mpiCheckPhaseHook: add parameters to bypass errors in sandbox (#350112)

pkgs/applications/science/chemistry/nwchem/default.nix

@@ -195,7 +195,7 @@ stdenv.mkDerivation rec {
   doCheck = false;
 
   doInstallCheck = true;
-  nativeCheckInputs = [ mpiCheckPhaseHook ];
+  nativeInstallCheckInputs = [ mpiCheckPhaseHook ];
   installCheckPhase = ''
     runHook preInstallCheck
 
@@ -211,7 +211,7 @@ stdenv.mkDerivation rec {
   meta = with lib; {
     description = "Open Source High-Performance Computational Chemistry";
     mainProgram = "nwchem";
-    platforms = [ "x86_64-linux" ];
+    platforms = [ "x86_64-linux" "aarch64-linux" ];
     maintainers = with maintainers; [ sheepforce markuskowa ];
     homepage = "https://nwchemgit.github.io";
     license = licenses.ecl20;

pkgs/build-support/setup-hooks/mpi-check-hook/default.nix

@@ -2,4 +2,8 @@
 
 makeSetupHook {
   name = "mpi-checkPhase-hook";
+
+  substitutions = {
+    topology = ./topology.xml;
+  };
 } ./mpi-check-hook.sh

pkgs/build-support/setup-hooks/mpi-check-hook/mpi-check-hook.sh

@@ -44,6 +44,17 @@ setupMpiCheck() {
       # Disable CPU pinning
       export OMPI_MCA_hwloc_base_binding_policy=none
       export PRTE_MCA_hwloc_default_binding_policy=none
+
+      # OpenMPI get confused by the sandbox environment and spew errors like this (both to stdout and stderr):
+      #     [hwloc/linux] failed to find sysfs cpu topology directory, aborting linux discovery.
+      #     [1729458724.473282] [localhost:78   :0]       tcp_iface.c:893  UCX  ERROR scandir(/sys/class/net) failed: No such file or directory
+      # These messages contaminate test output, which makes the difftest to fail.
+      # The solution is to use a preset cpu topology file and disable ucx model.
+
+      # Disable sysfs cpu topology directory discovery.
+      export PRTE_MCA_hwloc_use_topo_file="@topology@"
+      # Use the network model ob1 instead of ucx.
+      export OMPI_MCA_pml=ob1
       ;;
     MPICH)
       # Fix to make mpich run in a sandbox

pkgs/build-support/setup-hooks/mpi-check-hook/topology.xml

@@ -0,0 +1,10 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE topology SYSTEM "hwloc2.dtd">
+<topology version="2.0">
+  <object type="Machine" os_index="0" cpuset="0x00000001" complete_cpuset="0x00000001" allowed_cpuset="0x00000001" nodeset="0x00000001" complete_nodeset="0x00000001" allowed_nodeset="0x00000001" gp_index="1">
+    <object type="Core" cpuset="0x00000001" complete_cpuset="0x00000001" nodeset="0x00000001" complete_nodeset="0x00000001" gp_index="2">
+      <object type="NUMANode" os_index="0" cpuset="0x00000001" complete_cpuset="0x00000001" nodeset="0x00000001" complete_nodeset="0x00000001" gp_index="4"/>
+      <object type="PU" os_index="0" cpuset="0x00000001" complete_cpuset="0x00000001" nodeset="0x00000001" complete_nodeset="0x00000001" gp_index="3"/>
+    </object>
+  </object>
+</topology>

pkgs/by-name/pe/petsc/package.nix

@@ -9,6 +9,7 @@
   lapack,
   mpiSupport ? true,
   mpi, # generic mpi dependency
+  mpiCheckPhaseHook,
   openssh, # required for openmpi tests
   petsc-withp4est ? false,
   hdf5-support ? false,
@@ -52,12 +53,6 @@ stdenv.mkDerivation rec {
       --replace /usr/bin/install_name_tool ${cctools}/bin/install_name_tool
   '';
 
-  # Both OpenMPI and MPICH get confused by the sandbox environment and spew errors like this (both to stdout and stderr):
-  #     [hwloc/linux] failed to find sysfs cpu topology directory, aborting linux discovery.
-  #     [1684747490.391106] [localhost:14258:0]       tcp_iface.c:837  UCX  ERROR opendir(/sys/class/net) failed: No such file or directory
-  # These messages contaminate test output, which makes the quicktest suite to fail. The patch adds filtering for these messages.
-  patches = [ ./filter_mpi_warnings.patch ];
-
   configureFlags = [
     "--with-blas=1"
     "--with-lapack=1"
@@ -112,6 +107,7 @@ stdenv.mkDerivation rec {
   # the library is installed and available.
   doInstallCheck = true;
   installCheckTarget = "check_install";
+  nativeInstallCheckInputs = [ mpiCheckPhaseHook ];
 
   passthru = {
     inherit mpiSupport;