expdir}-tmp"); -{ - my $cmd = "./autogen.sh"; - $cmd .= " --with-autoconf=$with_autoconf" if $with_autoconf; - $cmd .= " --with-automake=$with_automake" if $with_automake; - run_cmd($cmd); -} -run_cmd("./configure --disable-fc --disable-f77 --disable-cxx"); +run_cmd("mkdir ${expdir}-build"); +chdir("${expdir}-build"); +run_cmd("${expdir}/configure --disable-fc --disable-f77 --disable-cxx"); run_cmd("(make mandoc && make htmldoc && make latexdoc)"); print("done\n"); print("===> Copying docs over... "); -chdir("${expdir}-tmp"); run_cmd("cp -a man ${expdir}"); run_cmd("cp -a www ${expdir}"); run_cmd("cp -a doc/userguide/user.pdf ${expdir}/doc/userguide"); ----------------------------------------------------------------------- Summary of changes: maint/release.pl | 16 +++------------- 1 files changed, 3 insertions(+), 13 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository annotated tag, v3.0.4, created. v3.0.4
by noreply＠mpich.org 24 Apr '13

24 Apr '13

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The annotated tag, v3.0.4 has been created at 2c1364662265cc95d72764bfa6456669b7a6d359 (tag) tagging 85acf6c54540d3e5e63deb4ef3a003dc78803659 (commit) replaces v3.0.3 tagged by Pavan Balaji on Wed Apr 24 21:06:28 2013 -0500 - Log ----------------------------------------------------------------- tagging 'v3.0.4' Dave Goodell (19): .gitignore: test directory autogenerated files add `mpiu_uthash.h` mark leaky cancel tests as xfail missed "mpiinfo.h" include in [9bb89788] .gitignore updates for the test suite make `MPIR_ContextMaskToStr` non-static add `MPIR_ContextMaskStats` debug routine report context ID counts on allocation failure TAP: comply better with "TAP Version 13" add `pipestatus` shell helper delete more "relative" svn:externals dirs un-break `--enable-strict` when using clang strict: add `-Werror-implicit-function-declaration` use C99-standard `__VA_ARGS__` fix uninitialized variable warnings tt#1817: ensure dependencies for PAC_PROG_CC fix PAC_PROG_CC ordering in configure.ac scripts tt#1817: move configure subdir exports earlier add clang/clang++ to compiler search lists James Dinan (1): Squash ARMCI-MPI build warnings Pavan Balaji (24): Bug-fix: We should use HYDU_FUNC instead of __func__ directly. We Fix typo -- the performance tests were not correctly disabled in Fix English typos caught by the Debian lintian checks. Added the per-communicator thresholds addition into the CHANGES file. Fix typo introduced in dd76a0f that's causing the nightly tests to Improvements to the release.pl script. Make kvs_name and kvsname consistent. Added ability to allgather codes to all the proxies during a PMI Cleanup memory allocation for the keyvals being forwarded to the CHANGES file updates for Hydra. Update mpich release and ABI versions for the 3.0.4 release. Move folks who are no longer contributing to the MPICH code to the Warning squash: unused variables. Bug-fix: PMI keyval allgather code. Single-quotes are used for manpage formatting. Yet another bug-fix in the commenting out the performance tests from Improve [480baf02] with Dave's suggestion. Bug fixes in the PMI keyval allgather code from [3ba0a7f]. Squash warnings. Static functions do not need to be name-spaced. Make the InitPG naming consistent. Reorder C and C++ compiler search list. Update CHANGES file about compiler reordering. Use VPATH builds for docs in release tarball creation. Ralf Gunter (3): VCs now use eager threshold from environment var. Add info hint support to communicators. Add per-communicator eager threshold support. William Gropp (4): Fix for #1772 - problem building on SunOS with F77=f90 Begin cleaning up error reporting Removed and fixed some FIXMEs Fix for #1813 ----------------------------------------------------------------------- hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.3-50-g0f1067d
by noreply＠mpich.org 24 Apr '13

24 Apr '13

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via 0f1067da5ffe74f6288f2767fe701d82a85ce99c (commit) from d4aa328f8e4ea24afdc60dc3d64182961828c178 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/0f1067da5ffe74f6288f2767fe701d82a… commit 0f1067da5ffe74f6288f2767fe701d82a85ce99c Author: William Gropp <wgropp(a)illinois.edu> Date: Tue Apr 23 09:17:49 2013 -0700 Fix for #1813 Pass the second argument to MPI_WIN_CREATE in Fortran as an address-sized integer to fix a test failure. diff --git a/test/mpi/errors/f77/errhan/uerrhandf.f b/test/mpi/errors/f77/errhan/uerrhandf.f index cd3c8bd..26eef2b 100644 --- a/test/mpi/errors/f77/errhan/uerrhandf.f +++ b/test/mpi/errors/f77/errhan/uerrhandf.f @@ -57,7 +57,8 @@ C endif endif - call mpi_win_create( winbuf, 2*sizeofint, sizeofint, MPI_INFO_NULL + asize = 2*sizeofint + call mpi_win_create( winbuf, asize, sizeofint, MPI_INFO_NULL $ , wdup, winh, ierr ) if (ierr .ne. MPI_SUCCESS) then call mtestprinterrormsg( "Win_create:", ierr ) ----------------------------------------------------------------------- Summary of changes: test/mpi/errors/f77/errhan/uerrhandf.f | 3 ++- 1 files changed, 2 insertions(+), 1 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.3-49-gd4aa328
by noreply＠mpich.org 23 Apr '13

23 Apr '13

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via d4aa328f8e4ea24afdc60dc3d64182961828c178 (commit) from c03fdd083f606d35199b98426a4d48ff8708ac31 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/d4aa328f8e4ea24afdc60dc3d64182961… commit d4aa328f8e4ea24afdc60dc3d64182961828c178 Author: Pavan Balaji <balaji(a)mcs.anl.gov> Date: Tue Apr 23 16:34:45 2013 -0500 Update CHANGES file about compiler reordering. Added a warning in the CHANGES file about the potential ABI breakage because of the compiler reordering if CC and friends are not added. Reviewed by goodell. diff --git a/CHANGES b/CHANGES index 94af968..51b88e3 100644 --- a/CHANGES +++ b/CHANGES @@ -2,6 +2,14 @@ Changes in 3.0.4 =============================================================================== + # BUILD SYSTEM: Reordered the default compiler search to prefer Intel + and PG compilers over GNU compilers because of the performance + difference. + + WARNING: If you do not explicitly specify the compiler you want + through CC and friends, this might break ABI for you relative to + the previous 3.0.x release. + # OVERALL: Added support to manage per-communicator eager-rendezvous thresholds. ----------------------------------------------------------------------- Summary of changes: CHANGES | 8 ++++++++ 1 files changed, 8 insertions(+), 0 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.3-48-gc03fdd0
by noreply＠mpich.org 23 Apr '13

23 Apr '13

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via c03fdd083f606d35199b98426a4d48ff8708ac31 (commit) from 0cca71f8e0cc0bbce77783f27b12aed1e4aa204e (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/c03fdd083f606d35199b98426a4d48ff8… commit c03fdd083f606d35199b98426a4d48ff8708ac31 Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Mon Apr 22 17:33:09 2013 -0500 add clang/clang++ to compiler search lists At Pavan's request, this is at a lower priority than gcc. Note, however, that clang might still be selected by default on some platforms if it is aliased to `cc`/`c++`, which are preferred over `gcc`/`g++` as of [1a1ff11e]. No reviewer. diff --git a/confdb/aclocal_cc.m4 b/confdb/aclocal_cc.m4 index 08439e4..fa44e6d 100644 --- a/confdb/aclocal_cc.m4 +++ b/confdb/aclocal_cc.m4 @@ -11,7 +11,7 @@ AC_DEFUN([PAC_PROG_CC],[ dnl developers notice this case. AC_BEFORE([$0],[AC_PROG_CC]) PAC_PUSH_FLAG([CFLAGS]) - AC_PROG_CC([icc pgcc xlc xlC pathcc cc gcc]) + AC_PROG_CC([icc pgcc xlc xlC pathcc cc gcc clang]) PAC_POP_FLAG([CFLAGS]) ]) dnl diff --git a/confdb/aclocal_cxx.m4 b/confdb/aclocal_cxx.m4 index 2f5cacc..279b668 100644 --- a/confdb/aclocal_cxx.m4 +++ b/confdb/aclocal_cxx.m4 @@ -1,7 +1,7 @@ dnl PAC_CXX_SEARCH_LIST - expands to a whitespace separated list of C++ dnl compilers for use with AC_PROG_CXX that is more suitable for HPC software dnl packages -AC_DEFUN([PAC_CXX_SEARCH_LIST],[$CCC icpc pgCC xlC pathCC c++ cxx CC g++ gcc cc++ cl]) +AC_DEFUN([PAC_CXX_SEARCH_LIST],[$CCC icpc pgCC xlC pathCC c++ cxx CC g++ clang++ gcc cc++ cl]) dnl PAC_PROG_CXX - reprioritize the C++ compiler search order dnl NOTE: this macro suffers from a basically intractable "expanded before it dnl was required" problem when libtool is also used ----------------------------------------------------------------------- Summary of changes: confdb/aclocal_cc.m4 | 2 +- confdb/aclocal_cxx.m4 | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.3-47-g0cca71f
by noreply＠mpich.org 23 Apr '13

23 Apr '13

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via 0cca71f8e0cc0bbce77783f27b12aed1e4aa204e (commit) via 00275e470873cd27b6366964233c16d7a19a1b45 (commit) via 9309120a2da4ba8da826437589490e1b7a538259 (commit) from 1a1ff11ed57a670118e7209c65d3ec7bdc8244c4 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/0cca71f8e0cc0bbce77783f27b12aed1e… commit 0cca71f8e0cc0bbce77783f27b12aed1e4aa204e Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Tue Apr 23 14:13:40 2013 -0500 tt#1817: move configure subdir exports earlier We already were exporting these variables for the "main" subdirs, such as the configured device/channel/netmod. The bug was that we weren't setting these exports in time for OPA or MPL to pick them up. When [1a1ff11e] came along and reordered the compiler search list, this caused MPICH to choose `cc` and OPA to choose `gcc`, which don't agree on the validity of certain warning flags. So `--enable-strict=c99` causes OPA's configure to fail because of a bad compiler flag exported by the top level. An alternative "fix" to this commit is to ensure that every single subdir package uses exactly the same logic in order to select compilers and compiler options. This seems much more difficult to ensure. Reviewed-by: balaji diff --git a/configure.ac b/configure.ac index d135372..35dc573 100644 --- a/configure.ac +++ b/configure.ac @@ -740,6 +740,37 @@ export master_top_builddir export master_top_srcdir # ---------------------------------------------------------------------------- +dnl Export important "precious" variables so that any directories configured via +dnl PAC_CONFIG_SUBDIR will agree with the top-level configure about these +dnl critical variables (esp. compiler selection). These exports should come +dnl before any subconfigures in this script. +dnl +dnl This list is arguably incomplete, and should possibly be automatically +dnl generated from "$ac_precious_vars" using code similar to the implementation +dnl of PAC_CONFIG_SUBDIR. +dnl +dnl To be clear, without these exports any variable values determined by this +dnl configure script will not be seen by child scripts. Instead they will dnl +dnl receive the only the original inherited environment and configure args used +dnl when this configure script was invoked. +export AR +export AR_FLAGS +export CC +export CFLAGS +export CPPFLAGS +export CXX +export CXXFLAGS +export F77 +export FC +export FCFLAGS +export FFLAGS +export LDFLAGS +export LIBS +export MPILIBNAME +export PMPILIBNAME +export RANLIB + +# ---------------------------------------------------------------------------- # with-device if test "$with_device" = "default" ; then # Pick the device. For now, always choose ch3 @@ -5170,7 +5201,6 @@ fi dnl Configure any subdirectories. Note that config.status will *not* dnl reexecute these! -dnl Export any important variables first: dnl dnl Gastly problem. CONFIG_SUBDIRS only adds the directories to the dnl list of directories to be configured. It does NOT control the @@ -5181,25 +5211,7 @@ dnl uses to handle the subdir configure. However, later versions of dnl autoconf did this in a way that caused problems, paritcularly with dnl errors reported as inconsistent cache files. Instead, we simply dnl invoke the configure scripts (if present) directly. -export AR -export AR_FLAGS -export RANLIB -export MPILIBNAME -export PMPILIBNAME -export CC -export CPPFLAGS -export LIBS -export CXX -export CXXFLAGS -export FFLAGS -export CFLAGS -export FCFLAGS -export LDFLAGS -# Make sure that any Fortran 77 and Fortran 90 compilers are exported to -# the subdir builds (in particular, ROMIO may try to use the Fortran 90 -# compiler to determine the Fortran 90 KINDS for MPI_OFFSET_KIND). -export F77 -export FC + # # ----------------------------------------------------------------------------- # Configure threads first. This is necessary to obtain all required http://git.mpich.org/mpich.git/commitdiff/00275e470873cd27b6366964233c16d7a… commit 00275e470873cd27b6366964233c16d7a19a1b45 Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Tue Apr 23 13:37:34 2013 -0500 fix PAC_PROG_CC ordering in configure.ac scripts If you get a message like this: ``` configure.ac:30: warning: AC_PROG_CC was called before PAC_PROG_CC confdb/aclocal_cc.m4:10: PAC_PROG_CC is expanded from... configure.ac:30: the top level ``` but can't seem to figure out where the earlier AC_PROG_CC macro is coming from, then you can run autoconf directly with the `--trace=AC_PROG_CC` argument for more information. Reviewed-by: balaji diff --git a/src/mpi/romio/configure.ac b/src/mpi/romio/configure.ac index 2ee57b0..e68f08a 100644 --- a/src/mpi/romio/configure.ac +++ b/src/mpi/romio/configure.ac @@ -25,6 +25,9 @@ AC_CONFIG_MACRO_DIR([confdb]) AM_INIT_AUTOMAKE([-Wall -Werror -Wno-portability-recursive foreign 1.12.3 silent-rules subdir-objects]) AM_MAINTAINER_MODE([enable]) +dnl must come before LT_INIT, which AC_REQUIREs AC_PROG_CC +PAC_PROG_CC + AM_PROG_AR LT_INIT([]) @@ -425,8 +428,6 @@ case $ARCH in ;; esac -PAC_PROG_CC - dnl AC_PROG_{CXX,F77,FC} must come early in configure.ac in order to control dnl compiler search order and avoid some esoteric autoconf macro expansion dnl errors diff --git a/src/mpl/configure.ac b/src/mpl/configure.ac index b147649..d048814 100644 --- a/src/mpl/configure.ac +++ b/src/mpl/configure.ac @@ -13,6 +13,9 @@ AC_CONFIG_AUX_DIR(confdb) AC_CONFIG_MACRO_DIR(confdb) AM_INIT_AUTOMAKE([-Wall -Werror foreign 1.12.3]) +dnl must come before LT_INIT, which AC_REQUIREs AC_PROG_CC +PAC_PROG_CC + AM_PROG_AR LT_PREREQ([2.2.6]) @@ -35,7 +38,6 @@ AX_PREFIX_CONFIG_H([include/mplconfig.h],[MPL]) # Non-verbose make m4_ifdef([AM_SILENT_RULES], [AM_SILENT_RULES([yes])]) -PAC_PROG_CC AC_C_CONST AC_C_RESTRICT diff --git a/src/pm/hydra/configure.ac b/src/pm/hydra/configure.ac index 1398dcc..892363f 100644 --- a/src/pm/hydra/configure.ac +++ b/src/pm/hydra/configure.ac @@ -14,7 +14,15 @@ AC_CONFIG_MACRO_DIR(confdb) # bizarre expansion ordering warnings AC_CANONICAL_TARGET AC_ARG_PROGRAM + +dnl must come before LT_INIT, which AC_REQUIREs AC_PROG_CC +dnl must also come before AC_USE_SYSTEM_EXTENSIONS +PAC_PROG_CC + +# also needed by hwloc in embedded mode, must also come early for expansion +# ordering reasons AC_USE_SYSTEM_EXTENSIONS + # Define -D_DARWIN_C_SOURCE on OS/X to ensure that hwloc will build even if we # are building under MPICH with --enable-strict that defined _POSIX_C_SOURCE. # Some standard Darwin headers don't build correctly under a strict posix @@ -36,7 +44,6 @@ m4_ifdef([AM_SILENT_RULES], [AM_SILENT_RULES([yes])]) # Reset link flags PAC_RESET_LINK_FLAGS() -PAC_PROG_CC AM_PROG_CC_C_O PAC_ARG_STRICT http://git.mpich.org/mpich.git/commitdiff/9309120a2da4ba8da826437589490e1b7… commit 9309120a2da4ba8da826437589490e1b7a538259 Author: Dave Goodell <goodell(a)mcs.anl.gov> Date: Tue Apr 23 13:04:41 2013 -0500 tt#1817: ensure dependencies for PAC_PROG_CC `PAC_PROG_CC` should always come before `AC_PROG_CC` if the former is used anywhere in a configure script. Otherwise one of the many (often standard) macros which `AC_REQUIRE([AC_PROG_CC])` might inadvertently bypass the user's desired logic. The `AC_BEFORE` call here will cause an autogen-time warning to be emitted if the order is accidentally interchanged. Reviewed-by: balaji diff --git a/confdb/aclocal_cc.m4 b/confdb/aclocal_cc.m4 index 07b5548..08439e4 100644 --- a/confdb/aclocal_cc.m4 +++ b/confdb/aclocal_cc.m4 @@ -3,6 +3,13 @@ ifdef([AC_PROG_CC_GNU],,[AC_DEFUN([AC_PROG_CC_GNU],)]) dnl PAC_PROG_CC - reprioritize the C compiler search order AC_DEFUN([PAC_PROG_CC],[ + dnl Many standard autoconf/automake/libtool macros, such as LT_INIT, + dnl perform an AC_REQUIRE([AC_PROG_CC]). If this macro (PAC_PROG_CC) + dnl comes after LT_INIT (or similar) then the default compiler search + dnl path will be used instead. This AC_BEFORE macro ensures that a + dnl warning will be emitted at autoconf-time (autogen.sh-time) to help + dnl developers notice this case. + AC_BEFORE([$0],[AC_PROG_CC]) PAC_PUSH_FLAG([CFLAGS]) AC_PROG_CC([icc pgcc xlc xlC pathcc cc gcc]) PAC_POP_FLAG([CFLAGS]) ----------------------------------------------------------------------- Summary of changes: confdb/aclocal_cc.m4 | 7 ++++++ configure.ac | 52 +++++++++++++++++++++++++++----------------- src/mpi/romio/configure.ac | 5 ++- src/mpl/configure.ac | 4 ++- src/pm/hydra/configure.ac | 9 ++++++- 5 files changed, 53 insertions(+), 24 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.3-44-g1a1ff11
by noreply＠mpich.org 23 Apr '13

23 Apr '13

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via 1a1ff11ed57a670118e7209c65d3ec7bdc8244c4 (commit) from fbc070994cc58f713ecb206721da42b9d8c95dcb (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/1a1ff11ed57a670118e7209c65d3ec7bd… commit 1a1ff11ed57a670118e7209c65d3ec7bdc8244c4 Author: Pavan Balaji <balaji(a)mcs.anl.gov> Date: Tue Apr 23 07:49:11 2013 -0500 Reorder C and C++ compiler search list. Use similar compiler search ordering for C and C++ as Fortran. Some of the compilers, such as gcc, do not make sense in the C++ compiler search list, but this patch does not eliminate them. Reviewed by goodell. diff --git a/confdb/aclocal_cc.m4 b/confdb/aclocal_cc.m4 index 237c435..07b5548 100644 --- a/confdb/aclocal_cc.m4 +++ b/confdb/aclocal_cc.m4 @@ -4,7 +4,7 @@ ifdef([AC_PROG_CC_GNU],,[AC_DEFUN([AC_PROG_CC_GNU],)]) dnl PAC_PROG_CC - reprioritize the C compiler search order AC_DEFUN([PAC_PROG_CC],[ PAC_PUSH_FLAG([CFLAGS]) - AC_PROG_CC([gcc icc pgcc xlc xlC pathcc cc]) + AC_PROG_CC([icc pgcc xlc xlC pathcc cc gcc]) PAC_POP_FLAG([CFLAGS]) ]) dnl diff --git a/confdb/aclocal_cxx.m4 b/confdb/aclocal_cxx.m4 index ec4a5a2..2f5cacc 100644 --- a/confdb/aclocal_cxx.m4 +++ b/confdb/aclocal_cxx.m4 @@ -1,7 +1,7 @@ dnl PAC_CXX_SEARCH_LIST - expands to a whitespace separated list of C++ dnl compilers for use with AC_PROG_CXX that is more suitable for HPC software dnl packages -AC_DEFUN([PAC_CXX_SEARCH_LIST],[$CCC c++ g++ gcc CC cxx cc++ cl pgCC xlC icpc pathCC]) +AC_DEFUN([PAC_CXX_SEARCH_LIST],[$CCC icpc pgCC xlC pathCC c++ cxx CC g++ gcc cc++ cl]) dnl PAC_PROG_CXX - reprioritize the C++ compiler search order dnl NOTE: this macro suffers from a basically intractable "expanded before it dnl was required" problem when libtool is also used ----------------------------------------------------------------------- Summary of changes: confdb/aclocal_cc.m4 | 2 +- confdb/aclocal_cxx.m4 | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.3-43-gfbc0709
by noreply＠mpich.org 22 Apr '13

22 Apr '13

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via fbc070994cc58f713ecb206721da42b9d8c95dcb (commit) via 1a95a84b7a83c1920da1264817aa96c6d0fb8f64 (commit) via 163170510458a9921dfd0bdfdc8255746a73704f (commit) from fd583f48d7e8b3009e359c35786a8fc0f9cc398c (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/fbc070994cc58f713ecb206721da42b9d… commit fbc070994cc58f713ecb206721da42b9d8c95dcb Author: Pavan Balaji <balaji(a)mcs.anl.gov> Date: Mon Apr 22 14:45:36 2013 -0500 Make the InitPG naming consistent. Avoid camel-cased names. Reviewed by goodell. diff --git a/src/mpid/ch3/src/mpid_init.c b/src/mpid/ch3/src/mpid_init.c index c37014b..414734f 100644 --- a/src/mpid/ch3/src/mpid_init.c +++ b/src/mpid/ch3/src/mpid_init.c @@ -30,7 +30,7 @@ char *MPIDI_DBG_parent_str = "?"; int MPIDI_Use_pmi2_api = 0; -static int InitPG( int *argc_p, char ***argv_p, +static int init_pg( int *argc_p, char ***argv_p, int *has_args, int *has_env, int *has_parent, int *pg_rank_p, MPIDI_PG_t **pg_p ); static int pg_compare_ids(void * id1, void * id2); @@ -133,7 +133,7 @@ int MPID_Init(int *argc, char ***argv, int requested, int *provided, /* * Perform channel-independent PMI initialization */ - mpi_errno = InitPG( argc, argv, + mpi_errno = init_pg( argc, argv, has_args, has_env, &has_parent, &pg_rank, &pg ); if (mpi_errno) { MPIU_ERR_SETANDJUMP(mpi_errno,MPI_ERR_OTHER, "**ch3|ch3_init"); @@ -352,7 +352,7 @@ int MPID_InitCompleted( void ) * process group structures. * */ -static int InitPG( int *argc, char ***argv, +static int init_pg( int *argc, char ***argv, int *has_args, int *has_env, int *has_parent, int *pg_rank_p, MPIDI_PG_t **pg_p ) { http://git.mpich.org/mpich.git/commitdiff/1a95a84b7a83c1920da1264817aa96c6d… commit 1a95a84b7a83c1920da1264817aa96c6d0fb8f64 Author: Pavan Balaji <balaji(a)mcs.anl.gov> Date: Mon Apr 22 14:44:47 2013 -0500 Static functions do not need to be name-spaced. Reviewed by goodell. diff --git a/src/mpid/ch3/src/mpid_init.c b/src/mpid/ch3/src/mpid_init.c index 8d0e748..c37014b 100644 --- a/src/mpid/ch3/src/mpid_init.c +++ b/src/mpid/ch3/src/mpid_init.c @@ -33,9 +33,9 @@ int MPIDI_Use_pmi2_api = 0; static int InitPG( int *argc_p, char ***argv_p, int *has_args, int *has_env, int *has_parent, int *pg_rank_p, MPIDI_PG_t **pg_p ); -static int MPIDI_CH3I_PG_Compare_ids(void * id1, void * id2); -static int MPIDI_CH3I_PG_Destroy(MPIDI_PG_t * pg ); -static int MPIDI_CH3_Set_eager_threshold(MPID_Comm *comm_ptr, MPID_Info *info, void *state); +static int pg_compare_ids(void * id1, void * id2); +static int pg_destroy(MPIDI_PG_t * pg ); +static int set_eager_threshold(MPID_Comm *comm_ptr, MPID_Info *info, void *state); MPIDI_Process_t MPIDI_Process = { NULL }; MPIDI_CH3U_SRBuf_element_t * MPIDI_CH3U_SRBuf_pool = NULL; @@ -59,10 +59,10 @@ static int finalize_failed_procs_group(void *param) } #undef FUNCNAME -#define FUNCNAME MPIDI_CH3_Set_eager_threshold +#define FUNCNAME set_eager_threshold #undef FCNAME #define FCNAME MPIU_QUOTE(FUNCNAME) -static int MPIDI_CH3_Set_eager_threshold(MPID_Comm *comm_ptr, MPID_Info *info, void *state) +static int set_eager_threshold(MPID_Comm *comm_ptr, MPID_Info *info, void *state) { int mpi_errno = MPI_SUCCESS; char *endptr; @@ -323,7 +323,7 @@ int MPID_Init(int *argc, char ***argv, int requested, int *provided, } mpi_errno = MPIR_Comm_register_hint("eager_rendezvous_threshold", - MPIDI_CH3_Set_eager_threshold, + set_eager_threshold, NULL); if (mpi_errno) MPIU_ERR_POP(mpi_errno); @@ -482,7 +482,7 @@ static int InitPG( int *argc, char ***argv, * Initialize the process group tracking subsystem */ mpi_errno = MPIDI_PG_Init(argc, argv, - MPIDI_CH3I_PG_Compare_ids, MPIDI_CH3I_PG_Destroy); + pg_compare_ids, pg_destroy); if (mpi_errno != MPI_SUCCESS) { MPIU_ERR_SETANDJUMP(mpi_errno,MPI_ERR_OTHER,"**dev|pg_init"); } @@ -582,13 +582,13 @@ int MPIDI_CH3I_BCFree( char *bc_val ) /* FIXME: The PG code should supply these, since it knows how the pg_ids and other data are represented */ -static int MPIDI_CH3I_PG_Compare_ids(void * id1, void * id2) +static int pg_compare_ids(void * id1, void * id2) { return (strcmp((char *) id1, (char *) id2) == 0) ? TRUE : FALSE; } -static int MPIDI_CH3I_PG_Destroy(MPIDI_PG_t * pg) +static int pg_destroy(MPIDI_PG_t * pg) { if (pg->id != NULL) { http://git.mpich.org/mpich.git/commitdiff/163170510458a9921dfd0bdfdc8255746… commit 163170510458a9921dfd0bdfdc8255746a73704f Author: Pavan Balaji <balaji(a)mcs.anl.gov> Date: Mon Apr 22 14:43:03 2013 -0500 Squash warnings. Reviewed by goodell. diff --git a/src/mpi/comm/comm_set_info.c b/src/mpi/comm/comm_set_info.c index 71445c0..5afbc61 100644 --- a/src/mpi/comm/comm_set_info.c +++ b/src/mpi/comm/comm_set_info.c @@ -32,7 +32,6 @@ int MPIR_Comm_set_info_impl(MPID_Comm * comm_ptr, MPID_Info * info_ptr) { int mpi_errno = MPI_SUCCESS; - MPID_Info *info; MPID_Info *curr_info = NULL; MPID_MPI_STATE_DECL(MPID_STATE_MPIR_COMM_SET_INFO_IMPL); diff --git a/src/mpid/ch3/src/mpid_init.c b/src/mpid/ch3/src/mpid_init.c index 2a7b24e..8d0e748 100644 --- a/src/mpid/ch3/src/mpid_init.c +++ b/src/mpid/ch3/src/mpid_init.c @@ -35,7 +35,7 @@ static int InitPG( int *argc_p, char ***argv_p, int *pg_rank_p, MPIDI_PG_t **pg_p ); static int MPIDI_CH3I_PG_Compare_ids(void * id1, void * id2); static int MPIDI_CH3I_PG_Destroy(MPIDI_PG_t * pg ); - +static int MPIDI_CH3_Set_eager_threshold(MPID_Comm *comm_ptr, MPID_Info *info, void *state); MPIDI_Process_t MPIDI_Process = { NULL }; MPIDI_CH3U_SRBuf_element_t * MPIDI_CH3U_SRBuf_pool = NULL; @@ -62,7 +62,7 @@ static int finalize_failed_procs_group(void *param) #define FUNCNAME MPIDI_CH3_Set_eager_threshold #undef FCNAME #define FCNAME MPIU_QUOTE(FUNCNAME) -int MPIDI_CH3_Set_eager_threshold(MPID_Comm *comm_ptr, MPID_Info *info, void *state) +static int MPIDI_CH3_Set_eager_threshold(MPID_Comm *comm_ptr, MPID_Info *info, void *state) { int mpi_errno = MPI_SUCCESS; char *endptr; ----------------------------------------------------------------------- Summary of changes: src/mpi/comm/comm_set_info.c | 1 - src/mpid/ch3/src/mpid_init.c | 24 ++++++++++++------------ 2 files changed, 12 insertions(+), 13 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.3-40-gfd583f4
by noreply＠mpich.org 22 Apr '13

22 Apr '13

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via fd583f48d7e8b3009e359c35786a8fc0f9cc398c (commit) from 1c156acf01ea5ed49252c10d2995d48f3f34c09f (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/fd583f48d7e8b3009e359c35786a8fc0f… commit fd583f48d7e8b3009e359c35786a8fc0f9cc398c Author: Pavan Balaji <balaji(a)mcs.anl.gov> Date: Mon Apr 22 13:21:52 2013 -0500 Bug fixes in the PMI keyval allgather code from [3ba0a7f]. 1. We had forgotten to distribute the pre-initialized keyvals in the spawn code to the new proxies. 2. We used a global static for keeping track of the previously distributed keyvals, instead of a per-KVS-space static. Fixes tt#1814. No reviewer. diff --git a/src/pm/hydra/pm/pmiserv/pmiserv_pmi.h b/src/pm/hydra/pm/pmiserv/pmiserv_pmi.h index f939033..3f91267 100644 --- a/src/pm/hydra/pm/pmiserv/pmiserv_pmi.h +++ b/src/pm/hydra/pm/pmiserv/pmiserv_pmi.h @@ -37,6 +37,7 @@ struct HYD_pmcd_pmi_pg_scratch { int dead_process_count; struct HYD_pmcd_pmi_kvs *kvs; + int keyval_dist_count; /* Number of keyvals distributed */ }; struct HYD_pmcd_pmi_publish { diff --git a/src/pm/hydra/pm/pmiserv/pmiserv_pmi_v1.c b/src/pm/hydra/pm/pmiserv/pmiserv_pmi_v1.c index dfbed96..cbc4c08 100644 --- a/src/pm/hydra/pm/pmiserv/pmiserv_pmi_v1.c +++ b/src/pm/hydra/pm/pmiserv/pmiserv_pmi_v1.c @@ -44,14 +44,13 @@ static HYD_status cmd_response(int fd, int pid, const char *cmd) goto fn_exit; } -static HYD_status fn_barrier_in(int fd, int pid, int pgid, char *args[]) +static HYD_status bcast_keyvals(int fd, int pid) { + int keyval_count, arg_count, i, j; + char **tmp = NULL, *cmd; + struct HYD_pmcd_pmi_kvs_pair *run; struct HYD_proxy *proxy, *tproxy; struct HYD_pmcd_pmi_pg_scratch *pg_scratch; - int proxy_count, keyval_count, i, j, arg_count; - struct HYD_pmcd_pmi_kvs_pair *run; - char **tmp = NULL, *cmd; - static int keyval_count_distributed = 0; HYD_status status = HYD_SUCCESS; HYDU_FUNC_ENTER(); @@ -60,88 +59,108 @@ static HYD_status fn_barrier_in(int fd, int pid, int pgid, char *args[]) HYDU_ASSERT(proxy, status); pg_scratch = (struct HYD_pmcd_pmi_pg_scratch *) proxy->pg->pg_scratch; - proxy_count = 0; - for (tproxy = proxy->pg->proxy_list; tproxy; tproxy = tproxy->next) - proxy_count++; + /* find the number of keyvals */ + keyval_count = 0; + for (run = pg_scratch->kvs->key_pair; run; run = run->next) + keyval_count++; - proxy->pg->barrier_count++; - if (proxy->pg->barrier_count == proxy_count) { - proxy->pg->barrier_count = 0; + keyval_count -= pg_scratch->keyval_dist_count; - /* find the number of keyvals */ - keyval_count = 0; - for (run = pg_scratch->kvs->key_pair; run; run = run->next) - keyval_count++; + /* Each keyval has the following four items: 'key' '=' 'val' + * '<space>'. Two additional items for the command at the start + * and the NULL at the end. */ + HYDU_MALLOC(tmp, char **, (4 * keyval_count + 3) * sizeof(char *), status); - keyval_count -= keyval_count_distributed; + /* send all available keyvals downstream */ + if (keyval_count) { + arg_count = 1; + i = 0; + tmp[i++] = HYDU_strdup("cmd=keyval_cache "); + for (run = pg_scratch->kvs->key_pair, j = 0; run; run = run->next, j++) { + if (j < pg_scratch->keyval_dist_count) + continue; - /* Each keyval has the following four items: 'key' '=' 'val' - * '<space>'. Two additional items for the command at the - * start and the NULL at the end. */ - HYDU_MALLOC(tmp, char **, (4 * keyval_count + 3) * sizeof(char *), status); + tmp[i++] = HYDU_strdup(run->key); + tmp[i++] = HYDU_strdup("="); + tmp[i++] = HYDU_strdup(run->val); + tmp[i++] = HYDU_strdup(" "); - /* send all available keyvals downstream */ - if (keyval_count) { - arg_count = 1; - i = 0; - tmp[i++] = HYDU_strdup("cmd=keyval_cache "); - for (run = pg_scratch->kvs->key_pair, j = 0; run; run = run->next, j++) { - if (j < keyval_count_distributed) - continue; - - tmp[i++] = HYDU_strdup(run->key); - tmp[i++] = HYDU_strdup("="); - tmp[i++] = HYDU_strdup(run->val); - tmp[i++] = HYDU_strdup(" "); - - arg_count++; - if (arg_count >= MAX_PMI_INTERNAL_ARGS) { - tmp[i++] = HYDU_strdup("\n"); - tmp[i++] = NULL; - - status = HYDU_str_alloc_and_join(tmp, &cmd); - HYDU_ERR_POP(status, "unable to join strings\n"); - HYDU_free_strlist(tmp); - - keyval_count_distributed += (arg_count - 1); - for (tproxy = proxy->pg->proxy_list; tproxy; tproxy = tproxy->next) { - status = cmd_response(tproxy->control_fd, pid, cmd); - HYDU_ERR_POP(status, "error writing PMI line\n"); - } - HYDU_FREE(cmd); - - i = 0; - tmp[i++] = HYDU_strdup("cmd=keyval_cache "); - arg_count = 1; - } - } - tmp[i++] = HYDU_strdup("\n"); - tmp[i++] = NULL; + arg_count++; + if (arg_count >= MAX_PMI_INTERNAL_ARGS) { + tmp[i++] = HYDU_strdup("\n"); + tmp[i++] = NULL; - if (arg_count > 1) { status = HYDU_str_alloc_and_join(tmp, &cmd); HYDU_ERR_POP(status, "unable to join strings\n"); + HYDU_free_strlist(tmp); - keyval_count_distributed += (arg_count - 1); + pg_scratch->keyval_dist_count += (arg_count - 1); for (tproxy = proxy->pg->proxy_list; tproxy; tproxy = tproxy->next) { status = cmd_response(tproxy->control_fd, pid, cmd); HYDU_ERR_POP(status, "error writing PMI line\n"); } HYDU_FREE(cmd); + + i = 0; + tmp[i++] = HYDU_strdup("cmd=keyval_cache "); + arg_count = 1; } - HYDU_free_strlist(tmp); } + tmp[i++] = HYDU_strdup("\n"); + tmp[i++] = NULL; + + if (arg_count > 1) { + status = HYDU_str_alloc_and_join(tmp, &cmd); + HYDU_ERR_POP(status, "unable to join strings\n"); + + pg_scratch->keyval_dist_count += (arg_count - 1); + for (tproxy = proxy->pg->proxy_list; tproxy; tproxy = tproxy->next) { + status = cmd_response(tproxy->control_fd, pid, cmd); + HYDU_ERR_POP(status, "error writing PMI line\n"); + } + HYDU_FREE(cmd); + } + HYDU_free_strlist(tmp); + } + + fn_exit: + if (tmp) + HYDU_FREE(tmp); + HYDU_FUNC_EXIT(); + return status; + + fn_fail: + goto fn_exit; +} + +static HYD_status fn_barrier_in(int fd, int pid, int pgid, char *args[]) +{ + struct HYD_proxy *proxy, *tproxy; + int proxy_count; + HYD_status status = HYD_SUCCESS; + + HYDU_FUNC_ENTER(); + + proxy = HYD_pmcd_pmi_find_proxy(fd); + HYDU_ASSERT(proxy, status); + + proxy_count = 0; + for (tproxy = proxy->pg->proxy_list; tproxy; tproxy = tproxy->next) + proxy_count++; + + proxy->pg->barrier_count++; + if (proxy->pg->barrier_count == proxy_count) { + proxy->pg->barrier_count = 0; + + bcast_keyvals(fd, pid); for (tproxy = proxy->pg->proxy_list; tproxy; tproxy = tproxy->next) { - /* complete barrier */ status = cmd_response(tproxy->control_fd, pid, "cmd=barrier_out\n"); HYDU_ERR_POP(status, "error writing PMI line\n"); } } fn_exit: - if (tmp) - HYDU_FREE(tmp); HYDU_FUNC_EXIT(); return status; @@ -510,6 +529,10 @@ static HYD_status fn_spawn(int fd, int pid, int pgid, char *args[]) HYDU_FREE(cmd); } + /* Cache the pre-initialized keyvals on the new proxies */ + if (preput_num) + bcast_keyvals(fd, pid); + fn_exit: HYD_pmcd_pmi_free_tokens(tokens, token_count); HYDU_free_strlist(proxy_args); diff --git a/src/pm/hydra/pm/pmiserv/pmiserv_utils.c b/src/pm/hydra/pm/pmiserv/pmiserv_utils.c index 46539da..45222aa 100644 --- a/src/pm/hydra/pm/pmiserv/pmiserv_utils.c +++ b/src/pm/hydra/pm/pmiserv/pmiserv_utils.c @@ -555,6 +555,8 @@ HYD_status HYD_pmcd_pmi_alloc_pg_scratch(struct HYD_pg *pg) status = HYD_pmcd_pmi_allocate_kvs(&pg_scratch->kvs, pg->pgid); HYDU_ERR_POP(status, "unable to allocate kvs space\n"); + pg_scratch->keyval_dist_count = 0; + fn_exit: HYDU_FUNC_EXIT(); return status; ----------------------------------------------------------------------- Summary of changes: src/pm/hydra/pm/pmiserv/pmiserv_pmi.h | 1 + src/pm/hydra/pm/pmiserv/pmiserv_pmi_v1.c | 149 +++++++++++++++++------------- src/pm/hydra/pm/pmiserv/pmiserv_utils.c | 2 + 3 files changed, 89 insertions(+), 63 deletions(-) hooks/post-receive -- MPICH primary repository

1 0

[mpich] MPICH primary repository branch, master, updated. v3.0.3-39-g1c156ac
by noreply＠mpich.org 22 Apr '13

22 Apr '13

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "MPICH primary repository". The branch, master has been updated via 1c156acf01ea5ed49252c10d2995d48f3f34c09f (commit) via 2bdd13ba0f7ff35fd567b0fd47862477e27ba811 (commit) from fec2569df449da7e1266e83be8900ef17f18456d (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://git.mpich.org/mpich.git/commitdiff/1c156acf01ea5ed49252c10d2995d48f3… commit 1c156acf01ea5ed49252c10d2995d48f3f34c09f Author: William Gropp <wgropp(a)illinois.edu> Date: Tue Mar 12 09:43:44 2013 -0500 Removed and fixed some FIXMEs diff --git a/src/mpi/errhan/errutil.c b/src/mpi/errhan/errutil.c index 2637fc0..0420fb1 100644 --- a/src/mpi/errhan/errutil.c +++ b/src/mpi/errhan/errutil.c @@ -406,13 +406,20 @@ static void handleFatalError( MPID_Comm *comm_ptr, { /* Define length of the the maximum error message line (or string with newlines?). This definition is used only within this routine. */ - /* FIXME: This should really be the same as MPI_MAX_ERROR_STRING, or in the - worst case, defined in terms of that */ + /* Ensure that the error message string is sufficiently long to + hold enough information about the error. Use the size of the + MPI error messages unless it is too short (defined as shown here) */ +#if MPI_MAX_ERROR_STRING < 4096 #define MAX_ERRMSG_STRING 4096 - char error_msg[ MAX_ERRMSG_STRING ]; +#else +#define MAX_ERRMSG_STRING MPI_MAX_ERROR_STRING +#endif + char error_msg[MAX_ERRMSG_STRING]; int len; - /* FIXME: Not internationalized */ + /* FIXME: Not internationalized. Since we are using MPIR_Err_get_string, + we are assuming that the code is still able to execute a full + MPICH error code to message conversion. */ MPIU_Snprintf(error_msg, MAX_ERRMSG_STRING, "Fatal error in %s: ", fcname); len = (int)strlen(error_msg); MPIR_Err_get_string(errcode, &error_msg[len], MAX_ERRMSG_STRING-len, NULL); @@ -491,8 +498,8 @@ int MPIR_Err_combine_codes(int error1, int error2) error2_class = MPI_ERR_OTHER; } - /* FIXME: This is the only difference between the two versions of - CombineCodes */ + /* Note that this call may simply discard an error code if the error + message level does not support multiple codes */ CombineSpecificCodes( error1_code, error2_code, error2_class ); if (MPIR_ERR_GET_CLASS(error1_code) == MPI_ERR_OTHER) @@ -856,7 +863,7 @@ int MPIR_Err_create_code_valist( int lastcode, int fatal, const char fcname[], reason = checkErrcodeIsValid(lastcode); if (reason) { /* --BEGIN ERROR HANDLING-- */ - MPIU_Error_printf( "Internal Error: invalid error code %x (%s) in %s:%d\n", + MPIU_Error_printf( "INTERNAL ERROR: invalid error code %x (%s) in %s:%d\n", lastcode, ErrcodeInvalidReasonStr( reason ), fcname, line ); lastcode = MPI_SUCCESS; @@ -933,9 +940,7 @@ int MPIR_Err_create_code_valist( int lastcode, int fatal, const char fcname[], { if (generic_msg[0] == '*' && generic_msg[1] == '*') { - /* FIXME : Internal error. Generate some debugging - information; Fix for the general release */ - fprintf( stderr, "Could not find %s in list of messages\n", generic_msg ); + MPIU_Error_printf( "INTERNAL ERROR: Could not find %s in list of messages\n", generic_msg ); } } # endif /* DBG_OUTPUT */ http://git.mpich.org/mpich.git/commitdiff/2bdd13ba0f7ff35fd567b0fd47862477e… commit 2bdd13ba0f7ff35fd567b0fd47862477e27ba811 Author: William Gropp <wgropp(a)illinois.edu> Date: Fri Mar 8 08:41:16 2013 -0600 Begin cleaning up error reporting diff --git a/src/mpi/errhan/errutil.c b/src/mpi/errhan/errutil.c index bbd7627..2637fc0 100644 --- a/src/mpi/errhan/errutil.c +++ b/src/mpi/errhan/errutil.c @@ -41,13 +41,16 @@ * The first group provides memory for the MPID_Errhandler objects * and the routines to free and manipulate them * - * MPIR_Err_return_xxx - For each of the MPI types on which an + * The second group provides routines to call the appropriate error handler, + * which may be predefined or user defined. These also return the + * appropriate return code. These routines have names of the form + * MPIR_Err_return_xxx. Specifically, for each of the MPI types on which an * error handler can be defined, there is an MPIR_Err_return_xxx routine * that determines what error handler function to call and whether to * abort the program. The comm and win versions are here; ROMIO * provides its own routines for invoking the error handlers for Files. * - * The next group of code handles the error messages. There are four + * The third group of code handles the error messages. There are four * options, controlled by the value of MPICH_ERROR_MSG_LEVEL. * * MPICH_ERROR_MSG_NONE - No text messages at all @@ -77,17 +80,16 @@ static int did_err_init = FALSE; /* helps us solve a bootstrapping problem */ /* A few prototypes. These routines are called from the MPIR_Err_return - routines. checkValidErrcode depends on the MPICH_ERROR_MSG_LEVEL */ + routines. checkValidErrcode depends on the MPICH_ERROR_MSG_LEVEL. + If the error code is *not* valid, checkValidErrcode may replace it + with a valid value. */ -static int checkValidErrcode( int error_class, const char fcname[], - int *errcode ); -static void handleFatalError( MPID_Comm *comm_ptr, - const char fcname[], int errcode ); +static int checkValidErrcode( int, const char [], int * ); +static void handleFatalError( MPID_Comm *, const char [], int ); #if MPICH_ERROR_MSG_LEVEL >= MPICH_ERROR_MSG_ALL -static int ErrGetInstanceString( int errorcode, char *msg, int num_remaining ); +static int ErrGetInstanceString( int, char [], int ); static void MPIR_Err_stack_init( void ); -static void CombineSpecificCodes( int, int, int ); static int checkForUserErrcode( int ); #else /* We only need special handling for user error codes when we support the @@ -98,7 +100,7 @@ static int checkForUserErrcode( int ); /* ------------------------------------------------------------------------- */ /* Provide the MPID_Errhandler space and the routines to free and set them - from C++ */ + from C++ and Fortran */ /* ------------------------------------------------------------------------- */ /* * Error handlers. These are handled just like the other opaque objects @@ -123,6 +125,22 @@ void MPID_Errhandler_free(MPID_Errhandler *errhan_ptr) MPIU_Handle_obj_free(&MPID_Errhandler_mem, errhan_ptr); } +void MPIR_Err_init( void ) +{ + /* these are "stub" objects, so the other fields (which are statically + * initialized to zero) don't really matter */ + MPID_Errhandler_builtin[0].handle = MPI_ERRORS_ARE_FATAL; + MPID_Errhandler_builtin[1].handle = MPI_ERRORS_RETURN; + MPID_Errhandler_builtin[2].handle = MPIR_ERRORS_THROW_EXCEPTIONS; + +# if MPICH_ERROR_MSG_LEVEL >= MPICH_ERROR_MSG_ALL + MPIR_Err_stack_init(); +# endif + did_err_init = TRUE; +} + +/* Language Callbacks */ + #ifdef HAVE_CXX_BINDING /* This routine is used to install a callback used by the C++ binding to invoke the (C++) error handler. The callback routine is a C routine, @@ -148,11 +166,13 @@ void MPIR_Errhandler_set_fc( MPI_Errhandler errhand ) } #endif + /* ------------------------------------------------------------------------- */ -/* These routines are called on error exit from most top-level MPI routines - to invoke the appropriate error handler. Also included is the routine - to call if MPI has not been initialized (MPIR_Err_preinit) and to - determine if an error code represents a fatal error (MPIR_Err_is_fatal). */ +/* Group 2: These routines are called on error exit from most + top-level MPI routines to invoke the appropriate error handler. + Also included is the routine to call if MPI has not been + initialized (MPIR_Err_preinit) and to determine if an error code + represents a fatal error (MPIR_Err_is_fatal). */ /* ------------------------------------------------------------------------- */ /* Special error handler to call if we are not yet initialized, or if we have finalized */ @@ -367,37 +387,47 @@ int MPIR_Err_return_win( MPID_Win *win_ptr, const char fcname[], int errcode ) return errcode; } -void MPIR_Err_init( void ) -{ - /* these are "stub" objects, so the other fields (which are statically - * initialized to zero) don't really matter */ - MPID_Errhandler_builtin[0].handle = MPI_ERRORS_ARE_FATAL; - MPID_Errhandler_builtin[1].handle = MPI_ERRORS_RETURN; - MPID_Errhandler_builtin[2].handle = MPIR_ERRORS_THROW_EXCEPTIONS; - -# if MPICH_ERROR_MSG_LEVEL >= MPICH_ERROR_MSG_ALL - MPIR_Err_stack_init(); -# endif - did_err_init = TRUE; -} - /* ------------------------------------------------------------------------- */ +/* Group 3: Routines to handle error messages. These are organized into + * several subsections: + * General service routines (used by more than one error reporting level) + * Routines of specific error message levels + * + */ /* ------------------------------------------------------------------------- */ -static int checkErrcodeIsValid( int errcode ); -static const char *ErrcodeInvalidReasonStr( int reason ); -static const char *get_class_msg( int error_class ); +/* Forward reference */ +static void CombineSpecificCodes( int, int, int ); +static const char *get_class_msg( int ); +/* --BEGIN ERROR HANDLING-- */ +/* This routine is called when there is a fatal error */ +static void handleFatalError( MPID_Comm *comm_ptr, + const char fcname[], int errcode ) +{ + /* Define length of the the maximum error message line (or string with + newlines?). This definition is used only within this routine. */ + /* FIXME: This should really be the same as MPI_MAX_ERROR_STRING, or in the + worst case, defined in terms of that */ +#define MAX_ERRMSG_STRING 4096 + char error_msg[ MAX_ERRMSG_STRING ]; + int len; -/* ------------------------------------------------------------------------- */ -/* The following block of code manages the instance-specific error messages */ -/* ------------------------------------------------------------------------- */ + /* FIXME: Not internationalized */ + MPIU_Snprintf(error_msg, MAX_ERRMSG_STRING, "Fatal error in %s: ", fcname); + len = (int)strlen(error_msg); + MPIR_Err_get_string(errcode, &error_msg[len], MAX_ERRMSG_STRING-len, NULL); + /* The third argument is a return code, a value of 1 usually indicates + an error */ + MPID_Abort(comm_ptr, MPI_SUCCESS, 1, error_msg); +} +/* --END ERROR HANDLING-- */ - /* Check for a valid error code. If the code is not valid, attempt to print out something sensible; reset the error code to have class ERR_UNKNOWN */ /* FIXME: Now that error codes are chained, this does not produce a valid error code since there is no valid ring index corresponding to this code */ +/* FIXME: No one uses the return value */ static int checkValidErrcode( int error_class, const char fcname[], int *errcode_p ) { @@ -419,6 +449,7 @@ static int checkValidErrcode( int error_class, const char fcname[], } else { + /* FIXME: The error stack comment only applies to MSG_ALL */ MPIU_Error_printf("INTERNAL ERROR: Invalid error class (%d) encountered while returning from\n" "%s. Please file a bug report. No error stack is available.\n", error_class, fcname); } @@ -432,177 +463,18 @@ static int checkValidErrcode( int error_class, const char fcname[], return rc; } -/* Check that an encoded error code is valid. Return 0 if valid, positive, - non-zero if invalid. Value indicates reason; see - ErrcodeInvalidReasonStr() */ - -#if MPICH_ERROR_MSG_LEVEL <= MPICH_ERROR_MSG_GENERIC -/* This is the shortened version when there are no error messages */ -static int checkErrcodeIsValid( int errcode ) -{ - /* MPICH_ERR_LAST_CLASS is the last of the *predefined* error classes. */ - /* FIXME: Should this check against dynamically-created error classes? */ - if (errcode < 0 || errcode >= MPICH_ERR_LAST_CLASS) { - return 3; - } - return 0; -} -static const char *ErrcodeInvalidReasonStr( int reason ) -{ - const char *str = 0; - - /* FIXME: These strings need to be internationalized */ - switch (reason) { - case 3: - str = "Message class is out of range"; - break; - default: - str = "Unknown reason for invalid errcode"; - break; - } - return str; -} -#endif /* MPICH_ERROR_MSG_LEVEL <= MPICH_ERROR_MSG_GENERIC */ - -/* --BEGIN ERROR HANDLING-- */ -/* This routine is called when there is a fatal error */ -static void handleFatalError( MPID_Comm *comm_ptr, - const char fcname[], int errcode ) -{ - /* Define length of the the maximum error message line (or string with - newlines?). This definition is used only within this routine. */ - /* FIXME: This should really be the same as MPI_MAX_ERROR_STRING, or in the - worst case, defined in terms of that */ -#define MAX_ERRMSG_STRING 4096 - char error_msg[ MAX_ERRMSG_STRING ]; - int len; - - /* FIXME: Not internationalized */ - MPIU_Snprintf(error_msg, MAX_ERRMSG_STRING, "Fatal error in %s: ", fcname); - len = (int)strlen(error_msg); - MPIR_Err_get_string(errcode, &error_msg[len], MAX_ERRMSG_STRING-len, NULL); - /* The third argument is a return code, a value of 1 usually indicates - an error */ - MPID_Abort(comm_ptr, MPI_SUCCESS, 1, error_msg); -} -/* --END ERROR HANDLING-- */ - -#if MPICH_ERROR_MSG_LEVEL >= MPICH_ERROR_MSG_GENERIC -/* - * Given a message string abbreviation (e.g., one that starts "**"), return - * the corresponding index. For the generic (non - * parameterized messages), use idx = FindGenericMsgIndex( "**msg" ); - * Returns -1 on failure to find the matching message - */ -static int FindGenericMsgIndex( const char *msg ) -{ - int i, c; - for (i=0; i<generic_msgs_len; i++) { - /* Check the sentinals to insure that the values are ok first */ - if (generic_err_msgs[i].sentinal1 != 0xacebad03 || - generic_err_msgs[i].sentinal2 != 0xcb0bfa11) { - /* Something bad has happened! Don't risk trying the - short_name pointer; it may have been corrupted */ - break; - } - c = strcmp( generic_err_msgs[i].short_name, msg ); - if (c == 0) return i; - if (c > 0) - { - /* don't return here if the string partially matches */ - /* FIXME: Why return failure instead of continuing to check? */ - if (strncmp(generic_err_msgs[i].short_name, msg, strlen(msg)) != 0) - return -1; - } - } - /* --BEGIN ERROR HANDLING-- */ - return -1; - /* --END ERROR HANDLING-- */ -} - -/* - Here is an alternate search routine based on bisection. - int i_low, i_mid, i_high, c; - i_low = 0; i_high = generic_msg_len - 1; - while (i_high - i_low >= 0) { - i_mid = (i_high + i_low) / 2; - c = strcmp( generic_err_msgs[i].short_name, msg ); - if (c == 0) return i_mid; - if (c < 0) { i_low = i_mid + 1; } - else { i_high = i_mid - 1; } - } - return -1; -*/ -#endif /* MPICH_ERROR_MSG_LEVEL >= MPICH_ERROR_MSG_GENERIC */ - -#if MPICH_ERROR_MSG_LEVEL == MPICH_ERROR_MSG_ALL -#endif /* MPICH_ERROR_MSG_LEVEL == MPICH_ERROR_MSG_ALL */ - - -/* ------------------------------------------------------------------------ */ -/* The following routines create an MPI error code, handling optional, */ -/* instance-specific error message information. There are two key routines:*/ -/* MPIR_Err_create_code - Create the error code; this is the routine used*/ -/* by most routines */ -/* MPIR_Err_create_code_valist - Create the error code; accept a valist */ -/* instead of a variable argument list (this is */ -/* used to allow this routine to be used from */ -/* within another varargs routine) */ -/* ------------------------------------------------------------------------ */ - -/* Err_create_code is just a shell that accesses the va_list and then - calls the real routine. */ -int MPIR_Err_create_code( int lastcode, int fatal, const char fcname[], - int line, int error_class, const char generic_msg[], - const char specific_msg[], ... ) -{ - int rc; - va_list Argp; - va_start(Argp, specific_msg); - MPIU_DBG_MSG_FMT(ERRHAND, TYPICAL, (MPIU_DBG_FDEST, "%sError created: last=%#010x class=%#010x %s(%d) %s", - fatal ? "Fatal " : "", lastcode, error_class, fcname, line, generic_msg)); - rc = MPIR_Err_create_code_valist( lastcode, fatal, fcname, line, - error_class, generic_msg, specific_msg, - Argp ); - va_end(Argp); - return rc; -} - -#if MPICH_ERROR_MSG_LEVEL < MPICH_ERROR_MSG_ALL -/* In this case, the routine ignores all but (possibly) the generic message. - It also returns lastcode *unless* lastcode is MPI_SUCCESS. Thus, the - error returned is the error specified at the point of detection. - */ -int MPIR_Err_create_code_valist( int lastcode, int fatal, const char fcname[], - int line, int error_class, - const char generic_msg[], - const char specific_msg[], va_list Argp ) -{ -#if MPICH_ERROR_MSG_LEVEL == MPICH_ERROR_MSG_GENERIC - int generic_idx; - int errcode = lastcode; - if (lastcode == MPI_SUCCESS) { - generic_idx = FindGenericMsgIndex(generic_msg); - if (generic_idx >= 0) { - errcode = (generic_idx << ERROR_GENERIC_SHIFT) | error_class; - if (fatal) - errcode |= ERROR_FATAL_MASK; - } - } - return errcode; -#else - int errcode = lastcode; - if (lastcode == MPI_SUCCESS) errcode = error_class; - return errcode; -#endif /* MSG_LEVEL == GENERIC */ -} -/* FIXME: Factor the error routines so that there is a single version of each */ +/* Append an error code, error2, to the end of a list of messages in the error + ring whose head endcoded in error1_code. An error code pointing at the + combination is returned. If the list of messages does not terminate cleanly + (i.e. ring wrap has occurred), then the append is not performed. and error1 + is returned (although it may include the class of error2 if the class of + error1 was MPI_ERR_OTHER). */ int MPIR_Err_combine_codes(int error1, int error2) { int error1_code = error1; int error2_code = error2; int error2_class; - + /* If either error code is success, return the other */ if (error1_code == MPI_SUCCESS) return error2_code; if (error2_code == MPI_SUCCESS) return error1_code; @@ -611,7 +483,7 @@ int MPIR_Err_combine_codes(int error1, int error2) error1. */ if (error1_code & ERROR_DYN_MASK) return error1_code; if (error2_code & ERROR_DYN_MASK) return error2_code; - + error2_class = MPIR_ERR_GET_CLASS(error2_code); if (MPIR_ERR_GET_CLASS(error2_class) < MPI_SUCCESS || MPIR_ERR_GET_CLASS(error2_class) > MPICH_ERR_LAST_CLASS) @@ -619,8 +491,10 @@ int MPIR_Err_combine_codes(int error1, int error2) error2_class = MPI_ERR_OTHER; } - /* If the class of error1 is OTHER, replace it with the class of - error2 */ + /* FIXME: This is the only difference between the two versions of + CombineCodes */ + CombineSpecificCodes( error1_code, error2_code, error2_class ); + if (MPIR_ERR_GET_CLASS(error1_code) == MPI_ERR_OTHER) { error1_code = (error1_code & ~(ERROR_CLASS_MASK)) | error2_class; @@ -629,45 +503,24 @@ int MPIR_Err_combine_codes(int error1, int error2) return error1_code; } -#endif /* msg_level < msg_all */ - -/* - * Accessor routines for the predefined mqessages. These can be - * used by the other routines (such as MPI_Error_string) to - * access the messages in this file, or the messages that may be - * available through any message catalog facility - */ -static const char *get_class_msg( int error_class ) -{ -#if MPICH_ERROR_MSG_LEVEL > MPICH_ERROR_MSG_NONE - if (error_class >= 0 && error_class < MPIR_MAX_ERROR_CLASS_INDEX) { -#if MPICH_ERROR_MSG_LEVEL == MPICH_ERROR_MSG_CLASS - return classToMsg[error_class]; -#else - return generic_err_msgs[class_to_index[error_class]].long_name; -#endif - } - else { - /* --BEGIN ERROR HANDLING-- */ - return "Unknown error class"; - /* --END ERROR HANDLING-- */ - } -#else - /* FIXME: Not internationalized */ - return "Error message texts are not available"; -#endif /* MSG_LEVEL > MSG_NONE */ -} - /* FIXME: This routine isn't quite right yet */ /* * Notes: * One complication is that in the instance-specific case, a ?? */ +/* + * Given an errorcode, place the corresponding message in msg[length]. + * The argument fn must be NULL and is otherwise ignored. + */ void MPIR_Err_get_string( int errorcode, char * msg, int length, MPIR_Err_get_class_string_func_t fn ) { int error_class; int len, num_remaining = length; + + /* The fn (fourth) argument was added improperly and is no longer + used. */ + MPIU_Assert( fn == NULL ); /* There was code to set num_remaining to MPI_MAX_ERROR_STRING if it was zero. But based on the usage of this routine, @@ -690,10 +543,10 @@ void MPIR_Err_get_string( int errorcode, char * msg, int length, */ if (errorcode & ERROR_DYN_MASK) { /* This is a dynamically created error code (e.g., with - MPI_Err_add_class) */ - /* If a dynamic error code was created, the function to convert - them into strings has been set. Check to see that it was; this - is a safeguard against a bogus error code */ + MPI_Err_add_class). If a dynamic error code was created, + the function to convert them into strings has been set. + Check to see that it was; this is a safeguard against a + bogus error code */ if (!MPIR_Process.errcode_to_string) { /* FIXME: not internationalized */ /* --BEGIN ERROR HANDLING-- */ @@ -716,33 +569,19 @@ void MPIR_Err_get_string( int errorcode, char * msg, int length, else if ( (errorcode & ERROR_CLASS_MASK) == errorcode) { error_class = MPIR_ERR_GET_CLASS(errorcode); - /* FIXME: Why was the last test commented out? */ - if (fn != NULL && error_class > MPICH_ERR_LAST_CLASS /*&& error_class < MPICH_ERR_MAX_EXT_CLASS*/) - { - fn(errorcode, msg, length); - } - else - { - if (MPIU_Strncpy(msg, get_class_msg( errorcode ), num_remaining)) + if (MPIU_Strncpy(msg, get_class_msg( errorcode ), num_remaining)) { msg[num_remaining - 1] = '\0'; } - } } else { /* print the class message first */ + /* FIXME: Why print the class message first? The instance + message is supposed to be complete by itself. */ error_class = MPIR_ERR_GET_CLASS(errorcode); - /* FIXME: Why was the last test commented out? */ - if (fn != NULL && error_class > MPICH_ERR_LAST_CLASS /*&& error_class < MPICH_ERR_MAX_EXT_CLASS*/) - { - fn(errorcode, msg, num_remaining); - } - else - { - MPIU_Strncpy(msg, get_class_msg(ERROR_GET_CLASS(errorcode)), num_remaining); - } + MPIU_Strncpy(msg, get_class_msg(error_class), num_remaining); msg[num_remaining - 1] = '\0'; len = (int)strlen(msg); msg += len; @@ -750,6 +589,8 @@ void MPIR_Err_get_string( int errorcode, char * msg, int length, /* then print the stack or the last specific error message */ + /* FIXME: Replace with function to add instance string or + error code string */ # if MPICH_ERROR_MSG_LEVEL >= MPICH_ERROR_MSG_ALL if (ErrGetInstanceString( errorcode, msg, num_remaining )) goto fn_exit; @@ -772,18 +613,139 @@ fn_exit: return; } -/* - * If the error message level is all, MPICH supports instance-specific - * error messages. Details above - */ -#if MPICH_ERROR_MSG_LEVEL == MPICH_ERROR_MSG_ALL +#if MPICH_ERROR_MSG_LEVEL == MPICH_ERROR_MSG_NONE +/* No error message support */ +int MPIR_Err_create_code( int lastcode, int fatal, const char fcname[], + int line, int error_class, const char generic_msg[], + const char specific_msg[], ... ) +{ + MPIU_DBG_MSG_FMT(ERRHAND, TYPICAL, (MPIU_DBG_FDEST, "%sError created: last=%#010x class=%#010x %s(%d) %s", + fatal ? "Fatal " : "", lastcode, error_class, fcname, line, generic_msg)); + return (lastcode == MPI_SUCCESS) ? error_class : lastcode; +} +int MPIR_Err_create_code_valist( int lastcode, int fatal, const char fcname[], + int line, int error_class, + const char generic_msg[], + const char specific_msg[], va_list Argp ) +{ + return (lastcode == MPI_SUCCESS) ? error_class : lastcode; +} +/* Internal routines */ +static void CombineSpecificCodes( int error1_code, int error2_code, + int error2_class ) +{ +} +static const char *get_class_msg( int error_class ) +{ + /* FIXME: Not internationalized */ + return "Error message texts are not available"; +} -static int convertErrcodeToIndexes( int errcode, int *ring_idx, int *ring_id, - int *generic_idx ); +#elif MPICH_ERROR_MSG_LEVEL == MPICH_ERROR_MSG_CLASS +/* Only class error messages. Note this is nearly the same as + MPICH_ERROR_MSG_LEVEL == NONE, since the handling of error codes + is the same */ +int MPIR_Err_create_code( int lastcode, int fatal, const char fcname[], + int line, int error_class, const char generic_msg[], + const char specific_msg[], ... ) +{ + MPIU_DBG_MSG_FMT(ERRHAND, TYPICAL, (MPIU_DBG_FDEST, "%sError created: last=%#010x class=%#010x %s(%d) %s", + fatal ? "Fatal " : "", lastcode, error_class, fcname, line, generic_msg)); + return (lastcode == MPI_SUCCESS) ? error_class : lastcode; +} +int MPIR_Err_create_code_valist( int lastcode, int fatal, const char fcname[], + int line, int error_class, + const char generic_msg[], + const char specific_msg[], va_list Argp ) +{ + return (lastcode == MPI_SUCCESS) ? error_class : lastcode; +} + +static void CombineSpecificCodes( int error1_code, int error2_code, + int error2_class ) +{ +} +static const char *get_class_msg( int error_class ) +{ + if (error_class >= 0 && error_class < MPIR_MAX_ERROR_CLASS_INDEX) { + return classToMsg[error_class]; + } + else { + /* --BEGIN ERROR HANDLING-- */ + return "Unknown error class"; + /* --END ERROR HANDLING-- */ + } +} + +#elif MPICH_ERROR_MSG_LEVEL == MPICH_ERROR_MSG_GENERIC +#define NEEDS_FIND_GENERIC_MSG_INDEX +static int FindGenericMsgIndex( const char [] ); + +/* Only generic error messages (more than class, but all predefined) */ + +int MPIR_Err_create_code( int lastcode, int fatal, const char fcname[], + int line, int error_class, const char generic_msg[], + const char specific_msg[], ... ) +{ + int rc; + va_list Argp; + va_start(Argp, specific_msg); + MPIU_DBG_MSG_FMT(ERRHAND, TYPICAL, (MPIU_DBG_FDEST, "%sError created: last=%#010x class=%#010x %s(%d) %s", + fatal ? "Fatal " : "", lastcode, error_class, fcname, line, generic_msg)); + rc = MPIR_Err_create_code_valist( lastcode, fatal, fcname, line, + error_class, generic_msg, specific_msg, + Argp ); + va_end(Argp); + return rc; +} +int MPIR_Err_create_code_valist( int lastcode, int fatal, const char fcname[], + int line, int error_class, + const char generic_msg[], + const char specific_msg[], va_list Argp ) +{ + int generic_idx; + int errcode = lastcode; + if (lastcode == MPI_SUCCESS) { + generic_idx = FindGenericMsgIndex(generic_msg); + if (generic_idx >= 0) { + errcode = (generic_idx << ERROR_GENERIC_SHIFT) | error_class; + if (fatal) + errcode |= ERROR_FATAL_MASK; + } + } + return errcode; +} +static void CombineSpecificCodes( int error1_code, int error2_code, + int error2_class ) +{ +} +static const char *get_class_msg( int error_class ) +{ + if (error_class >= 0 && error_class < MPIR_MAX_ERROR_CLASS_INDEX) { + return generic_err_msgs[class_to_index[error_class]].long_name; + } + else { + /* --BEGIN ERROR HANDLING-- */ + return "Unknown error class"; + /* --END ERROR HANDLING-- */ + } +} +#elif MPICH_ERROR_MSG_LEVEL == MPICH_ERROR_MSG_ALL +/* General error message support, including the error message stack */ -/* FIXME: Where is the documentation for this function? What is it for? */ +static int checkErrcodeIsValid( int ); +static const char *ErrcodeInvalidReasonStr( int ); +#define NEEDS_FIND_GENERIC_MSG_INDEX +static int FindGenericMsgIndex( const char [] ); +static int FindSpecificMsgIndex( const char [] ); +static int vsnprintf_mpi(char *str, size_t maxlen, const char *fmt_orig, + va_list list); +static void ErrcodeCreateID( int error_class, int generic_idx, + const char *msg, int *id, int *seq ); +static int convertErrcodeToIndexes( int errcode, int *ring_idx, int *ring_id, + int *generic_idx ); static void MPIR_Err_print_stack_string(int errcode, char *str, int maxlen); #define MAX_ERROR_RING ERROR_SPECIFIC_INDEX_SIZE @@ -854,131 +816,479 @@ static MPID_Thread_mutex_t error_ring_mutex; #define error_ring_mutex_unlock() #endif /* REQUIRES_THREAD_SAFETY */ -/* Create the ring id from information about the message */ -static void ErrcodeCreateID( int error_class, int generic_idx, - const char *msg, int *id, int *seq ) + +int MPIR_Err_create_code( int lastcode, int fatal, const char fcname[], + int line, int error_class, const char generic_msg[], + const char specific_msg[], ... ) { - int i; - int ring_seq = 0, ring_id; + int rc; + va_list Argp; + va_start(Argp, specific_msg); + MPIU_DBG_MSG_FMT(ERRHAND, TYPICAL, (MPIU_DBG_FDEST, "%sError created: last=%#010x class=%#010x %s(%d) %s", + fatal ? "Fatal " : "", lastcode, error_class, fcname, line, generic_msg)); + rc = MPIR_Err_create_code_valist( lastcode, fatal, fcname, line, + error_class, generic_msg, specific_msg, + Argp ); + va_end(Argp); + return rc; +} +/* + * This is the real routine for generating an error code. It takes + * a va_list so that it can be called by any routine that accepts a + * variable number of arguments. + */ +int MPIR_Err_create_code_valist( int lastcode, int fatal, const char fcname[], + int line, int error_class, + const char generic_msg[], + const char specific_msg[], va_list Argp ) +{ + int err_code; + int generic_idx; + int use_user_error_code = 0; + int user_error_code = -1; + char user_ring_msg[MPIR_MAX_ERROR_LINE+1]; - /* Create a simple hash function of the message to serve as the - sequence number */ - ring_seq = 0; - for (i=0; msg[i]; i++) - ring_seq += (unsigned int) msg[i]; + /* Create the code from the class and the message ring index */ - ring_seq %= ERROR_SPECIFIC_SEQ_SIZE; + /* Check that lastcode is valid */ + if (lastcode != MPI_SUCCESS) { + int reason; + reason = checkErrcodeIsValid(lastcode); + if (reason) { + /* --BEGIN ERROR HANDLING-- */ + MPIU_Error_printf( "Internal Error: invalid error code %x (%s) in %s:%d\n", + lastcode, ErrcodeInvalidReasonStr( reason ), + fcname, line ); + lastcode = MPI_SUCCESS; + /* --END ERROR HANDLING-- */ + } + } - ring_id = (error_class & ERROR_CLASS_MASK) | - ((generic_idx + 1) << ERROR_GENERIC_SHIFT) | - (ring_seq << ERROR_SPECIFIC_SEQ_SHIFT); + /* FIXME: ERR_OTHER is overloaded; this may mean "OTHER" or it may + mean "No additional error, just routine stack info" */ + if (error_class == MPI_ERR_OTHER) + { + if (MPIR_ERR_GET_CLASS(lastcode) > MPI_SUCCESS && + MPIR_ERR_GET_CLASS(lastcode) <= MPICH_ERR_LAST_CLASS) + { + /* If the last class is more specific (and is valid), then pass it + through */ + error_class = MPIR_ERR_GET_CLASS(lastcode); + } + else + { + error_class = MPI_ERR_OTHER; + } + } - *id = ring_id; - *seq = ring_seq; -} + /* Handle special case of MPI_ERR_IN_STATUS. According to the standard, + the code must be equal to the class. See section 3.7.5. + Information on the particular error is in the MPI_ERROR field + of the status. */ + if (error_class == MPI_ERR_IN_STATUS) + { + return MPI_ERR_IN_STATUS; + } -/* Convert an error code into ring_idx, ring_id, and generic_idx. - Return non-zero if there is a problem with the decode values - (e.g., out of range for the ring index) */ -static int convertErrcodeToIndexes( int errcode, int *ring_idx, int *ring_id, - int *generic_idx ) -{ - *ring_idx = (errcode & ERROR_SPECIFIC_INDEX_MASK) >> - ERROR_SPECIFIC_INDEX_SHIFT; - *ring_id = errcode & (ERROR_CLASS_MASK | - ERROR_GENERIC_MASK | ERROR_SPECIFIC_SEQ_MASK); - *generic_idx = ((errcode & ERROR_GENERIC_MASK) >> ERROR_GENERIC_SHIFT) - 1; - - /* Test on both the max_error_ring_loc and MAX_ERROR_RING to guard - against memory overwrites */ - if (*ring_idx < 0 || *ring_idx >= MAX_ERROR_RING || - *ring_idx > max_error_ring_loc) return 1; + err_code = error_class; - return 0; -} -static int checkErrcodeIsValid( int errcode ) -{ - int ring_id, generic_idx, ring_idx; + /* Handle the generic message. This selects a subclass, based on a text + string */ + generic_idx = FindGenericMsgIndex(generic_msg); + if (generic_idx >= 0) { + if (strcmp( generic_err_msgs[generic_idx].short_name, "**user" ) == 0) { + use_user_error_code = 1; + /* This is a special case. The format is + "**user", "**userxxx %d", intval + (generic, specific, parameter). In this + case we must ... save the user value because + we store it explicitly in the ring. + We do this here because we cannot both access the + user error code and pass the argp to vsnprintf_mpi . */ + if (specific_msg) { + const char *specific_fmt; + int specific_idx; + user_error_code = va_arg(Argp,int); + specific_idx = FindSpecificMsgIndex(specific_msg); + if (specific_idx >= 0) { + specific_fmt = specific_err_msgs[specific_idx].long_name; + } + else { + specific_fmt = specific_msg; + } + MPIU_Snprintf( user_ring_msg, sizeof(user_ring_msg), + specific_fmt, user_error_code ); + } + else { + user_ring_msg[0] = 0; + } + } + err_code |= (generic_idx + 1) << ERROR_GENERIC_SHIFT; + } + else { + /* TODO: lookup index for class error message */ + err_code &= ~ERROR_GENERIC_MASK; + +# ifdef MPICH_DBG_OUTPUT + { + if (generic_msg[0] == '*' && generic_msg[1] == '*') + { + /* FIXME : Internal error. Generate some debugging + information; Fix for the general release */ + fprintf( stderr, "Could not find %s in list of messages\n", generic_msg ); + } + } +# endif /* DBG_OUTPUT */ + } - /* If the errcode is a class, then it is valid */ - if (errcode <= MPIR_MAX_ERROR_CLASS_INDEX && errcode >= 0) return 0; + /* Handle the instance-specific part of the error message */ + { + int specific_idx; + const char * specific_fmt = 0; + int ring_idx, ring_seq=0; + char * ring_msg; + + error_ring_mutex_lock(); + { + /* Get the next entry in the ring; keep track of what part of the + ring is in use (max_error_ring_loc) */ + ring_idx = error_ring_loc++; + if (error_ring_loc >= MAX_ERROR_RING) + error_ring_loc %= MAX_ERROR_RING; + if (error_ring_loc > max_error_ring_loc) + max_error_ring_loc = error_ring_loc; + + ring_msg = ErrorRing[ring_idx].msg; - convertErrcodeToIndexes( errcode, &ring_idx, &ring_id, &generic_idx ); - MPIU_DBG_MSG_FMT(ERRHAND, VERBOSE, (MPIU_DBG_FDEST, "code=%#010x ring_idx=%d ring_id=%#010x generic_idx=%d", - errcode, ring_idx, ring_id, generic_idx)); + if (specific_msg != NULL) + { + specific_idx = FindSpecificMsgIndex(specific_msg); + if (specific_idx >= 0) + { + specific_fmt = specific_err_msgs[specific_idx].long_name; + } + else + { + specific_fmt = specific_msg; + } + /* See the code above for handling user errors */ + if (!use_user_error_code) { + vsnprintf_mpi( ring_msg, MPIR_MAX_ERROR_LINE, + specific_fmt, Argp ); + } + else { + MPIU_Strncpy( ring_msg, user_ring_msg, MPIR_MAX_ERROR_LINE ); + } + } + else if (generic_idx >= 0) + { + MPIU_Strncpy( ring_msg,generic_err_msgs[generic_idx].long_name, + MPIR_MAX_ERROR_LINE ); + } + else + { + MPIU_Strncpy( ring_msg, generic_msg, MPIR_MAX_ERROR_LINE ); + } - if (ring_idx < 0 || ring_idx >= MAX_ERROR_RING || - ring_idx > max_error_ring_loc) return 1; - if (ErrorRing[ring_idx].id != ring_id) return 2; - /* It looks like the code uses a generic idx of -1 to indicate no - generic message */ - if (generic_idx < -1 || generic_idx > generic_msgs_len) return 3; - return 0; -} + ring_msg[MPIR_MAX_ERROR_LINE] = '\0'; + + /* Get the ring sequence number and set the ring id */ + ErrcodeCreateID( error_class, generic_idx, ring_msg, + &ErrorRing[ring_idx].id, &ring_seq ); + /* Set the previous code. */ + ErrorRing[ring_idx].prev_error = lastcode; + + /* */ + if (use_user_error_code) + { + ErrorRing[ring_idx].use_user_error_code = 1; + ErrorRing[ring_idx].user_error_code = user_error_code; + } + else if (lastcode != MPI_SUCCESS) + { + int last_ring_idx; + int last_ring_id; + int last_generic_idx; + + if (convertErrcodeToIndexes( lastcode, &last_ring_idx, + &last_ring_id, + &last_generic_idx ) != 0) { + /* --BEGIN ERROR HANDLING-- */ + MPIU_Error_printf( + "Invalid error code (%d) (error ring index %d invalid)\n", + lastcode, last_ring_idx ); + /* --END ERROR HANDLING-- */ + } + else { + if (last_generic_idx >= 0 && + ErrorRing[last_ring_idx].id == last_ring_id) { + if (ErrorRing[last_ring_idx].use_user_error_code) { + ErrorRing[ring_idx].use_user_error_code = 1; + ErrorRing[ring_idx].user_error_code = + ErrorRing[last_ring_idx].user_error_code; + } + } + } + } + + if (fcname != NULL) + { + MPIU_Snprintf(ErrorRing[ring_idx].location, MAX_LOCATION_LEN, "%s(%d)", fcname, line); + ErrorRing[ring_idx].location[MAX_LOCATION_LEN] = '\0'; + } + else + { + ErrorRing[ring_idx].location[0] = '\0'; + } + { + MPIU_DBG_MSG_FMT(ERRHAND,VERBOSE,(MPIU_DBG_FDEST, "New ErrorRing[%d]", ring_idx)); + MPIU_DBG_MSG_FMT(ERRHAND,VERBOSE,(MPIU_DBG_FDEST, " id = %#010x", ErrorRing[ring_idx].id)); + MPIU_DBG_MSG_FMT(ERRHAND,VERBOSE,(MPIU_DBG_FDEST, " prev_error = %#010x", ErrorRing[ring_idx].prev_error)); + MPIU_DBG_MSG_FMT(ERRHAND,VERBOSE,(MPIU_DBG_FDEST, " user=%d", ErrorRing[ring_idx].use_user_error_code)); + } + } + error_ring_mutex_unlock(); + + err_code |= ring_idx << ERROR_SPECIFIC_INDEX_SHIFT; + err_code |= ring_seq << ERROR_SPECIFIC_SEQ_SHIFT; -/* --BEGIN ERROR HANDLING-- */ -static const char *ErrcodeInvalidReasonStr( int reason ) -{ - const char *str = 0; - switch (reason) { - case 1: - str = "Ring Index out of range"; - break; - case 2: - str = "Ring ids do not match"; - break; - case 3: - str = "Generic message index out of range"; - break; - default: - str = "Unknown reason for invalid errcode"; - break; } - return str; + + if (fatal || MPIR_Err_is_fatal(lastcode)) + { + err_code |= ERROR_FATAL_MASK; + } + + return err_code; } -/* --END ERROR HANDLING-- */ -/* Check to see if the error code is a user-specified error code - (e.g., from the attribute delete function) and if so, set the error code - to the value provide by the user */ -static int checkForUserErrcode( int errcode ) +/* FIXME: Shouldn't str be const char * ? - no, but you don't know that without + some documentation */ +static void MPIR_Err_print_stack_string(int errcode, char *str, int maxlen ) { + char *str_orig = str; + int len; + error_ring_mutex_lock(); { - if (errcode != MPI_SUCCESS) { + /* Find the longest fcname in the stack */ + int max_location_len = 0; + int tmp_errcode = errcode; + while (tmp_errcode != MPI_SUCCESS) { int ring_idx; int ring_id; int generic_idx; + if (convertErrcodeToIndexes( tmp_errcode, &ring_idx, &ring_id, + &generic_idx ) != 0) { + /* --BEGIN ERROR HANDLING-- */ + MPIU_Error_printf( + "Invalid error code (%d) (error ring index %d invalid)\n", + errcode, ring_idx ); + break; + /* --END ERROR HANDLING-- */ + } + + if (generic_idx < 0) { + break; + } + + if (ErrorRing[ring_idx].id == ring_id) { + len = (int)strlen(ErrorRing[ring_idx].location); + max_location_len = MPIR_MAX(max_location_len, len); + tmp_errcode = ErrorRing[ring_idx].prev_error; + } + else + { + break; + } + } + max_location_len += 2; /* add space for the ": " */ + /* print the error stack */ + while (errcode != MPI_SUCCESS) { + int ring_idx; + int ring_id; + int generic_idx; + int i; + char *cur_pos; + if (convertErrcodeToIndexes( errcode, &ring_idx, &ring_id, &generic_idx ) != 0) { /* --BEGIN ERROR HANDLING-- */ MPIU_Error_printf( - "Invalid error code (%d) (error ring index %d invalid)\n", - errcode, ring_idx ); + "Invalid error code (%d) (error ring index %d invalid)\n", + errcode, ring_idx ); /* --END ERROR HANDLING-- */ } - else { - /* Can we get a more specific error message */ - if (generic_idx >= 0 && - ErrorRing[ring_idx].id == ring_id && - ErrorRing[ring_idx].use_user_error_code) + + if (generic_idx < 0) + { + break; + } + + if (ErrorRing[ring_idx].id == ring_id) { + int nchrs; + MPIU_Snprintf(str, maxlen, "%s", ErrorRing[ring_idx].location); + len = (int)strlen(str); + maxlen -= len; + str += len; + nchrs = max_location_len - + (int)strlen(ErrorRing[ring_idx].location) - 2; + while (nchrs > 0 && maxlen > 0) { + *str++ = '.'; + nchrs--; + maxlen--; + } + if (maxlen > 0) { + *str++ = ':'; + maxlen--; + } + if (maxlen > 0) { + *str++ = ' '; + maxlen--; + } + + if (MPIR_PARAM_CHOP_ERROR_STACK > 0) + { + cur_pos = ErrorRing[ring_idx].msg; + len = (int)strlen(cur_pos); + if (len == 0 && maxlen > 0) { + *str++ = '\n'; + maxlen--; + } + while (len) { - errcode = ErrorRing[ring_idx].user_error_code; + if (len >= MPIR_PARAM_CHOP_ERROR_STACK - max_location_len) + { + if (len > maxlen) + break; + /* FIXME: Don't use Snprint to append a string ! */ + MPIU_Snprintf(str, MPIR_PARAM_CHOP_ERROR_STACK - 1 - max_location_len, "%s", cur_pos); + str[MPIR_PARAM_CHOP_ERROR_STACK - 1 - max_location_len] = '\n'; + cur_pos += MPIR_PARAM_CHOP_ERROR_STACK - 1 - max_location_len; + str += MPIR_PARAM_CHOP_ERROR_STACK - max_location_len; + maxlen -= MPIR_PARAM_CHOP_ERROR_STACK - max_location_len; + if (maxlen < max_location_len) + break; + for (i=0; i<max_location_len; i++) + { + MPIU_Snprintf(str, maxlen, " "); + maxlen--; + str++; + } + len = (int)strlen(cur_pos); + } + else + { + MPIU_Snprintf(str, maxlen, "%s\n", cur_pos); + len = (int)strlen(str); + maxlen -= len; + str += len; + len = 0; + } } + } + else + { + MPIU_Snprintf(str, maxlen, "%s\n", ErrorRing[ring_idx].msg); + len = (int)strlen(str); + maxlen -= len; + str += len; + } + errcode = ErrorRing[ring_idx].prev_error; + } + else + { + break; } } } error_ring_mutex_unlock(); - return errcode; + + if (errcode == MPI_SUCCESS) + { + goto fn_exit; + } + + /* FIXME: The following code is broken as described above (if the errcode + is not valid, then this code is just going to cause more problems) */ + { + int generic_idx; + + generic_idx = ((errcode & ERROR_GENERIC_MASK) >> ERROR_GENERIC_SHIFT) - 1; + + if (generic_idx >= 0) + { + const char *p; + /* FIXME: (Here and elsewhere) Make sure any string is + non-null before you use it */ + p = generic_err_msgs[generic_idx].long_name; + if (!p) { p = "<NULL>"; } + MPIU_Snprintf(str, maxlen, "(unknown)(): %s\n", p ); + len = (int)strlen(str); + maxlen -= len; + str += len; + goto fn_exit; + } + } + + { + int error_class; + + error_class = ERROR_GET_CLASS(errcode); + + if (error_class <= MPICH_ERR_LAST_CLASS) + { + MPIU_Snprintf(str, maxlen, "(unknown)(): %s\n", + get_class_msg(ERROR_GET_CLASS(errcode))); + len = (int)strlen(str); + maxlen -= len; + str += len; + } + else + { + /* FIXME: Not internationalized */ + MPIU_Snprintf(str, maxlen, + "Error code contains an invalid class (%d)\n", + error_class); + len = (int)strlen(str); + maxlen -= len; + str += len; + } + } + + fn_exit: + if (str_orig != str) + { + str--; + *str = '\0'; + } + return; +} + + +/* Internal Routines */ + +static const char *get_class_msg( int error_class ) +{ + if (error_class >= 0 && error_class < MPIR_MAX_ERROR_CLASS_INDEX) { + return generic_err_msgs[class_to_index[error_class]].long_name; + } + else { + /* --BEGIN ERROR HANDLING-- */ + return "Unknown error class"; + /* --END ERROR HANDLING-- */ + } } /* * Given a message string abbreviation (e.g., one that starts "**"), return * the corresponding index. For the specific * (parameterized messages), use idx = FindSpecificMsgIndex( "**msg" ); + * Note: Identical to FindGeneric, but with a different array. Should + * use a single routine. */ -static int FindSpecificMsgIndex( const char *msg ) +static int FindSpecificMsgIndex( const char msg[] ) { int i, c; for (i=0; i<specific_msgs_len; i++) { @@ -1003,6 +1313,7 @@ static int FindSpecificMsgIndex( const char *msg ) /* See FindGenericMsgIndex comments for a more efficient search routine that could be used here as well. */ +/* Support for the instance-specific messages */ /* ------------------------------------------------------------------------- */ /* Routines to convert instance-specific messages into a string */ /* This is the only case that supports instance-specific messages */ @@ -1172,6 +1483,8 @@ static const char * GetMPIOpString(MPI_Op o) return "MPI_MAXLOC"; case MPI_REPLACE: return "MPI_REPLACE"; + case MPI_NO_OP: + return "MPI_NO_OP"; } /* FIXME: default is not thread safe */ MPIU_Snprintf(default_str, sizeof(default_str), "op=0x%x", o); @@ -1283,11 +1596,11 @@ static int vsnprintf_mpi(char *str, size_t maxlen, const char *fmt_orig, case MPI_ANY_TAG: MPIU_Strncpy(str, "MPI_ANY_TAG", maxlen); break; - /* FIXME: Is MPI_UNDEFINED valid as a tag? */ - case MPI_UNDEFINED: - MPIU_Strncpy(str, "MPI_UNDEFINED", maxlen); - break; default: + /* Note that MPI_UNDEFINED is not a valid tag value, + though there is one example in the MPI-3.0 standard + that sets status.MPI_TAG to MPI_UNDEFINED in a + generalized request example. */ MPIU_Snprintf(str, maxlen, "%d", t); break; } @@ -1440,394 +1753,18 @@ static int vsnprintf_mpi(char *str, size_t maxlen, const char *fmt_orig, return mpi_errno; } -static void CombineSpecificCodes( int error1_code, int error2_code, - int error2_class ) -{ - int error_code; - - error_code = error1_code; - - error_ring_mutex_lock(); - { - for (;;) - { - int error_class; - int ring_idx; - int ring_id; - int generic_idx; - - if (convertErrcodeToIndexes(error_code, &ring_idx, &ring_id, - &generic_idx) != 0 || generic_idx < 0 || - ErrorRing[ring_idx].id != ring_id) - { - break; - } - - error_code = ErrorRing[ring_idx].prev_error; - - if (error_code == MPI_SUCCESS) - { - ErrorRing[ring_idx].prev_error = error2_code; - break; - } - - error_class = MPIR_ERR_GET_CLASS(error_code); - - if (error_class == MPI_ERR_OTHER) - { - ErrorRing[ring_idx].prev_error &= ~(ERROR_CLASS_MASK); - ErrorRing[ring_idx].prev_error |= error2_class; - } - } - } - error_ring_mutex_unlock(); -} - - /* ------------------------------------------------------------------------- */ +/* Manage the error reporting stack */ /* ------------------------------------------------------------------------- */ -/* - * Instance-specific error messages are stored in a ring. The elements of this - * ring are MPIR_Err_msg_t structures, which contain the following fields: - * id - this is used to check that the entry is valid; it is computed from - * the error code and location in the ring. The routine - * ErrcodeToId( errcode, &id ) is used to extract the id from an - * error code and - * ErrcodeCreateID( class, generic, msg, &id, &seq ) is used to - * create the id from an error class, generic index, and message - * string. The "seq" field is inserted into the error code as a - * check. - * - * prev_error - The full MPI error code of the previous error attached - * to this list of errors, or MPI_SUCCESSS (which has value 0). - * This is the last error code, not the index in the ring of the last - * error code. That's the right choice, because we want to ensure - * that the value is valid if the ring overflows. In addition, - * we allow this to be an error CLASS (one of the predefined MPI - * error classes). This is particularly important for - * MPI_ERR_IN_STATUS, which may be returned as a valid error code. - * (classes are valid error codes). - * - * use_user_error_code and user_error_code - Used to handle a few cases - * in MPI where a user-provided routine returns an error code; - * this allows us to provide information about the chain of - * routines that were involved, while returning the users prefered - * error value to the users environment. See the note below - * on user error codes. - * - * location - A string that indicates what function and line number - * where the error code was set. - * - * msg - A message about the error. This may be instance-specific (e.g., - * it may have been created at the time the error was detected with - * information about the parameters that caused the error). - * - * Note that both location and msg are defined as length MAX_xxx+1. This - * isn't really necessary (at least for msg), since the MPI standard - * requires that MAX_MPI_ERROR_STRING include the space for the trailing null, - * but using the extra byte makes the code a little simpler. - * - * The "id" value is used to keep a sort of "checkvalue" to ensure that the - * error code that points at this message is in fact for this particular - * message. This is used to handle the unlikely but possible situation where - * so many error messages are generated that the ring is overlapped. - * - * The message arrays are preallocated to ensure that there is space for these - * messages when an error occurs. One variation would be to allow these - * to be dynamically allocated, but it is probably better to either preallocate - * these or turn off all error message generation (which will eliminate these - * arrays). - * - * One possible alternative is to use the message ring *only* for instance - * messages and use the predefined messages in-place for the generic - * messages. The approach used here provides uniform handling of all - * error messages. - * - * Note on user error codes - * - * The "user error codes" is used to handle an ambiguity in the MPI-1 - * standard about the return value from the attribute callbacks. The - * standard does not specify what values, other than 'MPI_SUCCESS', are - * valid. Because the Intel MPI-1 test suite expected 'MPI_Comm_dup' to - * return the same non-zero value returned by the attribute callback routine, - * this is the behavior that many (if not all) MPI implementations provide. - * As a result, the return from those routines is - * - */ - -/* - * This is the real routine for generating an error code. It takes - * a va_list so that it can be called by any routine that accepts a - * variable number of arguments. +/* + * Support for multiple messages, including the error message ring. + * In principle, the error message ring could use used to provide + * support for multiple error classes or codes, without providing + * instance-specific support. However, for now, we combine the two + * capabilities. */ -int MPIR_Err_create_code_valist( int lastcode, int fatal, const char fcname[], - int line, int error_class, - const char generic_msg[], - const char specific_msg[], va_list Argp ) -{ - int err_code; - int generic_idx; - int use_user_error_code = 0; - int user_error_code = -1; - char user_ring_msg[MPIR_MAX_ERROR_LINE+1]; - - /* Create the code from the class and the message ring index */ - - /* Check that lastcode is valid */ - if (lastcode != MPI_SUCCESS) { - int reason; - reason = checkErrcodeIsValid(lastcode); - if (reason) { - /* --BEGIN ERROR HANDLING-- */ - MPIU_Error_printf( "Internal Error: invalid error code %x (%s) in %s:%d\n", - lastcode, ErrcodeInvalidReasonStr( reason ), - fcname, line ); - lastcode = MPI_SUCCESS; - /* --END ERROR HANDLING-- */ - } - } - - /* FIXME: ERR_OTHER is overloaded; this may mean "OTHER" or it may - mean "No additional error, just routine stack info" */ - if (error_class == MPI_ERR_OTHER) - { - if (MPIR_ERR_GET_CLASS(lastcode) > MPI_SUCCESS && - MPIR_ERR_GET_CLASS(lastcode) <= MPICH_ERR_LAST_CLASS) - { - /* If the last class is more specific (and is valid), then pass it - through */ - error_class = MPIR_ERR_GET_CLASS(lastcode); - } - else - { - error_class = MPI_ERR_OTHER; - } - } - - /* Handle special case of MPI_ERR_IN_STATUS. According to the standard, - the code must be equal to the class. See section 3.7.5. - Information on the particular error is in the MPI_ERROR field - of the status. */ - if (error_class == MPI_ERR_IN_STATUS) - { - return MPI_ERR_IN_STATUS; - } - - err_code = error_class; - - /* Handle the generic message. This selects a subclass, based on a text - string */ - generic_idx = FindGenericMsgIndex(generic_msg); - if (generic_idx >= 0) { - if (strcmp( generic_err_msgs[generic_idx].short_name, "**user" ) == 0) { - use_user_error_code = 1; - /* This is a special case. The format is - "**user", "**userxxx %d", intval - (generic, specific, parameter). In this - case we must ... save the user value because - we store it explicitly in the ring. - We do this here because we cannot both access the - user error code and pass the argp to vsnprintf_mpi . */ - if (specific_msg) { - const char *specific_fmt; - int specific_idx; - user_error_code = va_arg(Argp,int); - specific_idx = FindSpecificMsgIndex(specific_msg); - if (specific_idx >= 0) { - specific_fmt = specific_err_msgs[specific_idx].long_name; - } - else { - specific_fmt = specific_msg; - } - MPIU_Snprintf( user_ring_msg, sizeof(user_ring_msg), - specific_fmt, user_error_code ); - } - else { - user_ring_msg[0] = 0; - } - } - err_code |= (generic_idx + 1) << ERROR_GENERIC_SHIFT; - } - else { - /* TODO: lookup index for class error message */ - err_code &= ~ERROR_GENERIC_MASK; - -# ifdef MPICH_DBG_OUTPUT - { - if (generic_msg[0] == '*' && generic_msg[1] == '*') - { - /* FIXME : Internal error. Generate some debugging - information; Fix for the general release */ - fprintf( stderr, "Could not find %s in list of messages\n", generic_msg ); - } - } -# endif /* DBG_OUTPUT */ - } - - /* Handle the instance-specific part of the error message */ - { - int specific_idx; - const char * specific_fmt = 0; - int ring_idx, ring_seq=0; - char * ring_msg; - - error_ring_mutex_lock(); - { - /* Get the next entry in the ring; keep track of what part of the - ring is in use (max_error_ring_loc) */ - ring_idx = error_ring_loc++; - if (error_ring_loc >= MAX_ERROR_RING) - error_ring_loc %= MAX_ERROR_RING; - if (error_ring_loc > max_error_ring_loc) - max_error_ring_loc = error_ring_loc; - - ring_msg = ErrorRing[ring_idx].msg; - - if (specific_msg != NULL) - { - specific_idx = FindSpecificMsgIndex(specific_msg); - if (specific_idx >= 0) - { - specific_fmt = specific_err_msgs[specific_idx].long_name; - } - else - { - specific_fmt = specific_msg; - } - /* See the code above for handling user errors */ - if (!use_user_error_code) { - vsnprintf_mpi( ring_msg, MPIR_MAX_ERROR_LINE, - specific_fmt, Argp ); - } - else { - MPIU_Strncpy( ring_msg, user_ring_msg, MPIR_MAX_ERROR_LINE ); - } - } - else if (generic_idx >= 0) - { - MPIU_Strncpy( ring_msg,generic_err_msgs[generic_idx].long_name, - MPIR_MAX_ERROR_LINE ); - } - else - { - MPIU_Strncpy( ring_msg, generic_msg, MPIR_MAX_ERROR_LINE ); - } - - ring_msg[MPIR_MAX_ERROR_LINE] = '\0'; - - /* Get the ring sequence number and set the ring id */ - ErrcodeCreateID( error_class, generic_idx, ring_msg, - &ErrorRing[ring_idx].id, &ring_seq ); - /* Set the previous code. */ - ErrorRing[ring_idx].prev_error = lastcode; - - /* */ - if (use_user_error_code) - { - ErrorRing[ring_idx].use_user_error_code = 1; - ErrorRing[ring_idx].user_error_code = user_error_code; - } - else if (lastcode != MPI_SUCCESS) - { - int last_ring_idx; - int last_ring_id; - int last_generic_idx; - if (convertErrcodeToIndexes( lastcode, &last_ring_idx, - &last_ring_id, - &last_generic_idx ) != 0) { - /* --BEGIN ERROR HANDLING-- */ - MPIU_Error_printf( - "Invalid error code (%d) (error ring index %d invalid)\n", - lastcode, last_ring_idx ); - /* --END ERROR HANDLING-- */ - } - else { - if (last_generic_idx >= 0 && - ErrorRing[last_ring_idx].id == last_ring_id) { - if (ErrorRing[last_ring_idx].use_user_error_code) { - ErrorRing[ring_idx].use_user_error_code = 1; - ErrorRing[ring_idx].user_error_code = - ErrorRing[last_ring_idx].user_error_code; - } - } - } - } - - if (fcname != NULL) - { - MPIU_Snprintf(ErrorRing[ring_idx].location, MAX_LOCATION_LEN, "%s(%d)", fcname, line); - ErrorRing[ring_idx].location[MAX_LOCATION_LEN] = '\0'; - } - else - { - ErrorRing[ring_idx].location[0] = '\0'; - } - { - MPIU_DBG_MSG_FMT(ERRHAND,VERBOSE,(MPIU_DBG_FDEST, "New ErrorRing[%d]", ring_idx)); - MPIU_DBG_MSG_FMT(ERRHAND,VERBOSE,(MPIU_DBG_FDEST, " id = %#010x", ErrorRing[ring_idx].id)); - MPIU_DBG_MSG_FMT(ERRHAND,VERBOSE,(MPIU_DBG_FDEST, " prev_error = %#010x", ErrorRing[ring_idx].prev_error)); - MPIU_DBG_MSG_FMT(ERRHAND,VERBOSE,(MPIU_DBG_FDEST, " user=%d", ErrorRing[ring_idx].use_user_error_code)); - } - } - error_ring_mutex_unlock(); - - err_code |= ring_idx << ERROR_SPECIFIC_INDEX_SHIFT; - err_code |= ring_seq << ERROR_SPECIFIC_SEQ_SHIFT; - - } - - if (fatal || MPIR_Err_is_fatal(lastcode)) - { - err_code |= ERROR_FATAL_MASK; - } - - return err_code; -} - -/* Append an error code, error2, to the end of a list of messages in the error - ring whose head endcoded in error1_code. An error code pointing at the - combination is returned. If the list of messages does not terminate cleanly - (i.e. ring wrap has occurred), then the append is not performed. and error1 - is returned (although it may include the class of error2 if the class of - error1 was MPI_ERR_OTHER). */ -int MPIR_Err_combine_codes(int error1, int error2) -{ - int error1_code = error1; - int error2_code = error2; - int error2_class; - - /* If either error code is success, return the other */ - if (error1_code == MPI_SUCCESS) return error2_code; - if (error2_code == MPI_SUCCESS) return error1_code; - - /* If an error code is dynamic, return that. If both are, we choose - error1. */ - if (error1_code & ERROR_DYN_MASK) return error1_code; - if (error2_code & ERROR_DYN_MASK) return error2_code; - - error2_class = MPIR_ERR_GET_CLASS(error2_code); - if (MPIR_ERR_GET_CLASS(error2_class) < MPI_SUCCESS || - MPIR_ERR_GET_CLASS(error2_class) > MPICH_ERR_LAST_CLASS) - { - error2_class = MPI_ERR_OTHER; - } - - CombineSpecificCodes( error1_code, error2_code, error2_class ); - - if (MPIR_ERR_GET_CLASS(error1_code) == MPI_ERR_OTHER) - { - error1_code = (error1_code & ~(ERROR_CLASS_MASK)) | error2_class; - } - - return error1_code; -} - - -/* ------------------------------------------------------------------------- */ -/* Manage the error reporting stack */ -/* ------------------------------------------------------------------------- */ static void MPIR_Err_stack_init( void ) { @@ -1855,14 +1792,76 @@ static void MPIR_Err_stack_init( void ) } } -/**/ -/* Given an error code, print the stack of messages corresponding to this - error code. */ -void MPIR_Err_print_stack(FILE * fp, int errcode) +/* Create the ring id from information about the message */ +static void ErrcodeCreateID( int error_class, int generic_idx, + const char *msg, int *id, int *seq ) +{ + int i; + int ring_seq = 0, ring_id; + + /* Create a simple hash function of the message to serve as the + sequence number */ + ring_seq = 0; + for (i=0; msg[i]; i++) + ring_seq += (unsigned int) msg[i]; + + ring_seq %= ERROR_SPECIFIC_SEQ_SIZE; + + ring_id = (error_class & ERROR_CLASS_MASK) | + ((generic_idx + 1) << ERROR_GENERIC_SHIFT) | + (ring_seq << ERROR_SPECIFIC_SEQ_SHIFT); + + *id = ring_id; + *seq = ring_seq; +} + +/* Convert an error code into ring_idx, ring_id, and generic_idx. + Return non-zero if there is a problem with the decode values + (e.g., out of range for the ring index) */ +static int convertErrcodeToIndexes( int errcode, int *ring_idx, int *ring_id, + int *generic_idx ) +{ + *ring_idx = (errcode & ERROR_SPECIFIC_INDEX_MASK) >> + ERROR_SPECIFIC_INDEX_SHIFT; + *ring_id = errcode & (ERROR_CLASS_MASK | + ERROR_GENERIC_MASK | ERROR_SPECIFIC_SEQ_MASK); + *generic_idx = ((errcode & ERROR_GENERIC_MASK) >> ERROR_GENERIC_SHIFT) - 1; + + /* Test on both the max_error_ring_loc and MAX_ERROR_RING to guard + against memory overwrites */ + if (*ring_idx < 0 || *ring_idx >= MAX_ERROR_RING || + *ring_idx > max_error_ring_loc) return 1; + + return 0; +} +static int checkErrcodeIsValid( int errcode ) +{ + int ring_id, generic_idx, ring_idx; + + /* If the errcode is a class, then it is valid */ + if (errcode <= MPIR_MAX_ERROR_CLASS_INDEX && errcode >= 0) return 0; + + convertErrcodeToIndexes( errcode, &ring_idx, &ring_id, &generic_idx ); + MPIU_DBG_MSG_FMT(ERRHAND, VERBOSE, (MPIU_DBG_FDEST, "code=%#010x ring_idx=%d ring_id=%#010x generic_idx=%d", + errcode, ring_idx, ring_id, generic_idx)); + + if (ring_idx < 0 || ring_idx >= MAX_ERROR_RING || + ring_idx > max_error_ring_loc) return 1; + if (ErrorRing[ring_idx].id != ring_id) return 2; + /* It looks like the code uses a generic idx of -1 to indicate no + generic message */ + if (generic_idx < -1 || generic_idx > generic_msgs_len) return 3; + return 0; +} + +/* Check to see if the error code is a user-specified error code + (e.g., from the attribute delete function) and if so, set the error code + to the value provide by the user */ +static int checkForUserErrcode( int errcode ) { error_ring_mutex_lock(); { - while (errcode != MPI_SUCCESS) { + if (errcode != MPI_SUCCESS) { int ring_idx; int ring_id; int generic_idx; @@ -1871,278 +1870,92 @@ void MPIR_Err_print_stack(FILE * fp, int errcode) &generic_idx ) != 0) { /* --BEGIN ERROR HANDLING-- */ MPIU_Error_printf( - "Invalid error code (%d) (error ring index %d invalid)\n", - errcode, ring_idx ); - break; + "Invalid error code (%d) (error ring index %d invalid)\n", + errcode, ring_idx ); /* --END ERROR HANDLING-- */ } - - if (generic_idx < 0) - { - break; - } - - if (ErrorRing[ring_idx].id == ring_id) - { - fprintf(fp, "%s: %s\n", ErrorRing[ring_idx].location, - ErrorRing[ring_idx].msg); - errcode = ErrorRing[ring_idx].prev_error; - } - else - { - break; + else { + /* Can we get a more specific error message */ + if (generic_idx >= 0 && + ErrorRing[ring_idx].id == ring_id && + ErrorRing[ring_idx].use_user_error_code) + { + errcode = ErrorRing[ring_idx].user_error_code; + } } } } error_ring_mutex_unlock(); - - /* FIXME: This is wrong. The only way that you can get here without - errcode being MPI_SUCCESS is if there is an error in the - processing of the error codes. Dropping through into the next - level of code (particularly when that code doesn't check for - valid error codes!) is erroneous */ - if (errcode == MPI_SUCCESS) - { - goto fn_exit; - } + return errcode; +} - { - int generic_idx; - - generic_idx = ((errcode & ERROR_GENERIC_MASK) >> ERROR_GENERIC_SHIFT) - 1; - - if (generic_idx >= 0) - { - fprintf(fp, "(unknown)(): %s\n", generic_err_msgs[generic_idx].long_name); - goto fn_exit; - } - } - - { - int error_class; - error_class = ERROR_GET_CLASS(errcode); - - if (error_class <= MPICH_ERR_LAST_CLASS) - { - fprintf(fp, "(unknown)(): %s\n", get_class_msg(ERROR_GET_CLASS(errcode))); - } - else - { - /* FIXME: Not internationalized */ - fprintf(fp, "Error code contains an invalid class (%d)\n", error_class); - } +/* --BEGIN ERROR HANDLING-- */ +static const char *ErrcodeInvalidReasonStr( int reason ) +{ + const char *str = 0; + switch (reason) { + case 1: + str = "Ring Index out of range"; + break; + case 2: + str = "Ring ids do not match"; + break; + case 3: + str = "Generic message index out of range"; + break; + default: + str = "Unknown reason for invalid errcode"; + break; } - - fn_exit: - return; + return str; } +/* --END ERROR HANDLING-- */ -/* FIXME: Shouldn't str be const char * ? - no, but you don't know that without - some documentation */ -static void MPIR_Err_print_stack_string(int errcode, char *str, int maxlen ) +static void CombineSpecificCodes( int error1_code, int error2_code, + int error2_class ) { - char *str_orig = str; - int len; - + int error_code; + + error_code = error1_code; + error_ring_mutex_lock(); { - /* Find the longest fcname in the stack */ - int max_location_len = 0; - int tmp_errcode = errcode; - while (tmp_errcode != MPI_SUCCESS) { - int ring_idx; - int ring_id; - int generic_idx; - - if (convertErrcodeToIndexes( tmp_errcode, &ring_idx, &ring_id, - &generic_idx ) != 0) { - /* --BEGIN ERROR HANDLING-- */ - MPIU_Error_printf( - "Invalid error code (%d) (error ring index %d invalid)\n", - errcode, ring_idx ); - break; - /* --END ERROR HANDLING-- */ - } - - if (generic_idx < 0) { - break; - } - - if (ErrorRing[ring_idx].id == ring_id) { - len = (int)strlen(ErrorRing[ring_idx].location); - max_location_len = MPIR_MAX(max_location_len, len); - tmp_errcode = ErrorRing[ring_idx].prev_error; - } - else - { - break; - } - } - max_location_len += 2; /* add space for the ": " */ - /* print the error stack */ - while (errcode != MPI_SUCCESS) { - int ring_idx; - int ring_id; - int generic_idx; - int i; - char *cur_pos; - - if (convertErrcodeToIndexes( errcode, &ring_idx, &ring_id, - &generic_idx ) != 0) { - /* --BEGIN ERROR HANDLING-- */ - MPIU_Error_printf( - "Invalid error code (%d) (error ring index %d invalid)\n", - errcode, ring_idx ); - /* --END ERROR HANDLING-- */ - } - - if (generic_idx < 0) + for (;;) { - break; - } - - if (ErrorRing[ring_idx].id == ring_id) { - int nchrs; - MPIU_Snprintf(str, maxlen, "%s", ErrorRing[ring_idx].location); - len = (int)strlen(str); - maxlen -= len; - str += len; - nchrs = max_location_len - - (int)strlen(ErrorRing[ring_idx].location) - 2; - while (nchrs > 0 && maxlen > 0) { - *str++ = '.'; - nchrs--; - maxlen--; - } - if (maxlen > 0) { - *str++ = ':'; - maxlen--; - } - if (maxlen > 0) { - *str++ = ' '; - maxlen--; - } + int error_class; + int ring_idx; + int ring_id; + int generic_idx; - if (MPIR_PARAM_CHOP_ERROR_STACK > 0) - { - cur_pos = ErrorRing[ring_idx].msg; - len = (int)strlen(cur_pos); - if (len == 0 && maxlen > 0) { - *str++ = '\n'; - maxlen--; + if (convertErrcodeToIndexes(error_code, &ring_idx, &ring_id, + &generic_idx) != 0 || generic_idx < 0 || + ErrorRing[ring_idx].id != ring_id) + { + break; } - while (len) + + error_code = ErrorRing[ring_idx].prev_error; + + if (error_code == MPI_SUCCESS) { - if (len >= MPIR_PARAM_CHOP_ERROR_STACK - max_location_len) - { - if (len > maxlen) - break; - /* FIXME: Don't use Snprint to append a string ! */ - MPIU_Snprintf(str, MPIR_PARAM_CHOP_ERROR_STACK - 1 - max_location_len, "%s", cur_pos); - str[MPIR_PARAM_CHOP_ERROR_STACK - 1 - max_location_len] = '\n'; - cur_pos += MPIR_PARAM_CHOP_ERROR_STACK - 1 - max_location_len; - str += MPIR_PARAM_CHOP_ERROR_STACK - max_location_len; - maxlen -= MPIR_PARAM_CHOP_ERROR_STACK - max_location_len; - if (maxlen < max_location_len) - break; - for (i=0; i<max_location_len; i++) - { - MPIU_Snprintf(str, maxlen, " "); - maxlen--; - str++; - } - len = (int)strlen(cur_pos); - } - else - { - MPIU_Snprintf(str, maxlen, "%s\n", cur_pos); - len = (int)strlen(str); - maxlen -= len; - str += len; - len = 0; - } + ErrorRing[ring_idx].prev_error = error2_code; + break; + } + + error_class = MPIR_ERR_GET_CLASS(error_code); + + if (error_class == MPI_ERR_OTHER) + { + ErrorRing[ring_idx].prev_error &= ~(ERROR_CLASS_MASK); + ErrorRing[ring_idx].prev_error |= error2_class; } - } - else - { - MPIU_Snprintf(str, maxlen, "%s\n", ErrorRing[ring_idx].msg); - len = (int)strlen(str); - maxlen -= len; - str += len; - } - errcode = ErrorRing[ring_idx].prev_error; - } - else - { - break; } - } } error_ring_mutex_unlock(); - - if (errcode == MPI_SUCCESS) - { - goto fn_exit; - } - - /* FIXME: The following code is broken as described above (if the errcode - is not valid, then this code is just going to cause more problems) */ - { - int generic_idx; - - generic_idx = ((errcode & ERROR_GENERIC_MASK) >> ERROR_GENERIC_SHIFT) - 1; - - if (generic_idx >= 0) - { - const char *p; - /* FIXME: (Here and elsewhere) Make sure any string is - non-null before you use it */ - p = generic_err_msgs[generic_idx].long_name; - if (!p) { p = "<NULL>"; } - MPIU_Snprintf(str, maxlen, "(unknown)(): %s\n", p ); - len = (int)strlen(str); - maxlen -= len; - str += len; - goto fn_exit; - } - } - - { - int error_class; - - error_class = ERROR_GET_CLASS(errcode); - - if (error_class <= MPICH_ERR_LAST_CLASS) - { - MPIU_Snprintf(str, maxlen, "(unknown)(): %s\n", - get_class_msg(ERROR_GET_CLASS(errcode))); - len = (int)strlen(str); - maxlen -= len; - str += len; - } - else - { - /* FIXME: Not internationalized */ - MPIU_Snprintf(str, maxlen, - "Error code contains an invalid class (%d)\n", - error_class); - len = (int)strlen(str); - maxlen -= len; - str += len; - } - } - - fn_exit: - if (str_orig != str) - { - str--; - *str = '\0'; - } - return; } -static int ErrGetInstanceString( int errorcode, char *msg, int num_remaining ) +static int ErrGetInstanceString( int errorcode, char msg[], int num_remaining ) { int len; @@ -2200,4 +2013,50 @@ static int ErrGetInstanceString( int errorcode, char *msg, int num_remaining ) return 0; } -#endif /* MSG_LEVEL = MSG_ALL */ + +#else +#error MPICH_ERROR_MSG_LEVEL undefined or has invalid value +#endif + +/* Common routines that are used by two or more error-message levels. + Very simple routines are defined inline */ +#ifdef NEEDS_FIND_GENERIC_MSG_INDEX +/* + * Given a message string abbreviation (e.g., one that starts "**"), return + * the corresponding index. For the generic (non + * parameterized messages), use idx = FindGenericMsgIndex( "**msg" ); + * Returns -1 on failure to find the matching message + * + * The values are in increasing, sorted order, so once we find a + * comparison where the current generic_err_msg is greater than the + * message we are attempting to match, we have missed the match and + * there is an internal error (all short messages should exist in defmsg.h) + */ +/* Question: Could be a service routine for message level >= generic */ +static int FindGenericMsgIndex( const char msg[] ) +{ + int i, c; + for (i=0; i<generic_msgs_len; i++) { + /* Check the sentinals to insure that the values are ok first */ + if (generic_err_msgs[i].sentinal1 != 0xacebad03 || + generic_err_msgs[i].sentinal2 != 0xcb0bfa11) { + /* Something bad has happened! Don't risk trying the + short_name pointer; it may have been corrupted */ + break; + } + c = strcmp( generic_err_msgs[i].short_name, msg ); + if (c == 0) return i; + if (c > 0) { + /* In case the generic messages are not sorted exactly the + way that strcmp compares, we check for the case that + the short msg matches the current generic message. If + that is the case, we do *not* fail */ + if (strncmp(generic_err_msgs[i].short_name, msg, strlen(msg)) != 0) + return -1; + } + } + /* --BEGIN ERROR HANDLING-- */ + return -1; + /* --END ERROR HANDLING-- */ +} +#endif ----------------------------------------------------------------------- Summary of changes: src/mpi/errhan/errutil.c | 1842 +++++++++++++++++++++------------------------- 1 files changed, 853 insertions(+), 989 deletions(-) hooks/post-receive -- MPICH primary repository

1 0