userBlocks grew (P4 handle teardown) — bump baseline. flaky-relays grid-render
timeout 15s->30s so dead-relay connect timeouts don't cause a false-negative
(ux-harness with a healthy relay renders in ~1.2s; the harness now reliably
passes 4/4).
Stands up several mock relays — some alive (seeded), some dead (unreachable
ports) — points the app at all of them, logs in via the fake nip-07 extension,
and asserts the refactor invariants under reconnect churn:
1. grid renders from alive relays
2. subscribes on <= the capped core (not all configured relays)
3. no steady-state REQ storm despite dead relays (was 1000+/s)
4. report subs are batched (one kind-1984 filter with many #e, no per-id)
All four PASS on the current build. This is the realism the single perfect mock
relay lacked — the reason earlier 'fixed' verdicts didn't hold in production.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
Headless Playwright probes that reproduce the real login path via a fake
window.nostr extension (which loginWithNsec bypasses) and measure
steady-state extension-call and relay-REQ rates:
- nip07-heat.mjs, req-capture.mjs, ws-lifecycle.mjs, loopc-capture.mjs, etc.
Plus js/devReqMonitor.js: a localhost-only WebSocket REQ-rate monitor that
warns in-console when relay traffic spikes, naming the storming kinds — used
to capture the real-environment storm the mock-relay harness could not see.
Removable once the refactor lands.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
wh-repro.mjs uses a SEPARATE browser context per session (seed vs cold), so the
persisted decrypt cache doesn't warm between runs, and captures the page-side
stack of every window.nostr.nip44.decrypt call.
It root-caused the watch-history cold-load cost: history is bucketed by month
(one encrypted kind-30079 snapshot per month) and the load decrypts EVERY month
via the extension — 600 items / 18 months => 18 serialized decrypt calls. So the
cost is ~1 nip-07 decrypt per month of history; a multi-year history is the ~20s
cold load. (The earlier "~180" figure was a shared-context harness artifact.)
Co-Authored-By: Claude Opus 4.8 <[email protected]>
Consolidates the persisted-plaintext-decrypt-cache pattern (already used by watch
history) into a shared factory, persistedPlaintextCache.js, and adds a DM cache
on top of it. watchHistoryDecryptCache now delegates to the factory (same public
API; tests unchanged).
decryptDirectMessageEvent now also consults/writes a persisted cache keyed by the
immutable DM event id (JSON of the decrypt result; flushed once per
listDirectMessages batch; cleared on logout alongside the in-memory cache). DM
events never change, so a reload re-decrypts only messages it has not seen and
skips the (serialized, slow) nip-07 extension for the rest.
Harness (dm-repro, now cold/warm): 20 DMs -> COLD nip04.decrypt 50, WARM 0 —
reloads cost zero extension decrypt calls. Also excludes the actor's own key from
NIP-04 remote candidates (decrypting against self is always wasted). DM unit +
integration and watch-history cache tests green.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
The DM-repro harness (fake nip-07 extension) showed legacy DM history made ~12
extension decrypt round-trips PER message (20 DMs -> 241 calls: nip44.decrypt
121 + nip04.decrypt 120). At a real ~100ms extension latency that's ~24s, which
is why DM history "looked broken" — it was just catastrophically slow.
Two causes in decryptLegacyDm:
- It raced EVERY (decryptor x remote-pubkey) combination in parallel. That looked
like a latency win, but the nip-07 extension serializes calls, so racing N
combinations just queued N slow round-trips per message.
- orderDecryptors ranked nip44 first (by candidate priority), so every kind-4
(always NIP-04) message burned a failing nip44 extension call first.
Fix: order nip04 first for legacy DMs (preferLegacy) and try candidates
SEQUENTIALLY, stopping at the first success. With the right scheme and remote
party tried first, it's ~1-3 calls/message instead of ~12. Harness after: 20 DMs
-> 60 calls, zero nip44 attempts (was 241). DM unit + integration tests green.
Adds scripts/perf/dm-repro.mjs (seeds kind-4 DMs, logs in via the fake extension,
calls listDirectMessages, reports decrypted count + extension call counts).
Co-Authored-By: Claude Opus 4.8 <[email protected]>
Extends the fake-nip-07-extension harness to a multi-session flow that publishes
a watch history via the app (real chunking + encryption through the fake
extension), then reloads to measure the decrypt-cache effect by counting
nip44.decrypt extension round-trips:
seed : publishes the snapshot (nip44.encrypt per chunk)
COLD : first reload — decrypts chunks via the extension (~180 calls @ 300 items)
WARM : second reload — decrypt cache hits, ~2 calls
Validates the watch-history decrypt cache (watchHistoryDecryptCache.js): the
warm reload skips ~180 serialized extension round-trips (~9s @ 50ms, more at
real latency). Uses context-level bindings so localStorage (and the cache)
persists across reloads; bypasses feed signature verification of seeded data via
the test-only videoEventVerifier hook so the feed renders deterministically.
Also surfaced: cold load still pays the full decrypt cost, and the decrypt path
makes far more extension calls than there are chunks — an inefficiency to chase.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
The UX harness (fake nip-07 extension) revealed that the off-thread verification
could blank the feed: if nostr-tools is slow/unable to load inside the verify
worker, the worker hangs, each batch waits the full 10s timeout, and the
main-thread fallback could return an empty set (fail-closed) — dropping every
event so no videos ever render, silently (devLogger is suppressed in prod).
Harden it so verification can never stall or blank the read feed:
- Worker timeout 10s -> 2.5s, and disable the worker for the session after one
timeout/error so we never pay the timeout on every batch.
- Bound the main-thread nostr-tools wait (Promise.race ~2s) so a slow/failed
load can't hang the fallback.
- Fail OPEN when the verifier is genuinely unavailable: pass events through
unverified rather than dropping them (the pool already does not verify, so this
does not regress the prior posture). Real verification still runs whenever
nostr-tools is available.
Adds a regression test (hung worker -> resolves via fallback, worker disabled).
Also updates the harness to seed mode:"live" videos + a NIP-65 relay list for
mock-relay isolation.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
Lets the agent observe the real app the way a user does: runs bitvid in headless
Chromium against an in-process mock relay, injects a fake window.nostr extension
backed by Node-side nostr-tools crypto (via page.exposeFunction) with a
serialized ~100ms latency queue to model a real extension. Drives the real
nip-07 login flow, isolates to the mock relay via a seeded NIP-65 list, and
captures console logs, per-method extension call counts, feed-render timing, and
screenshots.
Phase 1 (this commit) verifies fake-extension login + relay isolation + metrics.
Already surfaced a stuck "Fetching recent videos…" state to investigate. Next
phases: seed encrypted watch-history chunks and DMs to reproduce/verify the
decrypt-cache wins and the DM-history bug, with before/after extension-call
counts.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
Standalone Playwright/CDP profilers used to find the main-thread bottlenecks:
- profile-live.mjs: cold-load CPU profile + long-task report against a URL.
- diagnose-feed.mjs: dumps feed DOM/app-state/console to explain empty feeds.
- profile-seeded.mjs: seeds an in-process mock relay, drives login/tab-switch/
modal-open in test mode, and attributes secp256k1 self-time to its nearest
app-level caller.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
thePR0M3TH3AN