Merge pull request #357 from xylar/fix-mache-deploy

A few fixes to `mache.deploy`

Author: xylar

docs/design/mache_deploy.md

@@ -159,7 +159,12 @@ After deployment, users need a simple way to load the environment.
 
 `mache.deploy` generates load scripts, for example:
 
-`load_<software>.sh`
+- `load_<software>.sh` (no toolchain-specific Spack environment)
+- `load_<software>_<machine>_<compiler>_<mpi>.sh` (toolchain-specific)
+
+Including `<machine>` avoids filename collisions on shared filesystems where
+multiple machines (or machine partitions exposed as distinct machine names)
+share compiler and MPI naming.
 
 These scripts encapsulate:
 

docs/developers_guide/deploy.md

@@ -111,7 +111,9 @@ That file is target-owned and is not generated from a package template.
 
 `load.sh.j2`
 : Package-owned template used by `run.py` to create final
-  `load_<software>*.sh` scripts in the target repo.
+  `load_<software>*.sh` scripts in the target repo
+  (toolchain-specific form:
+  `load_<software>_<machine>_<compiler>_<mpi>.sh`).
 
 `spack_install.bash.j2`
 : Package-owned template used by `spack.py` to create a temporary Spack build

docs/users_guide/deploy.md

@@ -343,6 +343,11 @@ created.
 10. Optionally installs the target software in editable mode.
 11. Writes one or more `load_<software>*.sh` scripts.
 
+When a toolchain pair is selected, script names include machine, compiler,
+and MPI tags, for example:
+
+`load_<software>_<machine>_<compiler>_<mpi>.sh`
+
 Those load scripts are the main artifact a downstream user consumes after the
 deployment completes.
 

mache/deploy/run.py

@@ -772,14 +772,20 @@ def _write_load_script(
     )
     pixi_toml = os.path.join(prefix_abs, 'pixi.toml')
 
-    # Keep the script name stable and discoverable.
-    safe_software = software.strip() or 'software'
     if toolchain_compiler and toolchain_mpi:
+        if machine is None:
+            raise ValueError(
+                'Cannot include toolchain in load script name without machine.'
+                ' Set a machine in deploy/config.yaml.j2 or pass --machine.'
+            )
+        machine_tag = _sanitize_script_tag(machine)
         compiler_tag = _sanitize_script_tag(toolchain_compiler)
         mpi_tag = _sanitize_script_tag(toolchain_mpi)
-        script_path = f'load_{safe_software}_{compiler_tag}_{mpi_tag}.sh'
+        script_path = (
+            f'load_{software}_{machine_tag}_{compiler_tag}_{mpi_tag}.sh'
+        )
     else:
-        script_path = f'load_{safe_software}.sh'
+        script_path = f'load_{software}.sh'
 
     template_text = (
         resources.files(__package__)
@@ -788,7 +794,7 @@ def _write_load_script(
     )
     tmpl = Template(template_text, keep_trailing_newline=True)
 
-    software_upper = safe_software.upper().replace('-', '_')
+    software_upper = software.upper().replace('-', '_')
     source_path = os.path.abspath(os.getcwd())
     target_load_snippet = os.path.join(source_path, 'deploy', 'load.sh')
 

mache/deploy/templates/load.sh.j2

@@ -50,10 +50,11 @@ if [[ -n "${MACHE_DEPLOY_RUNTIME_VERSION_CMD:-}" ]]; then
     return 1
   fi
 
+  # Use a non-login shell so pixi-provided PATH/env are not reset by site profiles.
   probe_output="$(
     env -u PIXI_PROJECT_MANIFEST -u PIXI_PROJECT_ROOT \
       "${PIXI}" run -m "${PIXI_TOML}" -- \
-      bash -lc "${MACHE_DEPLOY_RUNTIME_VERSION_CMD}" 2>&1
+      bash -c "${MACHE_DEPLOY_RUNTIME_VERSION_CMD}" 2>&1
   )"
   status=$?
 
@@ -103,7 +104,6 @@ echo "   spack env activated."
 if [[ -n "${MACHE_DEPLOY_TARGET_LOAD_SNIPPET:-}" \
       && -f "${MACHE_DEPLOY_TARGET_LOAD_SNIPPET}" ]]; then
   echo "loading ${MACHE_DEPLOY_TARGET_SOFTWARE} environment variables..."
-  echo "   ${MACHE_DEPLOY_TARGET_LOAD_SNIPPET}"
   # shellcheck source=/dev/null
   source "${MACHE_DEPLOY_TARGET_LOAD_SNIPPET}"
   echo "   ${MACHE_DEPLOY_TARGET_SOFTWARE} environment variables loaded."

tests/test_deploy_run.py

@@ -0,0 +1,48 @@
+from pathlib import Path
+
+from mache.deploy import run as deploy_run
+
+
+def test_write_load_script_includes_machine_for_toolchain(
+    tmp_path, monkeypatch
+):
+    monkeypatch.chdir(tmp_path)
+
+    script_path = deploy_run._write_load_script(
+        prefix=str(tmp_path / 'prefix'),
+        pixi_exe='/usr/bin/pixi',
+        software='polaris',
+        software_version='1.0.0',
+        runtime_version_cmd=None,
+        machine='pm-gpu',
+        toolchain_compiler='oneapi/2024.2',
+        toolchain_mpi='mpich@4.2',
+        spack_library_view=None,
+        spack_activation='',
+    )
+
+    assert script_path == 'load_polaris_pm-gpu_oneapi_2024.2_mpich_4.2.sh'
+    assert Path(script_path).is_file()
+
+
+def test_write_load_script_without_toolchain_keeps_default_name(
+    tmp_path,
+    monkeypatch,
+):
+    monkeypatch.chdir(tmp_path)
+
+    script_path = deploy_run._write_load_script(
+        prefix=str(tmp_path / 'prefix'),
+        pixi_exe='/usr/bin/pixi',
+        software='polaris',
+        software_version='1.0.0',
+        runtime_version_cmd=None,
+        machine='chrysalis',
+        toolchain_compiler=None,
+        toolchain_mpi=None,
+        spack_library_view=None,
+        spack_activation='',
+    )
+
+    assert script_path == 'load_polaris.sh'
+    assert Path(script_path).is_file()