Files
bitvid/docs/KNOWN_BUGS.md
T

88 lines
4.1 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# Known Bugs (deferred)
Running list of confirmed bugs/limitations found during the data-flow refactor
(see [architecture-refactor.md](./architecture-refactor.md)) that we chose to
fix later rather than block the current phase. Distinct from the *planned*
refactor work in that doc — these are defects.
> Format: **[severity]** short title — where — what — suggested fix / phase.
---
## Open
### 1. Logged-out: video grid stays empty until a manual refresh
- **Severity:** medium (first-load UX, logged-out).
- **Where:** initial feed render path (`js/index.js` disclaimer flow →
`viewManager` → `feedCoordinator`); rendering is a side-effect of event
ordering.
- **Symptom:** logged out, after dismissing the disclaimer the grid doesn't
populate until the page is refreshed.
- **Root cause (suspected):** the grid paints as a side-effect of a data event
that can fire before the grid container is mounted (render coupled to event
ordering, not to state).
- **Fix:** P5 — unidirectional flow: render from AppState whenever it changes,
decoupled from event ordering. Could not reproduce headlessly (real-relay
feed returned nothing in the sandbox); reproduce in a real browser first.
### 2. Residual full-feed reload loop on list-loaded signals
- **Severity:** low–medium (bounded by the 2s throttle, but wasteful — re-fetches
lists ~every 2s while signals keep arriving).
- **Where:** `js/app/authSessionCoordinator.js` `handleBlocksLoaded` (and peers)
call `onVideosShouldRefresh` **unconditionally**; the forced reload re-fetches
the lists, which re-emit "loaded".
- **Fix:** P5 — only trigger a reload when the underlying list data actually
changed (emit-on-change at the list-loaded sources + collapse the refresh
triggers into the Coordinator).
### 3. Direct-pool-access lint still misses MULTI-LINE access
- **Severity:** low (tooling gap — weakens the L1 chokepoint guarantee).
- **Status:** partially fixed — the regex now catches bare `pool.sub(` /
`pool.list(` (re-baselined to 22 files). Still MISSES access split across
lines, e.g. `js/services/profileMetadataService.js` writes `nostr.pool`
newline `.list(` — so profiles is an unflagged fetch source.
- **Fix:** normalize whitespace before matching (collapse `\.pool\s*\n\s*\.list`)
or move to an AST/`grep -A1` check. Add profileMetadataService (and any other
multi-line offenders) to the allowlist once detected.
### 4. Interim relay cap is static, not liveness-ranked
- **Severity:** low (acceptable interim).
- **Where:** `js/nostr/client.js` `capSubscribeRelays` — keeps the first 6
relays (known-good defaults prioritized) but does NOT re-rank by live health.
If a user's first relays are dead, reads degrade until other paths recover.
- **Fix:** P1's deferred `RelayHealth` — rank the subscribe set by rolling
liveness and re-resolve on health changes.
---
## Pre-existing test failures (predate this refactor)
Confirmed failing at the pre-session baseline (`4bd9503c`) — not caused by the
refactor, but should be triaged. Not in `KNOWN_ISSUES.md`.
### 5. `tests/nostr-delete-flow.test.mjs`
- "deleted events should not trigger subscription callbacks" — a deleted event
reaches the feed callback (buffer delete-filter or dedupe). Investigate the
`VideoEventBuffer` delete handling.
### 6. `tests/view-counter.test.mjs`
- "view event should append the session tag when requested" — view-event
**publishing** tag composition (unrelated to the P3 counting refactor).
### 7. `tests/nostr-count-fallback.test.mjs`
- Fails at baseline; NIP-45 COUNT fallback path. Triage.
### 8. `tests/nostr/nip07Permissions.test.js`
- `writeStoredNip07Permissions` / `clearStoredNip07Permissions` fail under Node
(no real `localStorage`). Likely environmental — provide a localStorage shim
(see `tests/test-helpers/setup-localstorage.mjs`) or guard the test.
---
## Perf follow-ups (not bugs, but tracked)
- **One-time login enrichment burst** (~75 REQ frames): batched and O(relays),
settles to idle, but could be trimmed by hydrating only on-screen cards.
- **Remove `js/devReqMonitor.js` import + `scripts/perf/*` diagnostics** in P6
cleanup once the refactor lands.