fix(decision): use the merge base as base_rev for GitHub pushes and pull requests (bug 2072920)

docs/reference/parameters.rst

@@ -28,6 +28,17 @@ base_rev
 ~~~~~~~~
    The previous revision before ``head_rev`` got merged into. This can be a short revision string.
 
+   For a ``github-push`` of a branch and for ``github-pull-request``, the
+   :term:`Decision Task` replaces the reported base with the merge base of
+   ``head_rev`` and the target branch, so ``files_changed`` covers only the
+   new commits. The target is ``base_ref`` for a pull request and the default
+   branch for a push. A push that only adds commits keeps its reported base,
+   and a force push to the default branch uses the merge base of its previous
+   and new tips. When no merge base can be found, the reported base is kept.
+
+   Set the ``GITHUB_TOKEN`` environment variable in the :term:`Decision Task`
+   to raise the rate limit of the GitHub API requests this makes.
+
 head_rev
 ~~~~~~~~
    The revision to check out, this can be a short revision string.

src/taskgraph/decision.py

@@ -153,6 +153,28 @@ def taskgraph_decision(options, parameters=None):
     )
 
 
+def derive_base_rev(repo, parameters):
+    """Return the base to use for a GitHub push or pull request, or ``None``
+    to keep ``parameters["base_rev"]`` as reported."""
+    tasks_for = parameters.get("tasks_for", "")
+    if tasks_for == "github-push":
+        # GitHub's event.before is not the base of a new or force pushed
+        # branch, see GitRepository.derive_base_rev.
+        return repo.derive_base_rev(
+            parameters["head_repository"],
+            parameters["head_rev"],
+            parameters["head_ref"],
+            parameters["base_rev"],
+        )
+    if tasks_for.startswith("github-pull-request"):
+        return repo.derive_pull_request_base_rev(
+            parameters["base_repository"],
+            parameters["head_rev"],
+            parameters.get("base_ref"),
+        )
+    return None
+
+
 def get_decision_parameters(graph_config, options):
     """
     Load parameters from the command-line options for 'taskgraph decision'.
@@ -188,6 +210,10 @@ def get_decision_parameters(graph_config, options):
     except UnicodeDecodeError:
         commit_message = ""
 
+    if base_rev := derive_base_rev(repo, parameters):
+        logger.info(f"Using {base_rev} as base_rev for {parameters['head_ref']}")
+        parameters["base_rev"] = base_rev
+
     # Define default filter list, as most configurations shouldn't need
     # custom filters.
     parameters["files_changed"] = repo.get_changed_files(