bug: `gbrain sync --strategy code` silently imports zero code files on first-sync (full path)

[lanceretter]

Summary

gbrain sync --strategy code --source <id> reports "X pages imported" and advances last_commit on the named source, but zero code files are actually persisted to the database. The flag is silently dropped on the full-sync code path. Reproduces consistently across multiple repos and multiple syncs.

Reproducer

# Fresh source, fresh sync
gbrain sources add gstack-code-myrepo --path /path/to/repo --federated
gbrain sync --strategy code --source gstack-code-myrepo
# → "216 pages imported, 1323 chunks created" (looks great)

# Verify
psql "$DATABASE_URL" -c "SELECT COUNT(*) FROM pages WHERE source_id='gstack-code-myrepo' AND page_kind='code';"
# → 0
psql "$DATABASE_URL" -c "SELECT COUNT(*) FROM pages WHERE page_kind='code';"
# → 0 across the entire brain
gbrain code-def someSymbol
# → { "count": 0, "results": [] }

I have three sources on the same brain (gstack-code-rv-helper, gstack-code-trashtastic-helix, gstack-code-conquest-lpr-ccfd94) all synced via this command path over the past 3 days. All three show last_commit and last_sync_at set, yet pages WHERE source_id=<X> AND page_kind='code' returns 0 for every one of them. pages WHERE page_kind='code' returns 0 globally.

Root cause

gbrain sync --strategy code --source X on a source with last_commit IS NULL:

1. Hits the full-sync path: commands/sync.ts:performFullSync.
2. Calls runImport(engine, [repoPath], { commit: headCommit }) at commands/sync.ts:892. Strategy is not plumbed through.
3. runImport in commands/import.ts:31 calls collectMarkdownFiles(dir) and logs "Found ${allFiles.length} markdown files". It walks markdown only — by name.
4. Result: 216 markdown files walked → all match existing default-source slugs by content hash → all marked skipped (unchanged) → zero code files ever touch the importCodeFile path.
5. last_commit gets set as if synced. Next incremental run via the diff path (sync.ts:467) finds no diff and does nothing. Code files remain forever absent.

The dry-run path at commands/sync.ts:861 has the same shape — isSyncable(rel) without the strategy arg — but the prod path is the bigger break because it actually writes the misleading last_commit.

The incremental path at line 471 does thread strategy via syncOpts. It's only the full-sync (first-sync) path that's broken.

gbrain reindex-code is not a workaround: it walks pages WHERE p.type='code', of which there are zero post-sync, so it reports "No code pages to reindex" and exits clean.

Verbose-mode evidence

$ gbrain sync --strategy code --source gstack-code-conquest-lpr-ccfd94 --verbose
Found 216 markdown files                          ← walker is markdown-only
[import.files] 216/216 (100%) imported=0 skipped=216 errors=0
Import complete (2.2s):
  0 pages imported
  216 pages skipped (216 unchanged, 0 errors)     ← all 216 match default-source slugs
  0 chunks created
First sync complete. Checkpoint: 6f462d2f         ← last_commit advances anyway

(File counts: this repo has 223 .md and 1109 .ts/.tsx/.py/.js. The "216 markdown files" matches the .md count after skipFiles exclusion of README.md/schema.md/etc., not the code count.)

Suggested fix locations

The cleanest shape is probably to plumb strategy through runImport so the walker can branch:

• commands/import.ts — accept strategy in opts; replace collectMarkdownFiles with a strategy-aware walker (or call a new collectCodeFiles for strategy='code').
• commands/sync.ts:892 — pass opts.strategy into runImport.
• commands/sync.ts:861 — pass { strategy: opts.strategy } into isSyncable (dry-run cosmetic fix).

Alternatively keep runImport markdown-only and split performFullSync so strategy='code' routes to a new performFullCodeSync that walks code files via isCodeFilePath and calls importCodeFile directly.

Either fix is bigger than a one-liner — happy to send a PR if you want, but flagging as an issue first since the right shape is your call.

Companion to recent PRs

This was discovered during the v0.26.6 → v0.30.0 upgrade dig in #740 + #741. Same upgrade exercise revealed three fixed bugs (deps, UNSAFE_TRANSACTION, missing connect) plus the bootstrap gap. This is the fourth real bug from that exercise. Not blocking the upgrade — markdown semantic search works fine — but gbrain code-def/code-refs/code-callers is unusable for any user who has gotten this far without realizing their code source has zero pages.

🤖 Reported by Claude Code

[garrytan]

Closing as a duplicate of #767 (sync --strategy code dropped on first sync).

Thanks for the report — please follow #767 for updates. If your case differs from the canonical one in a way that warrants its own issue, please reopen with details.