[autorevert] properly resolve github workflow names (#7010)

Introduces `WorkflowRestartChecker` that resolves githb workflows name
<-> display name. Uses it in the codebase instead of relying on naming
convention (that apparently is not reliable, see lint.yml).

### Testing

python -m unittest -v
pytorch_auto_revert/tests/test_workflow_resolver.py

```
python -m unittest -v pytorch_auto_revert/tests/test_workflow_resolver.py
test_resolve_pull_workflow (pytorch_auto_revert.tests.test_workflow_resolver.TestWorkflowResolverRealRepo.test_resolve_pull_workflow) ... ok
test_resolve_trunk_workflow (pytorch_auto_revert.tests.test_workflow_resolver.TestWorkflowResolverRealRepo.test_resolve_trunk_workflow) ... ok

----------------------------------------------------------------------
Ran 2 tests in 3.369s

OK
```


```
python -m pytorch_auto_revert
Fetching commits for workflow(s) 'Lint, trunk, pull, inductor' (last 48h)...

Commit data by workflow:
Fetching workflow data for 4 workflows since 2025-08-11T13:40:22.263487...
Found 115 commits with job data for workflow 'Lint'
Found 113 commits with job data for workflow 'inductor'
Found 117 commits with job data for workflow 'pull'
Found 112 commits with job data for workflow 'trunk'
  Lint: 115 commits
  trunk: 112 commits
  pull: 117 commits
  inductor: 113 commits
✓ 2 AUTOREVERT PATTERNS DETECTED

Pattern #1:
Failure rule: 'Lintrunner failure'
Recent commits with failure: a3df3aa8 d3329604
Older commit without failure: 01bcf9a4
✗ NOT REVERTED: d3329604e2f2c5523c68003a1093ba31e4937d42 was not reverted
INFO:root:Successfully dispatched workflow Lint for commit d3329604e2f2c5523c68003a1093ba31e4937d42
  View at: https://github.com/pytorch/pytorch/actions/workflows/lint.yml?query=branch%3Atrunk%2Fd3329604e2f2c5523c68003a1093ba31e4937d42
  ✓ RESTARTED: Lint for d3329604
INFO:root:Successfully dispatched workflow Lint for commit 01bcf9a40dea937637d2cdd530bed2652510943d
  View at: https://github.com/pytorch/pytorch/actions/workflows/lint.yml?query=branch%3Atrunk%2F01bcf9a40dea937637d2cdd530bed2652510943d
  ✓ RESTARTED: Lint for 01bcf9a4
Failed jobs (1):
  - lintrunner-noclang / linux-job

Pattern #2:
Failure rule: 'pytest failure'
Recent commits with failure: 9708fcf9 a288b15e
Older commit without failure: d2d29707
✗ NOT REVERTED: a288b15ea9f87ddd665f249d492e0fb0861f5a69 was not reverted
  ⟳ ALREADY RESTARTED: pull for a288b15e
INFO:root:Successfully dispatched workflow pull for commit d2d2970734e0c3d241002f386bb8ca208e2594bb
  View at: https://github.com/pytorch/pytorch/actions/workflows/pull.yml?query=branch%3Atrunk%2Fd2d2970734e0c3d241002f386bb8ca208e2594bb
  ✓ RESTARTED: pull for d2d29707
Failed jobs (8):
  - linux-jammy-py3.9-gcc11 / test (distributed, 2, 2, linux.2xlarge)
  - linux-jammy-py3.9-clang12 / test (default, 1, 5, linux.4xlarge)
  - linux-jammy-py3.10-clang18-asan / test (default, 2, 6, linux.4xlarge)
  - linux-jammy-py3.10-clang18-asan / test (default, 3, 6, linux.4xlarge)
  - linux-jammy-py3.13-clang12 / test (crossref, 1, 2, linux.2xlarge)
  ... and 3 more

```

---------

Co-authored-by: Jean Schmidt <[email protected]>

Author: izaitsevfb

aws/lambda/pytorch-auto-revert/pytorch_auto_revert/__main__.py

@@ -174,7 +174,7 @@ def main(*args, **kwargs) -> None:
             os.environ.get("WORKFLOWS", "Lint,trunk,pull,inductor").split(","),
             do_restart=True,
             do_revert=True,
-            hours=int(os.environ.get("HOURS", 48)),
+            hours=int(os.environ.get("HOURS", 16)),
             verbose=True,
             dry_run=opts.dry_run,
             ignore_common_errors=True,

aws/lambda/pytorch-auto-revert/pytorch_auto_revert/testers/autorevert.py

@@ -221,7 +221,6 @@ def autorevert_checker(
                     print(f"  ... and {len(pattern['failed_job_names']) - 5} more")
 
                 # Job coverage overlap logging removed (older_job_coverage dropped from pattern)
-
                 if revert_result and verbose:
                     print(f"Revert message: {revert_result['revert_message'][:100]}...")
 
@@ -281,7 +280,6 @@ def autorevert_checker(
             if len_non_ghfirst_reverts > 0
             else 0
         )
-        # recall_non_ghfirst = 1 - ratio_non_ghfirst_reverts
         print(
             "Reverts (excluding ghfirst) that dont match any auto revert pattern detected (%): "
             + f"({len_not_found_non_ghfirst}) ({ratio_non_ghfirst_reverts * 100:.1f}%)"
@@ -302,14 +300,19 @@ def autorevert_checker(
             else 0.0
         )
 
-        print()
-        print("*********************************************************************")
-        print("STATS SUMMARY:")
-        print(f" PRECISION: {stats_precision * 100:.1f}%")
-        print(f" RECALL: {stats_recall * 100:.1f}%")
-        print(f" F1: {stats_f1 * 100:.1f}%")
-        print("*********************************************************************")
-        print()
+        if verbose:
+            print()
+            print(
+                "*********************************************************************"
+            )
+            print("STATS SUMMARY:")
+            print(f" PRECISION: {stats_precision * 100:.1f}%")
+            print(f" RECALL: {stats_recall * 100:.1f}%")
+            print(f" F1: {stats_f1 * 100:.1f}%")
+            print(
+                "*********************************************************************"
+            )
+            print()
 
         workflow_statistics = defaultdict(
             lambda: {"match_pattern": 0, "reverts": 0, "reverts_non_ghfirst": 0}

aws/lambda/pytorch-auto-revert/pytorch_auto_revert/tests/test_workflow_resolver.py

@@ -0,0 +1,61 @@
+import os
+import sys
+import unittest
+
+
+# Ensure package import when running from repo root
+sys.path.insert(0, "aws/lambda/pytorch-auto-revert")
+
+from pytorch_auto_revert.github_client_helper import GHClientFactory
+from pytorch_auto_revert.workflow_resolver import WorkflowResolver
+
+
+class TestWorkflowResolverRealRepo(unittest.TestCase):
+    @classmethod
+    def setUpClass(cls):
+        token = os.environ.get("GITHUB_TOKEN") or os.environ.get("GH_TOKEN")
+        if token:
+            GHClientFactory.setup_client(
+                app_id="",
+                app_secret="",
+                installation_id=0,
+                token=token,
+            )
+
+    def test_resolve_pull_workflow(self):
+        resolver = WorkflowResolver.get("pytorch/pytorch")
+
+        # Resolve by display name
+        pull_by_name = resolver.resolve("pull")
+        self.assertIsNotNone(pull_by_name, "Expected to resolve 'pull' by display name")
+
+        # Resolve by basename
+        pull_by_file = resolver.resolve("pull.yml")
+        self.assertIsNotNone(
+            pull_by_file, "Expected to resolve 'pull.yml' by file name"
+        )
+        self.assertTrue(
+            pull_by_file.file_name.endswith("pull.yml"),
+            "Resolved file name should be 'pull.yml'",
+        )
+
+    def test_resolve_trunk_workflow(self):
+        resolver = WorkflowResolver.get("pytorch/pytorch")
+
+        trunk_by_name = resolver.resolve("trunk")
+        self.assertIsNotNone(
+            trunk_by_name, "Expected to resolve 'trunk' by display name"
+        )
+
+        trunk_by_file = resolver.resolve("trunk.yml")
+        self.assertIsNotNone(
+            trunk_by_file, "Expected to resolve 'trunk.yml' by file name"
+        )
+        self.assertTrue(
+            trunk_by_file.file_name.endswith("trunk.yml"),
+            "Resolved file name should be 'trunk.yml'",
+        )
+
+
+if __name__ == "__main__":
+    unittest.main()

aws/lambda/pytorch-auto-revert/pytorch_auto_revert/workflow_checker.py

@@ -1,12 +1,15 @@
 """
-WorkflowRestartChecker for querying restarted workflows via ClickHouse.
+WorkflowRestartChecker for querying restarted workflows via ClickHouse and
+dispatching workflows via GitHub with consistent workflow name resolution.
 """
 
 import logging
 from datetime import datetime, timedelta
+from functools import cached_property
 from typing import Dict, Set
 
 from .clickhouse_client_helper import CHCliFactory
+from .workflow_resolver import WorkflowResolver
 
 
 class WorkflowRestartChecker:
@@ -28,9 +31,10 @@ def has_restarted_workflow(self, workflow_name: str, commit_sha: str) -> bool:
         Returns:
             bool: True if workflow was restarted (workflow_dispatch with trunk/* branch)
         """
-        # Normalize workflow name - remove .yml extension for consistency
-        normalized_workflow_name = workflow_name.replace(".yml", "")
-        cache_key = f"{normalized_workflow_name}:{commit_sha}"
+        # Resolve to display name via GitHub (exact display or file name)
+        display_name = self.resolver.require(workflow_name).display_name
+
+        cache_key = f"{display_name}:{commit_sha}"
         if cache_key in self._cache:
             return self._cache[cache_key]
 
@@ -54,7 +58,7 @@ def has_restarted_workflow(self, workflow_name: str, commit_sha: str) -> bool:
                 "commit_sha": commit_sha,
                 "workflow_event": "workflow_dispatch",
                 "head_branch": f"trunk/{commit_sha}",
-                "workflow_name": normalized_workflow_name,
+                "workflow_name": display_name,
             },
         )
 
@@ -73,8 +77,7 @@ def get_restarted_commits(self, workflow_name: str, days_back: int = 7) -> Set[s
         Returns:
             Set of commit SHAs that have restarted workflows
         """
-        # Normalize workflow name - remove .yml extension for consistency
-        normalized_workflow_name = workflow_name.replace(".yml", "")
+        display_name = self.resolver.require(workflow_name).display_name
         since_date = datetime.now() - timedelta(days=days_back)
 
         query = """
@@ -87,14 +90,14 @@ def get_restarted_commits(self, workflow_name: str, days_back: int = 7) -> Set[s
         """
 
         result = CHCliFactory().client.query(
-            query, {"workflow_name": normalized_workflow_name, "since_date": since_date}
+            query, {"workflow_name": display_name, "since_date": since_date}
         )
 
         commits = {row[0] for row in result.result_rows}
 
         # Update cache
         for commit_sha in commits:
-            cache_key = f"{normalized_workflow_name}:{commit_sha}"
+            cache_key = f"{display_name}:{commit_sha}"
             self._cache[cache_key] = True
 
         return commits
@@ -114,13 +117,10 @@ def restart_workflow(self, workflow_name: str, commit_sha: str) -> bool:
         Returns:
             bool: True if workflow was successfully dispatched, False otherwise
         """
-        # Normalize workflow name
-        normalized_workflow_name = workflow_name.replace(".yml", "")
-
         # Check if already restarted
-        if self.has_restarted_workflow(normalized_workflow_name, commit_sha):
+        if self.has_restarted_workflow(workflow_name, commit_sha):
             logging.warning(
-                f"Workflow {normalized_workflow_name} already restarted for commit {commit_sha}"
+                f"Workflow {workflow_name} already restarted for commit {commit_sha}"
             )
             return False
 
@@ -144,33 +144,35 @@ def restart_workflow(self, workflow_name: str, commit_sha: str) -> bool:
             # Use trunk/{sha} tag format
             tag_ref = f"trunk/{commit_sha}"
 
-            # Add .yml extension for workflow name
-            workflow_file_name = f"{normalized_workflow_name}.yml"
+            # Resolve workflow
+            wf_ref = self.resolver.require(workflow_name)
 
-            # Get repo and workflow objects
-            repo = client.get_repo(f"{self.repo_owner}/{self.repo_name}")
-            workflow = repo.get_workflow(workflow_file_name)
-
-            # Dispatch the workflow
-            workflow.create_dispatch(ref=tag_ref, inputs={})
+            # Dispatch via file name
+            client.get_repo(f"{self.repo_owner}/{self.repo_name}").get_workflow(
+                wf_ref.file_name
+            ).create_dispatch(ref=tag_ref, inputs={})
 
             # Construct the workflow runs URL
             workflow_url = (
                 f"https://github.com/{self.repo_owner}/{self.repo_name}"
-                f"/actions/workflows/{workflow_file_name}"
+                f"/actions/workflows/{wf_ref.file_name}"
                 f"?query=branch%3Atrunk%2F{commit_sha}"
             )
             logging.info(
-                f"Successfully dispatched workflow {normalized_workflow_name} for commit {commit_sha}\n"
+                f"Successfully dispatched workflow {wf_ref.display_name} for commit {commit_sha}\n"
                 f"  View at: {workflow_url}"
             )
 
             # Invalidate cache for this workflow/commit
-            cache_key = f"{normalized_workflow_name}:{commit_sha}"
+            cache_key = f"{wf_ref.display_name}:{commit_sha}"
             if cache_key in self._cache:
                 del self._cache[cache_key]
             return True
 
         except Exception as e:
-            logging.error(f"Error dispatching workflow {normalized_workflow_name}: {e}")
+            logging.error(f"Error dispatching workflow {workflow_name}: {e}")
             return False
+
+    @cached_property
+    def resolver(self) -> WorkflowResolver:
+        return WorkflowResolver.get(f"{self.repo_owner}/{self.repo_name}")

aws/lambda/pytorch-auto-revert/pytorch_auto_revert/workflow_resolver.py

@@ -0,0 +1,99 @@
+"""
+WorkflowResolver: Resolve GitHub Actions workflows by exact display or file name.
+
+- Exact matches only (no lowercasing or fuzzy matching)
+- Caches per-repo resolver instances
+"""
+
+from __future__ import annotations
+
+import os
+from dataclasses import dataclass
+from functools import lru_cache
+from typing import Optional
+
+import github
+
+from .github_client_helper import GHClientFactory
+
+
+@dataclass(frozen=True)
+class WorkflowRef:
+    """Reference to a workflow's identities in a repository."""
+
+    display_name: str
+    file_name: str  # basename, e.g., "pull.yml"
+
+
+class WorkflowResolver:
+    """Caches workflows for a repo and resolves by exact names.
+
+    Usage:
+        resolver = WorkflowResolver.get("owner/repo")
+        wf = resolver.resolve("pull")        # display name
+        wf = resolver.resolve("pull.yml")    # file basename
+    """
+
+    def __init__(
+        self, repo_full_name: str, repository: "github.Repository.Repository"
+    ) -> None:
+        self._repo_full_name = repo_full_name
+        self._repository = repository
+        self._by_display: dict[str, WorkflowRef] = {}
+        self._by_file: dict[str, WorkflowRef] = {}
+        self._build_indices()
+
+    @staticmethod
+    @lru_cache(maxsize=None)
+    def get(repo: str) -> "WorkflowResolver":
+        """Get a cached resolver for a repo in owner/repo format.
+
+        Internally creates a GitHub Repository client using GHClientFactory when
+        available; otherwise falls back to an anonymous client for public repos.
+        """
+        # Build a client: prefer configured factory; fall back to anonymous
+        try:
+            client = GHClientFactory().client
+        except Exception:
+            # Anonymous client for public data; may be rate limited
+            client = github.Github()
+
+        repository = client.get_repo(repo)
+        return WorkflowResolver(repo_full_name=repo, repository=repository)
+
+    def resolve(self, input_name: str) -> Optional[WorkflowRef]:
+        """Resolve by exact display name, file basename, or full path.
+
+        Returns None if no exact match is found.
+        """
+        if input_name in self._by_display:
+            return self._by_display[input_name]
+        if input_name in self._by_file:
+            return self._by_file[input_name]
+        return None
+
+    def require(self, input_name: str) -> WorkflowRef:
+        """Resolve or raise ValueError with a helpful message."""
+        ref = self.resolve(input_name)
+        if ref is None:
+            # Build an informative message with available names
+            display = ", ".join(sorted(self._by_display))
+            files = ", ".join(sorted(self._by_file))
+            raise ValueError(
+                f"Workflow '{input_name}' not found in {self._repo_full_name}. "
+                f"Available display names: [{display}]. Available files: [{files}]"
+            )
+        return ref
+
+    # Internal helpers
+
+    def _build_indices(self) -> None:
+        for w in self._repository.get_workflows():
+            name = getattr(w, "name", "") or ""
+            path = getattr(w, "path", "") or ""
+            base = os.path.basename(path) if path else ""
+            if not (name and base):
+                continue
+            ref = WorkflowRef(display_name=name, file_name=base)
+            self._by_display[name] = ref
+            self._by_file[base] = ref