Move conditional expand logic to base class

Also add a gpu-per-node entry

lib/ramble/ramble/workflow_manager.py

@@ -14,6 +14,7 @@
 from ramble.util.naming import NS_SEPARATOR
 import ramble.util.class_attributes
 import ramble.util.directives
+from ramble.expander import ExpanderError
 
 
 class WorkflowManagerBase(metaclass=WorkflowManagerMeta):
@@ -48,6 +49,27 @@ def get_status(self, workspace):
         """Return status of a given job"""
         return None
 
+    def conditional_expand(self, templates):
+        """Return a (potentially empty) list of expanded strings
+
+        Args:
+            templates: A list of templates to expand.
+                If the template cannot be fully expanded, it's skipped.
+        Returns:
+            A list of expanded strings
+        """
+        expander = self.app_inst.expander
+        expanded = []
+        for tpl in templates:
+            try:
+                rendered = expander.expand_var(tpl, allow_passthrough=False)
+                if rendered:
+                    expanded.append(rendered)
+            except ExpanderError:
+                # Skip a particular entry if any of the vars are not defined
+                continue
+        return expanded
+
     def copy(self):
         """Deep copy a workflow manager instance"""
         new_copy = type(self)(self._file_path)

var/ramble/repos/builtin/workflow_managers/slurm/workflow_manager.py

@@ -37,12 +37,6 @@ def __init__(self, file_path):
 
         self.runner = SlurmRunner()
 
-    workflow_manager_variable(
-        name="partition",
-        default="",
-        description="Name of the slurm partition for job submission",
-    )
-
     workflow_manager_variable(
         name="job_name",
         default="{application_name}_{workload_name}_{experiment_name}",
@@ -96,31 +90,21 @@ def _slurm_execute_script(self):
         expander = self.app_inst.expander
         # Adding pre-defined sets of headers
         pragmas = [
-            ("#SBATCH -N {}", "n_nodes"),
-            ("#SBATCH -p {}", "partition"),
-            ("#SBATCH --ntasks-per-node {}", "processes_per_node"),
-            ("#SBATCH -J {}", "job_name"),
+            ("#SBATCH -N {n_nodes}"),
+            ("#SBATCH -p {partition}"),
+            ("#SBATCH --ntasks-per-node {processes_per_node}"),
+            ("#SBATCH -J {job_name}"),
+            ("#SBATCH --gpus-per-node {gpus_per_node}"),
         ]
-        for tpl, var in pragmas:
-            try:
-                val = expander.expand_var_name(var, allow_passthrough=False)
-            except ExpanderError:
-                # Skip a particular header if any of the vars are not defined
-                continue
-            if val:
-                headers.append(tpl.format(val))
-        # Adding extra arbitrary headers
         try:
             extra_sbatch_headers_raw = expander.expand_var_name(
                 "extra_sbatch_headers", allow_passthrough=False
             )
             extra_sbatch_headers = extra_sbatch_headers_raw.strip().split("\n")
-            extra_headers = [
-                expander.expand_var(h) for h in extra_sbatch_headers
-            ]
-            headers = headers + extra_headers
+            pragmas = pragmas + extra_sbatch_headers
         except ExpanderError:
             pass
+        headers = headers + self.conditional_expand(pragmas)
         header_str = "\n".join(headers)
         content = rf"""
 {header_str}