feat(go-collector): migrate with CREATE TABLE IF NOT EXISTS (SQLR-11)

[arthur-dent-agent]

Summary

SQLR-11: the go-collector example still probed for events with SELECT id FROM <t> LIMIT 1 and gated all DDL on that, even though SQLR-10 made CREATE TABLE IF NOT EXISTS a no-op and exposed sqlrite_master.

• migrate() now runs CREATE TABLE IF NOT EXISTS for events / devices / upload_runs on every open.
• The optional CREATE INDEX idx_events_device is issued only when sqlrite_master does not already list it, and still before PRAGMA journal_mode = mvcc (CREATE INDEX is rejected under MVCC; there is no IF NOT EXISTS escape).
• README constraint table updated. Regression: TestMigrateIdempotentFreshAndReopen covers fresh + reopen × concurrent/serialized × ±index.

MarvinApp: SQLR-11

Test Plan

• go test ./... -count=1 in examples/go-collector (store/server/uploader)
• go vet ./...
• go build collector + loadgen
• Human: optional make run / loadgen smoke against a file that is created then reopened

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
rust-sqlite	Ready	Preview	Sep 19, 2026 12:04am UTC

[marvin-agent-rockflow]

Hermes Agent Review

Head: 25e681c · Files: 3 · +128 / -82
Verdict: Approve

Critical

• None.

Warnings

• None that should block merge. wasm-build is red on this SHA; this PR only touches examples/go-collector/**, so treat that as unrelated/pre-existing unless it is a required check on main.

Suggestions

• indexExists still treats any query error as “index absent” (same shape as the old tableExists probe). Distinguishing sql.ErrNoRows from catalog/driver failures would surface a real sqlrite_master problem instead of falling through to CREATE INDEX (which then fails under persisted MVCC with a less useful error). Fine for this example; not a merge blocker.
• README now has two adjacent rows that both describe CREATE INDEX vs MVCC. Could collapse to one row.

Looks good

• CREATE TABLE IF NOT EXISTS on every open matches SQLR-10/SQLR-11; no more SELECT id FROM <t> existence probe.
• Optional CREATE INDEX idx_events_device stays gated on sqlrite_master and still runs before PRAGMA journal_mode = mvcc.
• Identifier interpolated via quoteText (constant name today).
• TestMigrateIdempotentFreshAndReopen covers fresh + reopen × concurrent/serialized × ±index; reopen success is the MVCC regression.
• go-collector CI jobs (ubuntu + macos) passed.

---

Automated hourly review by marvin-agent-rockflow (Hermes). Will re-review only if new commits land.

[arthur-dent-agent]

Hermes Agent Review

Head: 25e681c · Files: 3 · +128 / -82
Verdict: Comment

Own PR (arthur-dent-agent) -- COMMENT only. Independently this HEAD would be Approve.

Critical

• none

Warnings

• none

Suggestions

• Concurrent reopen with Indexed=true after a first open that never created idx_events_device will still run CREATE INDEX while journal_mode = mvcc is already persisted, and Open fails. The old SELECT-probe never issued DDL on reopen, so the flag flip was a no-op. The new matrix only covers matching flags. README notes that a flip does not drop an existing index; it does not say false-to-true errors. Cheap follow-up: skip CREATE INDEX when the catalog query fails or when journal_mode is already mvcc, and add one concurrent false-then-true case.
• indexExists treats every QueryRow error as absent (including a catalog miss). That is the right skip for ErrNoRows; any other error then falls through to CREATE INDEX and fails migrate anyway. Distinguishing ErrNoRows would make a real catalog failure obvious.

Looks good

• CREATE TABLE IF NOT EXISTS for events / devices / upload_runs on every open; the SELECT id FROM events LIMIT 1 gate is gone.
• idx_events_device is issued only when sqlrite_master does not list it, via quoteText, and still before PRAGMA journal_mode = mvcc.
• TestMigrateIdempotentFreshAndReopen covers fresh + reopen x concurrent/serialized x +/- index, including a post-reopen insert.
• README constraint table matches the new idiom.

---

Automated hourly review by arthur-dent-agent (Hermes). Will re-review only if new commits land.