Enable address sanitizer for mpi4py runs

.github/workflows/ompi_mpi4py.yaml

@@ -20,18 +20,25 @@ permissions:
 
 jobs:
   test:
-    runs-on: ubuntu-22.04
+    # We need Unbuntu 24.04 (over 22.04) due to a kernel bug,
+    # see https://github.com/google/sanitizers/issues/856.
+    runs-on: ubuntu-24.04
     timeout-minutes: 30
     env:
       MPI4PY_TEST_SPAWN: true
+      # disable ASAN while building
+      ASAN_OPTIONS: verify_asan_link_order=0,detect_odr_violation=0,abort_on_error=0
+      # disable leak detection
+      LSAN_OPTIONS: detect_leaks=0,exitcode=0
+
     steps:
     - name: Configure hostname
       run:  echo 127.0.0.1 `hostname` | sudo tee -a /etc/hosts > /dev/null
       if:   ${{ runner.os == 'Linux' || runner.os == 'macOS' }}
 
     - name: Install depencencies
       run:  sudo apt-get install -y -q
-              libnuma-dev
+              libnuma-dev libasan8
       if:   ${{ runner.os == 'Linux' }}
 
     - name: Checkout Open MPI
@@ -59,7 +66,8 @@ jobs:
               --disable-oshmem
               --disable-silent-rules
               --prefix=/opt/openmpi
-              LDFLAGS=-Wl,-rpath,/opt/openmpi/lib
+              CFLAGS="-O2 -fno-omit-frame-pointer -g -fsanitize=address"
+              LDFLAGS="-Wl,-rpath,/opt/openmpi/lib -fsanitize=address"
       working-directory: mpi-build
 
     - name: Build MPI
@@ -115,6 +123,21 @@ jobs:
       env:
         CFLAGS: "-O0"
 
+    - name: Setting up ASAN environment
+      # LD_PRELOAD is needed to make sure ASAN is the first thing loaded
+      # as it will otherwise complain.
+      # Leak detection is currently disabled because of the size of the report.
+      # The patcher is disabled because ASAN fails if code mmaps data at fixed
+      # memory addresses, see https://github.com/open-mpi/ompi/issues/12819.
+      # ODR violation detection is disabled until #13469 is fixed
+      # Disabling stack use after return detection to reduce slowdown, per
+      # https://github.com/llvm/llvm-project/issues/64190.
+      run: |
+        echo LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libasan.so.8 >> $GITHUB_ENV
+        echo ASAN_OPTIONS=detect_odr_violation=0,abort_on_error=1,detect_stack_use_after_return=0 >> $GITHUB_ENV
+        echo LSAN_OPTIONS=detect_leaks=0,exitcode=0 >> $GITHUB_ENV
+        echo OMPI_MCA_memory=^patcher >> $GITHUB_ENV
+
     - name: Test mpi4py (singleton)
       run:  python test/main.py -v -x TestExcErrhandlerNull
       if:   ${{ true }}
@@ -145,6 +168,18 @@ jobs:
       if:   ${{ true }}
       timeout-minutes: 10
 
+    - name: Show MPI (ASAN)
+      run:  ompi_info
+
+    - name: Show MPICC (ASAN)
+      run:  mpicc -show
+
+    - name: Disabling ASAN environment
+      run: |
+        echo LD_PRELOAD= >> $GITHUB_ENV
+        echo ASAN_OPTIONS=verify_asan_link_order=0,detect_odr_violation=0,abort_on_error=0 >> $GITHUB_ENV
+        echo LSAN_OPTIONS=detect_leaks=0,exitcode=0 >> $GITHUB_ENV
+
     - name: Relocate Open MPI installation
       run:  mv /opt/openmpi /opt/ompi
     - name: Update PATH and set OPAL_PREFIX and LD_LIBRARY_PATH
@@ -157,4 +192,3 @@ jobs:
       run:  python test/main.py -v -x TestExcErrhandlerNull
       if:   ${{ true }}
       timeout-minutes: 10
-

.gitignore

@@ -517,6 +517,7 @@ docs/_static
 docs/_static/css/custom.css
 docs/_templates
 docs/man-openmpi/man3/bindings
+docs/*.inv
 
 # Common Python virtual environment and cache directory names
 venv

config/ompi_setup_prrte.m4

@@ -19,7 +19,7 @@ dnl Copyright (c) 2019-2020 Intel, Inc.  All rights reserved.
 dnl Copyright (c) 2020-2022 Amazon.com, Inc. or its affiliates.  All Rights reserved.
 dnl Copyright (c) 2021      Nanook Consulting.  All rights reserved.
 dnl Copyright (c) 2021-2022 IBM Corporation.  All rights reserved.
-dnl Copyright (c) 2023-2024 Jeffrey M. Squyres.  All rights reserved.
+dnl Copyright (c) 2023-2025 Jeffrey M. Squyres.  All rights reserved.
 dnl Copyright (c) 2025      Advanced Micro Devices, Inc. All rights reserved.
 dnl $COPYRIGHT$
 dnl
@@ -39,7 +39,8 @@ dnl results of the build.
 AC_DEFUN([OMPI_SETUP_PRRTE],[
     AC_REQUIRE([AC_PROG_LN_S])
 
-OPAL_VAR_SCOPE_PUSH([prrte_setup_internal_happy prrte_setup_external_happy target_rst_dir])
+    OPAL_VAR_SCOPE_PUSH([prrte_setup_internal_happy prrte_setup_external_happy target_rst_dir ompi_external_prrte_docs_url])
+    ompi_external_prrte_docs_url="https://docs.prrte.org/en/latest/"
 
     opal_show_subtitle "Configuring PRRTE"
 
@@ -120,6 +121,8 @@ OPAL_VAR_SCOPE_PUSH([prrte_setup_internal_happy prrte_setup_external_happy targe
 
     AC_SUBST(OMPI_PRRTE_RST_CONTENT_DIR)
     AC_SUBST(OMPI_SCHIZO_OMPI_RST_CONTENT_DIR)
+    AC_SUBST(OMPI_PRRTE_DOCS_URL_BASE)
+    AC_SUBST(OMPI_USING_INTERNAL_PRRTE)
     AM_CONDITIONAL(OMPI_HAVE_PRRTE_RST, [test $OMPI_HAVE_PRRTE_RST -eq 1])
 
     AS_IF([test "$OMPI_USING_INTERNAL_PRRTE" = "1"],
@@ -250,8 +253,30 @@ AC_DEFUN([_OMPI_SETUP_PRRTE_INTERNAL], [
                  [OMPI_HAVE_PRRTE_RST=1
                   OMPI_PRRTE_RST_CONTENT_DIR="$OMPI_TOP_SRCDIR/3rd-party/prrte/src/docs/prrte-rst-content"
                   OMPI_SCHIZO_OMPI_RST_CONTENT_DIR="$OMPI_TOP_SRCDIR/3rd-party/prrte/src/mca/schizo/ompi"
+
+                  # If we're building the OMPI Sphinx docs, and also
+                  # building the internal PRRTE, then we're *also*
+                  # building the internal PRRTE docs.
+                  #
+                  # In this case, the OMPI docs/conf.py will do a
+                  # bunch of processing that is a lot easier to do in
+                  # Python than Bourne shell (e.g., use the convenient
+                  # os.path.relpath() to compute the relative path
+                  # that we need, as well as dynamically create a
+                  # Sphinx link inventory file).  Hence, we skip doing
+                  # all that work here and just set a sentinel value
+                  OMPI_PRRTE_DOCS_URL_BASE="../../prrte/html"
                   AC_MSG_RESULT([found])],
-		 [AC_MSG_RESULT([not found])])
+                 [ # If we are not building the Sphinx docs, default
+                   # to using the external PRRTE docs URL.  This is
+                   # actually moot because we won't be building the
+                   # docs, but we might as well be complete in the
+                   # logic / cases.
+                  OMPI_PRRTE_DOCS_URL_BASE=$ompi_external_prrte_docs_url
+                  AC_MSG_RESULT([not found])])
+
+           AC_MSG_CHECKING([for internal PRRTE docs link URL base])
+           AC_MSG_RESULT([$OMPI_PRRTE_DOCS_URL_BASE])
            $1],
 	  [$2])
 
@@ -273,7 +298,7 @@ dnl _OMPI_SETUP_PRRTE_EXTERNAL([action if success], [action if not success])
 dnl
 dnl Try to find an external prrte with sufficient version.
 AC_DEFUN([_OMPI_SETUP_PRRTE_EXTERNAL], [
-    OPAL_VAR_SCOPE_PUSH([ompi_prte_min_version ompi_prte_min_num_version setup_prrte_external_happy opal_prrte_CPPFLAGS_save])
+    OPAL_VAR_SCOPE_PUSH([ompi_prte_min_version ompi_prte_min_num_version setup_prrte_external_happy opal_prrte_CPPFLAGS_save ompi_prrte_docdir])
 
     opal_prrte_CPPFLAGS_save=$CPPFLAGS
 
@@ -321,6 +346,10 @@ AC_DEFUN([_OMPI_SETUP_PRRTE_EXTERNAL], [
           [ # Determine if this external PRRTE has installed the RST
             # directories that we care about
 
+           # In the external case, initially assume we'll use the
+           # web-based docs
+           OMPI_PRRTE_DOCS_URL_BASE=$ompi_external_prrte_docs_url
+
            AC_MSG_CHECKING([for external PRRTE RST files])
            prrte_install_dir=${with_prrte}/share/prte/rst
            AS_IF([test -n "$SPHINX_BUILD"],
@@ -329,13 +358,27 @@ AC_DEFUN([_OMPI_SETUP_PRRTE_EXTERNAL], [
                         [OMPI_HAVE_PRRTE_RST=1
                          OMPI_PRRTE_RST_CONTENT_DIR="$prrte_install_dir/prrte-rst-content"
                          OMPI_SCHIZO_OMPI_RST_CONTENT_DIR="$prrte_install_dir/schizo-ompi-rst-content"
+                         # If the external PRTE docs dir exists where
+                         # a simple heuristic thinks it should be
+                         # (i.e., the default docdir location), use
+                         # it.  This will be an absolute path, which
+                         # is fine (because we're building against an
+                         # external PRRTE).  If we don't find it,
+                         # we'll fall back to the above-set HTTPS
+                         # internet PRRTE docs URL.
+                         ompi_prrte_docdir="$with_prrte/share/doc/prrte/html"
+                         AS_IF([test -d "$ompi_prrte_docdir"],
+                                [OMPI_PRRTE_DOCS_URL_BASE="$ompi_prrte_docdir"])
                          AC_MSG_RESULT([found])
                          ],
                         [ # This version of PRRTE doesn't have installed RST
                           # files.
                           AC_MSG_RESULT([not found])
                         ])
                  ])
+
+           AC_MSG_CHECKING([for external PRRTE docs link URL base])
+           AC_MSG_RESULT([$OMPI_PRRTE_DOCS_URL_BASE])
            $1],
 	  [$2])
 

config/opal_config_pmix.m4

@@ -21,6 +21,7 @@ dnl Copyright (c) 2020      Triad National Security, LLC. All rights
 dnl                         reserved.
 dnl Copyright (c) 2020-2022 Amazon.com, Inc. or its affiliates.  All Rights reserved.
 dnl Copyright (c) 2021      Nanook Consulting.  All rights reserved.
+dnl Copyright (c) 2025      Jeffrey M. Squyres.  All rights reserved.
 dnl $COPYRIGHT$
 dnl
 dnl Additional copyrights may follow
@@ -57,7 +58,8 @@ dnl         other execution tests later in configure (there are sadly
 dnl         some) would fail if the path in LDFLAGS was not added to
 dnl         LD_LIBRARY_PATH.
 AC_DEFUN([OPAL_CONFIG_PMIX], [
-    OPAL_VAR_SCOPE_PUSH([external_pmix_happy internal_pmix_happy internal_pmix_args internal_pmix_wrapper_libs internal_pmix_CPPFLAGS opal_pmix_STATIC_LDFLAGS opal_pmix_LIBS opal_pmix_STATIC_LIBS])
+    OPAL_VAR_SCOPE_PUSH([external_pmix_happy internal_pmix_happy internal_pmix_args internal_pmix_wrapper_libs internal_pmix_CPPFLAGS opal_pmix_STATIC_LDFLAGS opal_pmix_LIBS opal_pmix_STATIC_LIBS opal_external_pmix_docs_url])
+    opal_external_pmix_docs_url="https://docs.openpmix.org/en/latest/"
 
     opal_show_subtitle "Configuring PMIx"
 
@@ -154,6 +156,8 @@ AC_DEFUN([OPAL_CONFIG_PMIX], [
     AC_DEFINE_UNQUOTED([OPAL_USING_INTERNAL_PMIX],
                        [$OPAL_USING_INTERNAL_PMIX],
                        [Whether or not we are using the internal PMIx])
+    AC_SUBST(OPAL_PMIX_DOCS_URL_BASE)
+    AC_SUBST(OPAL_USING_INTERNAL_PMIX)
 
     OPAL_SUMMARY_ADD([Miscellaneous], [pmix], [], [$opal_pmix_mode])
 
@@ -216,8 +220,22 @@ AC_DEFUN([_OPAL_CONFIG_PMIX_EXTERNAL], [
                   dnl it will screw up other tests (like the pthread tests)
                   opal_pmix_BUILD_LIBS="${opal_pmix_LIBS}"
 
+                  # If the external PMIx docs dir exists where
+                  # a simple heuristic thinks it should be
+                  # (i.e., the default docdir location), use
+                  # it.  This will be an absolute path, which
+                  # is fine (because we're building against an
+                  # external PMIx).  If we don't find it,
+                  # we'll fall back to the HTTPS internet PMIx
+                  # docs URL.
+                  opal_pmix_docdir="$with_pmix/share/doc/pmix/html"
+                  AS_IF([test -d "$opal_pmix_docdir"],
+                        [OPAL_PMIX_DOCS_URL_BASE="$opal_pmix_docdir"],
+                        [OPAL_PMIX_DOCS_URL_BASE=$opal_external_pmix_docs_url])
+
                   $1],
-                 [$2])])
+                 [$2])
+     ])
 
     OPAL_VAR_SCOPE_POP
 ])
@@ -238,7 +256,7 @@ AC_DEFUN([_OPAL_CONFIG_PMIX_INTERNAL_POST], [
 
     pmix_internal_happy=1
 
-    dnl Don't pull LDFLAGS, because we don't have a good way to avoid
+    dnl Do not pull LDFLAGS, because we don't have a good way to avoid
     dnl a -L to our install directory, which can cause some weirdness
     dnl if there's an old OMPI install there.  And it makes filtering
     dnl redundant flags easier.
@@ -279,6 +297,31 @@ AC_DEFUN([_OPAL_CONFIG_PMIX_INTERNAL_POST], [
 
     opal_pmix_BUILD_LIBS="$OMPI_TOP_BUILDDIR/3rd-party/openpmix/src/libpmix.la"
 
+    AS_IF([test -n "$SPHINX_BUILD"],
+          [ # If we're building the OMPI Sphinx docs, and also
+            # building the internal PMIx, then we're *also*
+            # building the internal PMIx docs.
+            #
+            # In this case, the OMPI docs/conf.py will do a
+            # bunch of processing that is a lot easier to do in
+            # Python than Bourne shell (e.g., use the convenient
+            # os.path.relpath() to compute the relative path
+            # that we need, as well as dynamically create a
+            # Sphinx link inventory file).  Hence, we skip doing
+            # all that work here and just set a sentinel value
+            OPAL_PMIX_DOCS_URL_BASE="../../pmix/html"
+            AC_MSG_RESULT([found])],
+          [ # If we are not building the Sphinx docs, default
+            # to using the external PMIx docs URL.  This is
+            # actually moot because we won't be building the
+            # docs, but we might as well be complete in the
+            # logic / cases.
+            OPAL_PMIX_DOCS_URL_BASE=$opal_external_pmix_docs_url
+            AC_MSG_RESULT([not found])])
+
+    AC_MSG_CHECKING([for internal PMIx docs link URL base])
+    AC_MSG_RESULT([$OPAL_PMIX_DOCS_URL_BASE])
+
     OPAL_3RDPARTY_SUBDIRS="$OPAL_3RDPARTY_SUBDIRS openpmix"
 ])