Vendor .golangci.yml byte-identical from sneak/prompts at cc440118 and
move the Dockerfile lint phase to golangci-lint v2.14.0 by the digest
REPO_POLICIES.md names. Fix the code to that config with flags, help
text, output files, SQL and the order of steps unchanged; long
functions are split into named steps.
Judgement call: the extraction transaction is now rolled back on every
early return; the old deferred rollback missed most failures and could
dereference a nil transaction.
Wording change: the reversed-range error reads "--from is after --to
(--from X, --to Y)", since err113 requires a wrapped sentinel.
Ten //nolint directives (gosec, mnd, unqueryvet), each with its reason.
Model: opus-5-5
Handle every error return flagged by golangci-lint's errcheck rather
than discarding it:
- run.go: add a cleanup() helper that logs a warning (and ignores
ErrNotExist) when removing a temp file fails, so leaked scratch files
are surfaced; use it for all best-effort removals.
- copy.go / extract.go: log a warning on deferred Close() failures for
the source file, destination DB, and result-set rows.
- extract.go: on the rollback path, ignore the benign sql.ErrTxDone
(already committed) and log any other rollback failure.
- verify.go: add killCat() which ignores os.ErrProcessDone (zstdcat
already exited via SIGPIPE) and logs any unexpected kill failure.
make check is clean (0 lint issues, tests pass).
Add a smoke test that references the package's exported surface so go
test fails if it stops compiling. Run go mod tidy to promote cobra to
a direct dependency.
Add a detailed README, WTFPL LICENSE, and build/CI tooling modeled on
the vaultik repo (Makefile, multi-stage digest-pinned Dockerfile,
.gitea/workflows/check.yml). Bump Go to 1.26.4 and pin golangci-lint
to v2.12.2. gofmt existing sources so the new fmt-check gate passes.
- Use tx.QueryRow instead of db.QueryRow within transaction for media table check
- Switch to WAL mode instead of OFF for better concurrency handling
- Add busy_timeout pragma to wait up to 5 seconds if database is locked
- Clear transaction variable after commit to prevent defer rollback issues
- Remove parallel.go entirely
- Process days strictly sequentially to avoid SQLITE_BUSY errors
- Complete each day fully (extract, dump, compress, verify, write) before starting next
- Clean up temp files immediately after each day to save disk space
- SQLite cannot handle multiple concurrent ATTACH operations on same database
- Increase copy buffer from 4MB to 256MB for NVMe-speed transfers
- Add Linux-specific fadvise hints for sequential reads and prefetching
- Pre-allocate destination file to avoid fragmentation
- Add throughput monitoring to track copy performance
- Platform-specific build tags for Linux optimizations
Disable crash-safety pragmas (synchronous=OFF, journal_mode=MEMORY),
use a 200MB page cache, disable foreign keys to avoid per-row CASCADE
overhead, bulk-delete junction tables explicitly, and wrap all deletes
in a single transaction.
clawbot
sneak