Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
104 changes: 103 additions & 1 deletion .github/workflows/release-please.yml
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,8 @@ on:
permissions:
contents: write
pull-requests: write
# verify-ci's `gh api` calls read the CI workflow's runs + jobs.
actions: read

concurrency:
group: ${{ github.workflow }}-${{ github.ref }}
Expand Down Expand Up @@ -72,6 +74,106 @@ jobs:
echo "version=${{ steps.release.outputs.version }}" >> "$GITHUB_OUTPUT"
fi

# ----------------------------------------------------------------
# verify-ci: gate the release on the SAME tag commit's CI result.
#
# We do NOT re-run cargo tests here — we QUERY the existing `CI` workflow
# run for the exact tag commit SHA (event=push, branch=main) and wait for
# its `CI Success` job to conclude. Gating on that single job (which itself
# needs test/audit/windows) is the right wait point: it excludes the slow
# informational Coverage job.
#
# Topology: this runs IN PARALLEL with build-binaries (build-binaries needs
# only release-please), so the ~430s cross-compile overlaps with the CI wait.
# The gate is enforced downstream at upload-assets (which lists verify-ci in
# `needs`) — the first public-facing step that touches the GitHub Release.
# build-binaries only produces throwaway artifacts (retention-days: 1), so a
# failed/timed-out verify-ci blocks upload-assets + publish-release and leaves
# the Release a DRAFT, without weakening the gate.
#
# This job ALWAYS RUNS when a tag was cut (build_artifacts=='true'); it must
# never be SKIPPED, because upload-assets lists it in `needs` and a skipped
# needed job would cascade-skip upload-assets. On the workflow_dispatch
# (manual re-attach) path there is no new CI run, so the job short-circuits
# to success as a no-op; the polling steps run only on the push path.
# ----------------------------------------------------------------
verify-ci:
name: Verify CI Passed
needs: release-please
if: needs.release-please.outputs.build_artifacts == 'true'
runs-on: ubuntu-latest
timeout-minutes: 30
steps:
- name: Manual re-attach — skip CI gate (no-op success)
if: github.event_name == 'workflow_dispatch'
run: |
echo "workflow_dispatch: re-attaching binaries to an existing tag; no new CI run to gate on — passing."

- name: Checkout tag
if: github.event_name == 'push'
uses: actions/checkout@v5
with:
fetch-depth: 0
ref: ${{ needs.release-please.outputs.tag_name }}

- name: Resolve tag commit SHA
id: sha
if: github.event_name == 'push'
run: echo "sha=$(git rev-list -n1 '${{ needs.release-please.outputs.tag_name }}')" >> "$GITHUB_OUTPUT"

- name: Wait for CI Success on the tag commit
if: github.event_name == 'push'
env:
GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
REPO: ${{ github.repository }}
HEAD_SHA: ${{ steps.sha.outputs.sha }}
run: |
set -euo pipefail
# Bounded poll: ~25 min cap (100 iterations x 15s) inside timeout-minutes: 30.
max_iters=100
sleep_secs=15
run_id=""
for i in $(seq 1 "$max_iters"); do
# Select the CI workflow's push run on main for the exact tag SHA;
# if several exist, take the most recent by created_at.
run_id=$(gh api "repos/${REPO}/actions/runs?head_sha=${HEAD_SHA}&event=push&branch=main&per_page=100" \
--jq '[.workflow_runs[] | select(.name == "CI")] | sort_by(.created_at) | last | .id // empty')
if [ -n "$run_id" ]; then
echo "Found CI run ${run_id} for ${HEAD_SHA}"
break
fi
echo "No CI run for ${HEAD_SHA} yet (attempt ${i}/${max_iters}); sleeping ${sleep_secs}s..."
sleep "$sleep_secs"
done
if [ -z "$run_id" ]; then
echo "❌ Timed out waiting for a CI run on ${HEAD_SHA}; leaving release as draft."
exit 1
fi
for i in $(seq 1 "$max_iters"); do
status=$(gh api "repos/${REPO}/actions/runs/${run_id}/jobs?per_page=100" \
--jq '.jobs[] | select(.name == "CI Success") | .status')
conclusion=$(gh api "repos/${REPO}/actions/runs/${run_id}/jobs?per_page=100" \
--jq '.jobs[] | select(.name == "CI Success") | .conclusion // ""')
if [ -z "$status" ]; then
echo "CI Success job not present in run ${run_id} yet (attempt ${i}/${max_iters}); sleeping ${sleep_secs}s..."
sleep "$sleep_secs"
continue
fi
if [ "$status" != "completed" ]; then
echo "CI Success job status=${status} (attempt ${i}/${max_iters}); sleeping ${sleep_secs}s..."
sleep "$sleep_secs"
continue
fi
if [ "$conclusion" = "success" ]; then
echo "✅ CI Success passed for ${HEAD_SHA}."
exit 0
fi
echo "❌ CI Success concluded '${conclusion}' for ${HEAD_SHA}; blocking release (draft preserved)."
exit 1
done
echo "❌ Timed out waiting for CI Success to complete on ${HEAD_SHA}; leaving release as draft."
exit 1

# ----------------------------------------------------------------
# build-binaries: cross-compile release artifacts for all targets.
# Gated on a cut tag. The CLI crate is `codegraph-rs`, binary `codegraph`.
Expand Down Expand Up @@ -182,7 +284,7 @@ jobs:
# ----------------------------------------------------------------
upload-assets:
name: Upload Release Assets
needs: [release-please, build-binaries]
needs: [release-please, build-binaries, verify-ci]
if: needs.release-please.outputs.build_artifacts == 'true'
runs-on: ubuntu-latest
timeout-minutes: 15
Expand Down
37 changes: 20 additions & 17 deletions crates/codegraph-mcp/tests/reopen.rs
Original file line number Diff line number Diff line change
Expand Up @@ -134,31 +134,34 @@ fn reopens_cached_engine_when_db_replaced_with_new_inode() {
);
let reopens_before = reopen_count();

// When: the db is REPLACED on disk with a fresh index (new identity) whose
// content differs (a new symbol `betaSymbol`). The cached engine still holds
// an open WAL connection, which on windows blocks deleting the db dir
// (OS error 32). Drop the live connections first — `close_cached_handles`
// keeps the cache RECORD (with its old identity) while releasing the file
// handle, so the next call still detects the identity change and counts
// exactly one reopen. Removing the dir first guarantees a freshly built db.
let id_before = db_identity(project.path());
// When: the db is REPLACED on disk with a fresh index whose content differs
// (a new symbol `betaSymbol`). The cached engine still holds an open WAL
// connection, which on windows blocks deleting the db dir (OS error 32).
// Drop the live connections first — `close_cached_handles` releases the file
// handle while keeping the cache RECORD, so the next call reopens exactly
// once and re-reads the freshly built db from disk. Removing the dir first
// guarantees a freshly built db.
//
// We deliberately do NOT assert the raw `DbIdentity` changed here. On unix
// the identity is the inode, and the OS is free to RECYCLE the just-deleted
// db's inode for the rebuilt file (observed under `cargo llvm-cov`: both =
// 1343671), so an `assert_ne!(id_before, id_after)` gambles on an OS
// implementation detail that delete+rebuild does not guarantee. That inode
// assertion was the sole flake source and is redundant with #925's real
// contract, which is the BEHAVIORAL outcome asserted below and does not
// depend on inode luck: the SAME cached server reopens its engine and serves
// the NEW on-disk content. Serving `betaSymbol` (not `alphaSymbol`, not an
// open error) is itself the proof the engine reopened against the fresh db.
server.close_cached_handles();
fs::remove_dir_all(project.path().join(".codegraph")).unwrap();
index_into(
project.path(),
&[("src/b.ts", "export function betaSymbol() {}\n")],
);
let id_after = db_identity(project.path());
assert_ne!(
id_before, id_after,
"replace must change DbIdentity (inode on unix; \
len/creation_time/header-sig fold on non-unix — the rebuild changes \
the SQLite page count + schema cookie deterministically)"
);

// Then: the SAME server's SAME tool call must REOPEN the engine — without the
// #925 fix the cached engine keeps serving inode A and `reopen_count` never
// advances (this delta is the deterministic RED probe).
// #925 fix the cached engine keeps serving the old db and `reopen_count`
// never advances (this delta is the deterministic RED probe).
let after = search(&mut server, project.path(), "betaSymbol");
let reopens_after = reopen_count();
assert_eq!(
Expand Down
Loading