Evaluate hints and resources at the workflow level, using the correct inputs.

Author: kinow

cwltool/workflow.py

@@ -35,6 +35,7 @@
     aslist,
 )
 from .workflow_job import WorkflowJob
+from cwl_utils import expression
 
 
 def default_make_tool(
@@ -156,6 +157,73 @@ def job(
         output_callbacks: Optional[OutputCallbackType],
         runtimeContext: RuntimeContext,
     ) -> JobsGeneratorType:
+        # TODO: This is not very efficient and could be improved.
+        #
+        # See issue #1330. We needed to evaluate the requirements
+        # at the workflow level, so that it was not re-evaluated at
+        # each step level (since a command-line tool, for instance,
+        # won't have the inputs from args, but instead will have the
+        # empty/default inputs of a workflow step).
+        #
+        # The solution below evaluates the requirements and hints for
+        # the workflow (parent), keeping track of the name of the
+        # requirements and hints. For each step of the workflow and of
+        # the embedded tool (command-line or expression tools) it will
+        # then evaluate the requirements or hints that have the same
+        # name - even though they may be re-evaluated at the step
+        # level (e.g. a workflow defines a requirement resource that
+        # uses inputs.threads_max, and a command-line tool of the same
+        # workflow also defines a requirement with the same name, but
+        # using the command-line tool input values).
+        #
+        # This prevents evaluation at the step level (i.e. the values
+        # were already loaded earlier).
+        def _fix_hints_and_requirements(
+            hints_or_requirements: List[CWLObjectType],
+            requirements_or_hints_to_evaluate: List[str],
+        ) -> None:
+            """Fix hints and requirements of a workflow.
+
+            Internal function to iterate the hints or requirements
+            of steps provided and evaluate the ones that exist in
+            the parent process.
+            """
+            for hint_or_requirement in hints_or_requirements:
+                for key, value in hint_or_requirement.items():
+                    if key in requirements_or_hints_to_evaluate:
+                        hint_or_requirement[key] = expression.do_eval(
+                            ex=value,
+                            jobinput=job_order,
+                            requirements=self.requirements,
+                            outdir=runtimeContext.outdir,
+                            tmpdir=runtimeContext.tmpdir,
+                            resources={},
+                            context=None,
+                            timeout=runtimeContext.eval_timeout,
+                        )
+
+        for attr_key in ["hints", "requirements"]:
+            parent_entries = []
+            for hint_or_requirement in getattr(self, attr_key):
+                for key, value in hint_or_requirement.items():
+                    hint_or_requirement[key] = expression.do_eval(
+                        ex=value,
+                        jobinput=job_order,
+                        requirements=self.requirements,
+                        outdir=runtimeContext.outdir,
+                        tmpdir=runtimeContext.tmpdir,
+                        resources={},
+                        context=None,
+                        timeout=runtimeContext.eval_timeout,
+                    )
+                    parent_entries.append(key)
+
+            for step in self.steps:
+                _fix_hints_and_requirements(getattr(step, attr_key), parent_entries)
+                _fix_hints_and_requirements(
+                    getattr(step.embedded_tool, attr_key), parent_entries
+                )
+
         builder = self._init_job(job_order, runtimeContext)
 
         if runtimeContext.research_obj is not None:

tests/test_reqs_hints.py

@@ -0,0 +1,58 @@
+"""Test for Requirements and Hints in cwltool."""
+import json
+from io import StringIO
+
+from cwltool.main import main
+from .util import get_data
+
+
+def test_workflow_reqs_are_evaluated_earlier_default_args() -> None:
+    """Test that a Workflow process will evaluate the requirements earlier.
+
+    Uses the default input values.
+
+    This means that workflow steps, such as Expression and Command Line Tools
+    can both use resources without re-evaluating expressions. This is useful
+    when you have an expression that, for instance, dynamically decides
+    how many threads/cpus to use.
+
+    Issue: https://github.com/common-workflow-language/cwltool/issues/1330
+    """
+    stream = StringIO()
+
+    assert (
+        main(
+            [get_data("tests/wf/1330.cwl")],
+            stdout=stream,
+        )
+        == 0
+    )
+
+    out = json.loads(stream.getvalue())
+    assert out["out"] == "2\n"
+
+
+def test_workflow_reqs_are_evaluated_earlier_provided_inputs() -> None:
+    """Test that a Workflow process will evaluate the requirements earlier.
+
+    Passes inputs via a job file.
+
+    This means that workflow steps, such as Expression and Command Line Tools
+    can both use resources without re-evaluating expressions. This is useful
+    when you have an expression that, for instance, dynamically decides
+    how many threads/cpus to use.
+
+    Issue: https://github.com/common-workflow-language/cwltool/issues/1330
+    """
+    stream = StringIO()
+
+    assert (
+        main(
+            [get_data("tests/wf/1330.cwl"), get_data("tests/wf/1330.json")],
+            stdout=stream,
+        )
+        == 0
+    )
+
+    out = json.loads(stream.getvalue())
+    assert out["out"] == "1\n"

tests/wf/1330.cwl

@@ -0,0 +1,39 @@
+# Original file: tests/eager-eval-reqs-hints/wf-reqs.cwl
+# From: https://github.com/common-workflow-language/cwl-v1.2/pull/195
+cwlVersion: v1.2
+class: Workflow
+
+requirements:
+  ResourceRequirement:
+    coresMax: $(inputs.threads_max)
+
+inputs:
+  threads_max:
+    type: int
+    default: 2
+
+steps:
+  one:
+    in: []
+    run:
+      class: CommandLineTool
+      inputs:
+        other_input:
+          type: int
+          default: 8
+      baseCommand: echo
+      arguments: [ $(runtime.cores) ]
+      stdout: out.txt
+      outputs:
+        out:
+          type: string
+          outputBinding:
+            glob: out.txt
+            loadContents: true
+            outputEval: $(self[0].contents)
+    out: [out]
+
+outputs:
+  out:
+    type: string
+    outputSource: one/out

tests/wf/1330.json

@@ -0,0 +1,3 @@
+{
+  "threads_max": 1
+}