Fix module cleanup workflow (open-telemetry#18658)

Author: trask

.github/scripts/module-cleanup/build-cleanup-matrix.py

@@ -13,8 +13,7 @@
   GITHUB_OUTPUT   - path to the GitHub Actions output file
   GH_TOKEN        - token for `gh` CLI (set automatically by the workflow)
   REVIEW_PROGRESS - newline-separated list of processed module names
-                    (contents of processed.txt on the memory branch, plus
-                    shorts already in inflight module-cleanup PR bodies)
+                    (contents of processed.txt on the memory branch)
 
 Outputs (to $GITHUB_OUTPUT):
   has_work        - "true" if a module was picked, "false" otherwise

.github/scripts/module-cleanup/export-cleanup-patch.sh

@@ -2,7 +2,7 @@
 # Final action invoked by the LLM agent: format-patch the cleanup commit
 # range into /tmp/gh-aw/agent/cleanup.patch so gh-aw's auto-uploader
 # includes it in the `agent` workflow artifact. The finalize job then
-# downloads that artifact and applies the patch onto module-cleanup-wip.
+# downloads that artifact and applies the patch onto otelbot/module-cleanup-wip.
 #
 # Idempotent and write-only to /tmp. Does NOT push anything.
 #

.github/scripts/module-cleanup/finalize.sh

@@ -1,5 +1,5 @@
 #!/bin/bash
-# Finalize: single writer for both module-cleanup-wip and the
+# Finalize: single writer for both otelbot/module-cleanup-wip and the
 # memory/module-cleanup branch. Runs after the agent job (regardless of
 # whether the agent succeeded, no-oped, or failed).
 #
@@ -9,24 +9,23 @@
 #      module is recorded as "processed" (so it isn't retried in a loop)
 #      AND logged as a failure for diagnostics.
 #   2. If the agent produced a cleanup patch, apply it onto the fixed
-#      module-cleanup-wip branch and push.
+#      otelbot/module-cleanup-wip branch and push.
 #   3. If wip diff vs origin/main has reached FLUSH_THRESHOLD files OR
 #      the queue is empty, atomically rename wip to a
 #      module-cleanup-batch-<run_id> branch and open the PR. The wip
 #      branch ceases to exist on remote until the next run recreates
 #      it from main.
-#   4. Self-dispatch the workflow unless the queue is empty. The chain
-#      stops on its own once MAX_OPEN_PRS is reached (matrix script
-#      returns has_work=false and finalize is skipped).
+#   4. Leave self-dispatch to the workflow step that follows this script,
+#      so PR creation can use the app token while workflow dispatch uses
+#      the normal GitHub Actions token.
 #
 # No rebase-retry loops on push: the workflow uses
 # concurrency.group=module-cleanup with cancel-in-progress=false, so this
 # job is the only writer of either branch and runs serialized across
 # workflow runs.
 #
 # Required env:
-#   GH_TOKEN          - token with contents:write, pull-requests:write,
-#                       and actions:write
+#   GH_TOKEN          - token with contents:write and pull-requests:write
 #   GITHUB_REPOSITORY - owner/repo
 #   SHORT_NAME        - the module short_name processed this run
 #   AGENT_RESULT      - github.needs.agent.result ('success'|'failure'|...)
@@ -36,25 +35,26 @@
 #
 # Optional env:
 #   FLUSH_THRESHOLD   - file count that triggers a PR (default 10)
-#   WORKFLOW_FILE     - workflow file name for self-dispatch
 #   MEMORY_BRANCH     - default: memory/module-cleanup
-#   WIP_BRANCH        - default: module-cleanup-wip
+#   WIP_BRANCH        - default: otelbot/module-cleanup-wip
 
 set -euo pipefail
 
 MEMORY_BRANCH="${MEMORY_BRANCH:-memory/module-cleanup}"
-WIP_BRANCH="${WIP_BRANCH:-module-cleanup-wip}"
+WIP_BRANCH="${WIP_BRANCH:-otelbot/module-cleanup-wip}"
 THRESHOLD="${FLUSH_THRESHOLD:-10}"
 QUEUE_REMAINING="${QUEUE_REMAINING:-0}"
 REPO="${GITHUB_REPOSITORY:?GITHUB_REPOSITORY required}"
-WORKFLOW_FILE="${WORKFLOW_FILE:-module-cleanup.lock.yml}"
 SHORT="${SHORT_NAME:?SHORT_NAME required}"
 AGENT_RESULT="${AGENT_RESULT:-failure}"
 ARTIFACT_DIR="${ARTIFACT_DIR:-./agent-artifact}"
 
-git fetch origin main --depth=1
-git fetch origin "$MEMORY_BRANCH" --depth=1 2>/dev/null || true
-git fetch origin "$WIP_BRANCH"    --depth=1 2>/dev/null || true
+# Full history is required for `origin/main..origin/$WIP_BRANCH` log/diff
+# below. The finalize job's checkout uses `fetch-depth: 0`, so don't
+# re-shallow any of these refs with `--depth`.
+git fetch origin main
+git fetch origin "$MEMORY_BRANCH" 2>/dev/null || true
+git fetch origin "$WIP_BRANCH"    2>/dev/null || true
 
 # ---- 1. Update processed.txt (and failed.txt on failure) ----
 
@@ -136,7 +136,7 @@ if [ -n "$PATCH_SRC" ]; then
     )
 fi
 
-git fetch origin "$WIP_BRANCH" --depth=50 2>/dev/null || true
+git fetch origin "$WIP_BRANCH" 2>/dev/null || true
 
 # ---- 3. Decide flush ----
 
@@ -170,6 +170,8 @@ if [ "$SHOULD_FLUSH" = "true" ]; then
 
     MODULE_COUNT=$(git -C "$WIP_WT" rev-list --count "origin/main..origin/$WIP_BRANCH")
 
+    # processed.txt on the memory branch is the sole source of truth for
+    # which modules dispatch skips, so no hidden marker block is needed here.
     BODY_FILE=$(mktemp)
     {
         echo "Automated module-cleanup batch."
@@ -210,21 +212,5 @@ if [ "$SHOULD_FLUSH" = "true" ]; then
     OPENED_PR=true
 fi
 
-# ---- 4. Self-dispatch ----
-
-# Always self-dispatch when there's more queued work. The next run will
-# pick up wherever wip is: if we just flushed, wip is gone and the run
-# starts a fresh wip from main; otherwise it appends to the same wip.
-# The chain stops on its own when build-cleanup-matrix.py sees
-# MAX_OPEN_PRS reached and returns has_work=false (no agent, no
-# finalize, no self-dispatch). Cron picks back up later.
-
-if [ "$QUEUE_REMAINING" -le 0 ]; then
-    echo "Queue empty; nothing to dispatch."
-else
-    echo "Self-dispatching workflow for next module."
-    gh workflow run "$WORKFLOW_FILE" --repo "$REPO" --ref main
-fi
-
 git worktree remove --force "$MEM_WT" 2>/dev/null || true
 git worktree remove --force "$WIP_WT" 2>/dev/null || true