fix(release): watch the run for this tag, and let a re-run converge

.github/workflows/release.yml

@@ -131,11 +131,26 @@ jobs:
              found {print}' CHANGELOG.md > release-notes.md
           cat release-notes.md
 
-      - name: Create GitHub Release
+      # Converge rather than create. `gh release create` refuses an existing
+      # tag, so re-dispatching a release that already got this far failed here
+      # every time — after build and publish had both succeeded, which is the
+      # most confusing place to stop. workflow_dispatch was added to this
+      # workflow so a release could be re-driven through a webhook outage; a
+      # final step that cannot survive a second run takes that back.
+      - name: Create or update GitHub Release
         env:
           GH_TOKEN: ${{ github.token }}
         run: |
-          gh release create "${TAG}" just-bashit.tar.gz \
-            --title "just-bashit ${TAG}" \
-            --latest \
-            --notes-file release-notes.md
+          if gh release view "${TAG}" >/dev/null 2>&1; then
+            echo "Release ${TAG} already exists — updating it in place"
+            gh release edit "${TAG}" \
+              --title "just-bashit ${TAG}" \
+              --latest \
+              --notes-file release-notes.md
+            gh release upload "${TAG}" just-bashit.tar.gz --clobber
+          else
+            gh release create "${TAG}" just-bashit.tar.gz \
+              --title "just-bashit ${TAG}" \
+              --latest \
+              --notes-file release-notes.md
+          fi

CHANGELOG.md

@@ -2,6 +2,28 @@
 
 ## [Unreleased]
 
+### Fixed
+
+- **`make ship` could report a release as failed while it succeeded.**
+    `release-watch` took the newest release run with `--limit 1`, but `ship`
+    is `tag-release` then `release-watch`, and the tag push returns before
+    GitHub has created the run — so the newest row was still the *previous*
+    release. Shipping v0.4.1 watched a run that had failed the night before
+    and exited 1 while the real release went on to succeed. A watcher that can
+    report the wrong run is worse than no watcher, because the failure it
+    invents looks exactly like a real one. It now selects the run by tag —
+    `--branch "v$VERSION"`, which is what a tag-triggered run carries — and
+    waits up to 60s for it to exist, erroring clearly if it never does.
+
+- **A re-dispatched release always failed at the last step.** Since
+    `gh release create` refuses an existing tag, re-driving a release that had
+    already got as far as publishing died on "a release with the same tag name
+    already exists" — after build and publish had both succeeded. The
+    `workflow_dispatch` trigger was added to this workflow so a release could
+    be re-driven through a webhook outage; a final step that cannot survive a
+    second run takes that ability back. It now updates the release in place
+    when one already exists.
+
 ## [0.4.1] - 2026-08-06
 
 ### Added

Makefile

@@ -152,7 +152,34 @@ just-runit|sed -n 's/^_VERSION="\(.*\)"/\1/p' src/just_bashit/just-runit
 src/ headers|grep -h '^# PACKAGE' src/just_bashit/*.sh | sed 's/.*version \([0-9.]*\).*/\1/' | sort -u
 endef
 
-RELEASE_WATCH_CMD = gh run watch --exit-status $$(gh run list \
-    --workflow=release.yml --limit 1 --json databaseId -q '.[0].databaseId')
+# Select the run BY TAG, and wait for it to exist.
+#
+# `--limit 1` alone is a race: `ship` is `tag-release` then `release-watch`, and
+# the push returns before GitHub has created the workflow run, so the newest
+# row is still the PREVIOUS release. Shipping v0.4.1 watched the run that had
+# failed the night before and reported the release as failed while it went on
+# to succeed — a watcher that can report the wrong run is worse than none,
+# because the failure it invents is indistinguishable from a real one.
+#
+# A tag-triggered run carries the tag in headBranch, so --branch pins it to
+# exactly this release. The wait covers creation latency, which is a second or
+# two; 60s of headroom costs nothing on the happy path and turns "no run yet"
+# into a clear error rather than an empty run ID.
+define RELEASE_WATCH_CMD
+@tag="v$(VERSION)"; id=""; \
+ for _ in $$(seq 1 30); do \
+     id=$$(gh run list --workflow=release.yml --branch "$$tag" \
+           --limit 1 --json databaseId -q '.[0].databaseId' 2>/dev/null); \
+     [ -n "$$id" ] && break; \
+     sleep 2; \
+ done; \
+ if [ -z "$$id" ]; then \
+     echo "ERROR: no release run for $$tag after 60s"; \
+     echo "  check: gh run list --workflow=release.yml --branch $$tag"; \
+     exit 1; \
+ fi; \
+ echo "Watching release run $$id ($$tag)"; \
+ gh run watch --exit-status "$$id"
+endef
 
 include standard.mk