Move note language computation off the main thread

NLLanguageRecognizer.processString() is an expensive NLP operation that
was moved to @MainActor in 5058fb33, causing UI jank when scrolling.

This moves language detection back to the async preload path where it
runs off the main thread. get_preload_plan is synchronous again and
defers language computation to preload_event. Translations still happen
on the first preload pass since preload_event now checks .havent_tried
directly rather than relying on the load_translations flag from the plan.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
William Casarin
2026-02-10 10:16:25 -08:00
parent 2d5460b654
commit be6b0e2702
3 changed files with 24 additions and 39 deletions

View File

@@ -434,7 +434,7 @@ struct NoteContentView: View {
}
// always reload artifacts on load
let plan = get_preload_plan(ndb: damus_state.ndb, evcache: damus_state.events, ev: event, our_keypair: damus_state.keypair, settings: damus_state.settings)
let plan = get_preload_plan(evcache: damus_state.events, ev: event, our_keypair: damus_state.keypair, settings: damus_state.settings)
// TODO: make this cleaner
Task {

View File

@@ -311,24 +311,21 @@ func load_preview(artifacts: NoteArtifactsSeparated) async -> Preview? {
return Preview(meta: meta)
}
@MainActor
private func compute_note_language(ndb: Ndb, ev: NostrEvent, keypair: Keypair) -> String? {
return ev.note_language(ndb: ndb, keypair)
private func compute_note_language(ev: NostrEvent, keypair: Keypair) -> String? {
assert(!Thread.isMainThread, "note_language must not be run on the main thread.")
return ev.note_language(keypair)
}
@MainActor
func get_preload_plan(ndb: Ndb, evcache: EventCache, ev: NostrEvent, our_keypair: Keypair, settings: UserSettingsStore) -> PreloadPlan? {
func get_preload_plan(evcache: EventCache, ev: NostrEvent, our_keypair: Keypair, settings: UserSettingsStore) -> PreloadPlan? {
let cache = evcache.get_cache_data(ev.id)
let load_artifacts = cache.artifacts.should_preload
if load_artifacts {
cache.artifacts_model.state = .loading
}
// Cached event might not have the note language determined yet, so determine the language here before figuring out if translations should be preloaded.
let note_lang = cache.translations_model.note_language ?? compute_note_language(ndb: ndb, ev: ev, keypair: our_keypair)
if cache.translations_model.note_language == nil {
cache.translations_model.note_language = note_lang
}
// Use cached language if available; actual computation deferred to async preload_event()
let note_lang = cache.translations_model.note_language
let needs_language = note_lang == nil && settings.auto_translate
let load_translations = should_preload_translation(event: ev, our_keypair: our_keypair, current_status: cache.translations, settings: settings, note_lang: note_lang)
if load_translations {
@@ -352,7 +349,7 @@ func get_preload_plan(ndb: Ndb, evcache: EventCache, ev: NostrEvent, our_keypair
cache.preview_model.state = .loading
}
if !load_artifacts && !load_translations && !load_preview && load_urls.count == 0 {
if !load_artifacts && !load_translations && !load_preview && !needs_language && load_urls.count == 0 {
return nil
}
@@ -436,15 +433,15 @@ func preload_event(plan: PreloadPlan, state: DamusState) async {
if let cached_language = plan.data.translations_model.note_language {
note_language = cached_language
} else {
note_language = await compute_note_language(ndb: state.ndb, ev: plan.event, keypair: our_keypair)
note_language = compute_note_language(ev: plan.event, keypair: our_keypair)
}
var translations: TranslateStatus? = nil
// We have to recheck should_translate here now that we have note_language
if plan.load_translations,
let note_language,
// Check if we should translate now that we have note_language (which may have just been computed)
if let note_language,
can_and_should_translate(event: plan.event, our_keypair: our_keypair, settings: settings, note_lang: note_language),
settings.auto_translate
settings.auto_translate,
plan.data.translations == .havent_tried
{
translations = await translate_note(profiles: profiles, keypair: our_keypair, event: plan.event, settings: settings, note_lang: note_language, purple: state.purple)
}
@@ -468,27 +465,15 @@ func preload_events(state: DamusState, events: [NostrEvent]) {
let our_keypair = state.keypair
let settings = state.settings
let plans = events.compactMap { ev in
get_preload_plan(evcache: event_cache, ev: ev, our_keypair: our_keypair, settings: settings)
}
if plans.count == 0 {
return
}
Task {
let plans = await withTaskGroup(of: PreloadPlan?.self) { group in
for ev in events {
group.addTask {
await get_preload_plan(ndb: state.ndb, evcache: event_cache, ev: ev, our_keypair: our_keypair, settings: settings)
}
}
var results: [PreloadPlan] = []
for await plan in group {
if let plan {
results.append(plan)
}
}
return results
}
if plans.count == 0 {
return
}
for plan in plans {
await preload_event(plan: plan, state: state)
}

View File

@@ -568,8 +568,8 @@ extension NdbNote {
return thread_reply() != nil
}
@MainActor
func note_language(ndb: Ndb, _ keypair: Keypair) -> String? {
func note_language(_ keypair: Keypair) -> String? {
assert(!Thread.isMainThread, "This function must not be run on the main thread.")
// Rely on Apple's NLLanguageRecognizer to tell us which language it thinks the note is in
// and filter on only the text portions of the content as URLs and hashtags confuse the language recognizer.