Repository navigation
v0.2.0 — release-prep hardening
Shipped by the 2026-09-11 scheduled maintenance cycle. Anchored to [[github-portfolio-review-2026-09-02]] §3.1 (hoot tier-1 polish) + issue #49 (release prep).
Bug fixes
- dedupe defaultRelays (#61) —
wss://nostr.winewas listed twice, causing every post to double-publish to the same relay and double-count in the success tally. Newinit()dedupes the slice at process start. Pinned byTestDefaultRelaysAreUnique. - generateProfileID error path (#61) —
crypto/rand.Readerror was discarded, falling back to the hex of an all-zero buffer on rare entropy failure. Every such ID collided onprofiles.jsonand silently overwrote other profiles' encrypted keys. Now falls back to a nanosecond-timestamp ID, unique within a process. Pinned byTestGenerateProfileIDIsUnique(1000-iteration uniqueness check). - NWC payment honesty (#61 + #63) —
hoot -tippreviously exited 0 after publishing the NWC request without ever listening for the wallet's kind 23195 response. Lightning is irreversible; a payment that looks successful but isn't is a real-money bug. The fix:- 30 s context timeout so a hung wallet can't freeze
-tipforever (wascontext.Background()— no timeout). event.Sign()error now checked (previously discarded).- Post-publish message is honest: "relay accepted the request, not that the payment was made; check your wallet."
- The kind-23195 subscribe-and-listen loop is its own design discussion (timeout policy, disconnected-relay handling) — flagged for a future PR.
- 30 s context timeout so a hung wallet can't freeze
- publish path error surfacing (#62) —
main()usedlog.Fatalfon publish failure, which skips every pendingdefer(notablyeventCache.Close). Replaced with explicitlog.Printf+os.Exit(1)so the deferred cleanup runs. Same shape as #37/#38 refactors; this closes the lastlog.Fatalfin the publish path.
Chore
- untrack built binaries (#64) —
hoot.exe(13 MB) andhoot-test(13 MB) were tracked inmaster, bloating every clone + PR diff by 26 MB for files that should never be in source control.git rm --cachedkeeps them on existing clones but stops syncing. New.gitignorecovers all common Go build outputs, test artifacts, Go workspace files, IDE scratch, vendor caches, and.envoverrides. No history rewrite — past commits still contain the blobs (intentional;git filter-repopurge is a coordinate-with-clones task, flagged separately).
Upgrade notes
No breaking API changes. Tests pass locally; the repo still has no CI workflow (adding .github/workflows/ci.yml running go test -race ./... is the obvious next step and will be proposed separately).
What's next
- Release infrastructure:
.goreleaser.yml+ release workflow (issue #49). - LICENSE file (MIT, Go convention).
- Topics + homepage URL on the repo (
hoot-docs.vercel.app). - Lean / Ethereum / Nostr community cross-posts.