ci: Fix !bencher-thresholds-reset PR comment

.github/actions/bencher-track/action.yml

@@ -1,18 +1,18 @@
 name: Track benchmarks on bencher
 description: >-
   Run `bencher run` for the `ix` project. Every measure's baseline is windowed
-  to the data points recorded since the workload's `baseline-reset-<workload>`
+  to the data points recorded since the workload's `bencher-thresholds-reset-<workload>`
   tag, so a toolchain bump (or any permanent shift) is recalibrated by moving
   that one tag — via the !bencher-thresholds-reset PR comment or the
-  bench-thresholds-reset workflow. The calling job must check out with
+  bencher-thresholds-reset workflow. The calling job must check out with
   fetch-depth: 0 and fetch-tags: true (the tag is read here).
 
 inputs:
   testbed:
     description: Bencher testbed slug.
     required: true
   workload:
-    description: Workload key for the `baseline-reset-<workload>` tag (e.g. ix-compile, aiur).
+    description: Workload key for the `bencher-thresholds-reset-<workload>` tag (e.g. ix-compile, aiur).
     required: true
   file:
     description: Bencher Metric Format JSON file to upload.
@@ -45,7 +45,7 @@ runs:
         THRESHOLDS: ${{ inputs.thresholds }}
       run: |
         # Baseline window: data points for this benchmark since the workload's
-        # `baseline-reset-<workload>` tag (moved on a toolchain bump or any
+        # `bencher-thresholds-reset-<workload>` tag (moved on a toolchain bump or any
         # permanent shift). The same window applies to every measure. We count
         # actual data points, not commits, so the baseline never reaches past
         # the anchor (a failed run produces no metric, so a commit count would
@@ -54,7 +54,7 @@ runs:
         # project is public, so no token is needed. With no tag yet, fall back
         # to Bencher's 64-point window.
         sample=64
-        if reset_epoch=$(git log -1 --format=%ct "baseline-reset-$WORKLOAD" 2>/dev/null); then
+        if reset_epoch=$(git log -1 --format=%ct "bencher-thresholds-reset-$WORKLOAD" 2>/dev/null); then
           bench=$(jq -r '[keys[]][0]' "$FILE")
           sample=$(curl -fsS "https://api.bencher.dev/v0/projects/ix/reports?branch=${GITHUB_REF_NAME}&testbed=${TESTBED}&start_time=$((reset_epoch * 1000))&per_page=255" \
             | jq --arg b "$bench" '[.[] | select([.results[]?[]?.benchmark.name] | index($b))] | length')

.github/workflows/bench-main.yml

@@ -81,7 +81,7 @@ jobs:
       - uses: actions/checkout@v6
         with:
           fetch-depth: 0
-          fetch-tags: true   # bencher-track reads the baseline-reset tag
+          fetch-tags: true   # bencher-track reads the bencher-thresholds-reset tag
       - uses: actions/cache/restore@v5
         with:
           path: ~/.local/bin
@@ -182,7 +182,7 @@ jobs:
       - uses: actions/checkout@v6
         with:
           fetch-depth: 0     # full history for the baseline-anchor lookup
-          fetch-tags: true   # bencher-track reads the baseline-reset tag
+          fetch-tags: true   # bencher-track reads the bencher-thresholds-reset tag
       - uses: actions/cache/restore@v5
         with:
           path: ~/.local/bin

.github/workflows/bencher-thresholds-reset.yml

@@ -0,0 +1,146 @@
+name: Bencher thresholds reset
+
+# Move `bencher-thresholds-reset-<workload>` tag(s) so bencher-track re-windows those
+# measures from a chosen commit — after a toolchain bump or any permanent shift.
+# Other workloads are unaffected; historical data stays in Bencher, only the
+# alert baseline is re-anchored. Two ways to trigger:
+#
+#   - Maintainer comment `!bencher-thresholds-reset <workload|all>` on a PR:
+#     the `ack` job (issue_comment) is the single place that parses/validates the
+#     command. It records each valid workload as a `bencher-thresholds-reset:<workload>`
+#     PR label and replies to confirm (or warns on an unknown/missing workload,
+#     catching a typo immediately). On merge, the `reset` job simply reads those
+#     labels from the event payload — no re-parsing — and anchors the tags at the
+#     merge commit. The label is the shared flag between the two events, and is
+#     human-visible/cancelable (remove the label to cancel a queued reset).
+#   - Manual label (no comment needed): the merge step honors any
+#     `bencher-thresholds-reset:<token>` label on the PR, whatever added it — so a
+#     Triage+ collaborator can queue a reset by applying the label directly, and
+#     cancel by removing it before merge. Naming convention: one label per token,
+#     `bencher-thresholds-reset:<token>` where <token> is `ix-compile`, `aiur`, or
+#     `all` (the merge step expands an `all` label into every workload). Labeling
+#     requires Triage+, so PR authors from forks cannot self-queue a reset. The
+#     label shares the command/workflow name; the git tag it moves is the same
+#     stem with a dash: `bencher-thresholds-reset-<workload>`.
+#   - workflow_dispatch: reset the chosen workload at a given commit (default
+#     HEAD); does not read PR contents. For ad-hoc or bootstrap resets.
+#
+# pull_request_target / issue_comment both run in the base-repo context with a
+# write token (works for fork PRs too). Safe here because the jobs run no
+# PR-provided code — only `gh` calls against the trusted event payload, with the
+# comment body used purely as data (never evaluated).
+on:
+  workflow_dispatch:
+    inputs:
+      workload:
+        description: Workload baseline to reset
+        required: true
+        type: choice
+        options: [ix-compile, aiur, all]
+      sha:
+        description: "Commit to anchor to (default: HEAD)"
+        required: false
+  pull_request_target:
+    types: [closed]
+  issue_comment:
+    types: [created]
+
+permissions:
+  contents: write
+  pull-requests: write
+  issues: write # ack creates/applies the bencher-thresholds-reset:<workload> label
+
+jobs:
+  reset:
+    if: github.event_name == 'workflow_dispatch' || github.event.pull_request.merged == true
+    runs-on: ubuntu-latest
+    env:
+      GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+      REPO: ${{ github.repository }}
+      EVENT: ${{ github.event_name }}
+      INPUT_WORKLOAD: ${{ inputs.workload }}
+      INPUT_SHA: ${{ inputs.sha }}
+      HEAD_SHA: ${{ github.sha }}
+      PR: ${{ github.event.pull_request.number }}
+      MERGE_SHA: ${{ github.event.pull_request.merge_commit_sha }}
+    steps:
+      - run: |
+          valid="ix-compile aiur"
+          if [ "$EVENT" = workflow_dispatch ]; then
+            # Reset the chosen workload(s) at the given commit; no PR scan.
+            sha="${INPUT_SHA:-$HEAD_SHA}"
+            [ "$INPUT_WORKLOAD" = all ] && workloads="$valid" || workloads="$INPUT_WORKLOAD"
+          else
+            # Reset the workloads the `ack` job recorded as `bencher-thresholds-reset:*`
+            # labels (already parsed/validated at comment time), anchored at the
+            # merge commit. Read straight from the event payload — no re-parsing.
+            # A `bencher-thresholds-reset:all` label expands to every workload.
+            sha="$MERGE_SHA"
+            workloads=$(jq -r '.pull_request.labels[].name
+              | select(startswith("bencher-thresholds-reset:")) | ltrimstr("bencher-thresholds-reset:")' \
+              "$GITHUB_EVENT_PATH")
+            echo "$workloads" | grep -qx all && workloads="$valid"
+          fi
+
+          done_list=""
+          for w in $workloads; do
+            case " $valid " in *" $w "*) ;; *) echo "skip unknown workload: $w"; continue ;; esac
+            tag="bencher-thresholds-reset-$w"
+            echo "Anchoring $tag -> $sha"
+            # Update the tag if it exists, else create it. (Checking first avoids
+            # the spurious 422 a PATCH-then-POST logs on first creation.)
+            if gh api --silent "repos/$REPO/git/refs/tags/$tag" 2>/dev/null; then
+              gh api --silent -X PATCH "repos/$REPO/git/refs/tags/$tag" -f sha="$sha" -F force=true
+            else
+              gh api --silent -X POST "repos/$REPO/git/refs" -f ref="refs/tags/$tag" -f sha="$sha"
+            fi
+            done_list="$done_list $w"
+          done
+          [ -z "$done_list" ] && { echo "Nothing to reset."; exit 0; }
+
+          # Confirm on the PR for the merge-triggered path.
+          [ "$EVENT" = pull_request_target ] && gh pr comment "$PR" --repo "$REPO" \
+            --body "♻️ Baseline reset to \`${sha:0:7}\` for:$done_list"
+          true
+
+  # Immediate acknowledgment: when a maintainer comments the reset command on a
+  # still-open PR, reply that the named baseline(s) will be reset on merge — or
+  # warn if the workload is unknown/missing. The `user.type != 'Bot'` guard keeps
+  # this reply (which may echo the command) from re-triggering the job.
+  ack:
+    if: github.event_name == 'issue_comment' && github.event.issue.pull_request && github.event.issue.state == 'open' && github.event.comment.user.type != 'Bot' && contains(github.event.comment.body, '!bencher-thresholds-reset') && (github.event.comment.author_association == 'OWNER' || github.event.comment.author_association == 'MEMBER' || github.event.comment.author_association == 'COLLABORATOR')
+    runs-on: ubuntu-latest
+    env:
+      GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+      REPO: ${{ github.repository }}
+      PR: ${{ github.event.issue.number }}
+      BODY: ${{ github.event.comment.body }}
+    steps:
+      - run: |
+          # Accepted command tokens — applied verbatim as labels (incl. `all`,
+          # which the merge job expands into every workload).
+          accepted="ix-compile aiur all"
+          # Parse the workload token(s) after the command, lowercased.
+          workloads=$(printf '%s' "$BODY" \
+            | grep -oiE '!bencher-thresholds-reset[[:space:]]+[a-z0-9 -]+' \
+            | sed -E 's/^!bencher-thresholds-reset[[:space:]]+//I' \
+            | tr '[:upper:] ' '[:lower:]\n' | sort -u)
+          ok=""
+          for w in $workloads; do
+            case " $accepted " in *" $w "*) ok="$ok $w" ;; *) ;; esac
+          done
+          if [ -n "$ok" ]; then
+            # Record each workload as a label — the flag the merge-time `reset`
+            # job reads. Create the label on first use (idempotent), then apply.
+            for w in $ok; do
+              label="bencher-thresholds-reset:$w"
+              gh label create "$label" --repo "$REPO" --color BFD4F2 \
+                --description "bencher baseline reset queued on merge" 2>/dev/null || true
+              gh pr edit "$PR" --repo "$REPO" --add-label "$label"
+            done
+            gh pr comment "$PR" --repo "$REPO" \
+              --body "♻️ Baseline reset queued for:$ok — will anchor to the merge commit when this PR merges."
+          else
+            gh pr comment "$PR" --repo "$REPO" \
+              --body "⚠️ Reset command matched no known workload (expected \`ix-compile\`, \`aiur\`, or \`all\`). Nothing will reset on merge."
+          fi