Skip to content

Speakr is a personal, self-hosted web application designed for transcribing audio recordings

Python

Commits on master

Detail header metadata row: new .chip primitive matching sidebar style + flatten so items wrap individually — addresses three things the user pointed out; (1) detail-header chips were rounded-full + px-2 but sidebar chips were rounded-md + px-1.5 with 9px icons — two different styles for the same data; (2) the metadata row's file-metadata items (date created / file size / folder path / etc.) lived in their own nested flex-wrap div which forced them ALL to wrap together as a block even when individual entries could have stayed on the first row; (3) general crudeness — the dot-badges (incoming-share / group / outgoing-share / public-link) used inconsistent w-5 h-5 boxes that didn't share a primitive; CHANGES; (A) primitives.css new .chip primitive — display: inline-flex, gap: 4px, padding: 1px var(--sp-2), border-radius: var(--radius-sm), font-size: var(--fs-label), font-weight: 600, line-height: 1.55, white-space: nowrap, max-width: 14rem for truncation, plus .chip > i auto-sized to 0.78em and .chip__label that handles overflow with ellipsis; .chip--clickable adds the hover/opacity affordance + active:scale; (B) new .chip-dot primitive for the small round icon-only badges (18px circle, semantic background via inline class or bg-utility); (C) desktop-header.html metadata row completely flattened — every entry is a direct sibling of the outer flex-wrap container (was: folder/tags/dots nested in one div, file-metadata nested in another div); each item now wraps independently so the wrap point honours the actual width of each item; chips switched from inline rounded-full classes to the .chip primitive, dot badges switched to .chip-dot; icons in non-chip items pinned at text-[11px] for visual rhythm; (D) sidebar.html ported its inline chip classes to .chip chip--clickable so both surfaces share one source of truth — future style tweaks to chips only need to touch primitives.css
Murtaza Nasir committed 5 months ago
1293287
Revert ASR editor to virtual scrolling; keep speaker modal on content-visibility + bounded hydration — user's correct observation: the ASR editor is spreadsheet-like (uniform single-line inputs + a textarea capped at max-h-20: 80px), so row heights are bounded and the fixed-itemHeight virtual scroll is the right tool; user explicitly said 'earlier I wasn't having any scrolling issues with this EditASR transcription model'; the issues were always with the speaker modal (variable-height segments); my content-visibility approach was overkill for the editor and was caught producing CLS 0.50–3.02 in the latest metrics; CHANGES; (1) styles.css: removed 'table-layout: fixed' and the 'content-visibility: auto; contain-intrinsic-size' rule from .asr-editor-table; back to the original simple style; (2) asr-editor-modal.html: restored the virtual-scroll structure — asrEditorSpacerBefore / asrEditorVisibleSegments v-for with segment._originalIndex / asrEditorSpacerAfter, plus @scroll='onAsrEditorScroll($event)' on the container; removed the hydrated/placeholder template split; rows are now mounted by virtual scroll's window (~30 rows total in DOM); (3) app.modular.js: utils.scrollAsrEditorToIndex back to asrEditorVirtualScroll.scrollToIndex(index-2, 'auto') so highlight-jump still leaves 2 rows of clearance under the sticky table header; utils.resetAsrEditorScroll back to asrEditorVirtualScroll.reset(); removed the ASR-editor _makeRowHydrator wiring and the showAsrEditorModal watcher; hydrateAsrEditorRow kept as a no-op for any leftover template refs (none now, but defensive); kept the _makeRowHydrator factory + showSpeakerModal wiring — speaker modal continues to use content-visibility + bounded hydration which is correct for its variable-height segments
Murtaza Nasir committed 5 months ago
d664c09
Bounded lazy-hydration for ASR editor + speaker modal rows — addresses the 1408 ms INP / 142k DOM nodes / 71% CPU regressions the user hit at 2500 segments; content-visibility skipped the BROWSER's paint cost but Vue still mounted every row's reactive bindings; THE FIX is a generic _makeRowHydrator factory + IntersectionObserver that promotes a row to its full interactive form (inputs, buttons, dropdown teleport, edit handlers) ONLY when it's near the viewport (within 600 px above and below), and DEMOTES the row back to a lightweight static placeholder when it scrolls out — bounded membership so the hydrated set stays around one viewport's worth (~50-100 rows) regardless of transcript length; without the dehydration step the user's scroll top→bottom would gradually leave every row hydrated and re-introduce the regression; rows containing document.activeElement are never dehydrated so the user typing into an input mid-scroll never gets it yanked away; ASR editor placeholder is 4 read-only <td>s rendering segment fields, hydrated version has the 4 v-models + dropdown teleport + 3 action buttons; speaker-modal placeholder is the speaker tag + text spans (no hover buttons, no edit dropdown), hydrated version has the hover edit buttons + change-speaker dropdown; both modals' v-memo dependency lists now include the hydratedRows.has(idx) check so promotion/demotion correctly triggers re-render; speaker modal also keeps a row hydrated whenever editingSpeakerIndex === idx so an open dropdown survives a scroll; one shared factory keeps both modals' implementations identical (different containerRef + hydratedRows) for clarity
Murtaza Nasir committed 5 months ago
4787ec8
ASR Editor modal: drop virtual scroll, use content-visibility + v-memo + scrollIntoView (same fix as Speaker modal) — addresses user-reported CLS 1.60 (poor) with 20 layout shifts in the worst cluster during scroll; root cause was the same broken pattern as the speaker modal — fixed itemHeight: 44 estimated row height against variable real heights (textareas auto-resize via autoResizeTextarea, so a row with one line of sentence is ~44 px but a row with 5 lines is ~120 px), so as the user scrolls and different real heights enter the visible range, the table's total height changes (spacerBefore + visible rows real + spacerAfter is not stable), the scrollbar position recomputes, layout shifts every frame, CLS accumulates; THE FIX (3 parts); (1) styles.css gets two additions on .asr-editor-table; table-layout: fixed (required for content-visibility on <tr> — without it the browser still has to lay out every row to compute column widths, which defeats the off-screen skip) and 'content-visibility: auto; contain-intrinsic-size: auto 44px' on tbody > tr (browser skips layout + paint for off-screen rows so they don't pay the layout cost but they DO contribute their 44 px placeholder to scrollHeight, keeping the scrollbar position stable); (2) asr-editor-modal.html template drops the asrEditorSpacerBefore/After tr spacers and the asrEditorVisibleSegments iteration, iterates the FULL editingSegments array with (segment, idx), replaces segment._originalIndex with idx throughout, and adds v-memo per row so typing in one row's input/textarea doesn't fan out a re-render to the other ~4500 rows; @scroll handler no longer calls onAsrEditorScroll (no virtual range to update), only closeAllSpeakerSuggestions (keeps the user-experience parity); (3) app.modular.js utils.scrollAsrEditorToIndex switches from asrEditorVirtualScroll.scrollToIndex(index-2, 'auto') to container.querySelector([data-segment-index]) + scrollIntoView({block: 'center'}) — block: 'center' centres the row in the viewport so it lands clear of the sticky table header without needing to know the header's height; resetAsrEditorScroll just sets scrollTop=0; columns are still explicitly sized (w-40 / w-20 / w-24 + flexible sentence) so table-layout: fixed produces the same visual result as the previous auto layout; SW cache bumped to v10
Murtaza Nasir committed 5 months ago
2781259
Speaker modal rewrite: drop virtual scroll, use content-visibility + v-memo + scrollIntoView — full design-system rebuild that fixes the variable-height scroll bugs without regressing #157 perf; THE PROBLEM; the previous virtual scroller used a constant 52px itemHeight estimate but real segments range from 40px (Yes.) to 600+px (long podcast paragraphs), so scrollToIndex(N) = scrollTop = N×52 landed at wrong positions, the scrollbar visually jumped as scrollHeight recomputed mid-scroll when items with different real heights rendered, and smooth-scroll animations fought their own moving target; THE REWRITE uses three browser-native primitives that handle variable heights correctly; (1) .speaker-segment in styles.css gets 'content-visibility: auto; contain-intrinsic-size: auto 60px' so ALL ~4500 segments stay in the DOM with a 60px placeholder size for off-screen elements (browser skips paint + layout for them — the perf win virtual scroll was after) but on-screen ones use real measured heights, so document scrollHeight is stable; (2) v-memo on each row in the v-for: only re-renders a row when ITS specific data changes (segment data, highlightedSpeaker for this row's speaker, edit-mode for this row's index) — so typing into a speaker name no longer fans out to 4500 rows; (3) scrollToSegmentIndex is just querySelector + scrollIntoView({block: 'center', behavior: 'smooth'}) — measured positions, no math, smooth animation works because nothing fights it; getSpeakerModalVisibleRange replaced with a DOM-bounding-rect walk that returns the [start, end) range of currently-visible segments; resetSpeakerModalScroll just sets scrollTop=0; UX REBUILD; outer chrome now uses modal-overlay / modal-panel modal-panel--xl (with height: 90vh max-height: 90vh inline to lock the panel size so the inner flex column has a stable boundary) / modal-header / modal-title / modal-close / modal-footer modal-footer--between; speaker name inputs use the field primitive; footer buttons use btn btn--primary / btn--ghost; the 3/5 navigation position indicator is a pill-status pill-status--info; Prev/Next use btn btn--quiet btn--sm so they read as nav rather than primary actions; the Repeat warning for >16 speakers uses pill-status pill-status--warn; the modal feels like the rest of the redesigned app instead of a v0.8.21 leftover; SW cache bumped to v7
Murtaza Nasir committed 5 months ago
3adec73
Speaker modal: revert HEADER + FOOTER + restore modal-content class to exact v0.8.21 markup — actual side-by-side diff found three more genuine regressions I had missed; (1) the main content div was supposed to carry the .modal-content class which supplies max-height: calc(100vh - 200px) + overflow-y: auto from styles.css — I removed it in an earlier 'fix' thinking it was fighting the layout, but it was actually load-bearing for the virtual-scroll container size (the user remembers it working in v0.8.21 because v0.8.21 had this class); (2) the header was using the new .modal-header primitive with different padding (var(--sp-4) var(--sp-5)) and a separate banner sibling for the >16 speakers warning — v0.8.21 nested the warning INSIDE the header div with p-3 sm:p-5; restored the exact v0.8.21 header markup; (3) the footer was using .modal-footer modal-footer--between primitive — v0.8.21 used custom 'bg-tertiary px-3 sm:px-6 py-2 sm:py-4 flex flex-row justify-between rounded-b-xl' chrome with the original purple split-button and gradient apply-suggested button; restored to exact v0.8.21; net effect: the speaker-modal.html is now BYTE-for-byte the v0.8.21 markup with the only addition being the video-collapse toggle I added recently (kept because it's user-requested functionality); SW cache bumped to v6 so users pick up the new modal template via the cached homepage HTML; this is the last big revert — if it's still buggy after a hard SW refresh, the bug is somewhere else and we'll need to actually instrument scroll events to find it
Murtaza Nasir committed 5 months ago
c76b4e3
Speaker modal: revert outer wrapper + scrollToSegmentIndex + highlightSpeakerInTranscript to literal v0.8.21-alpha — last resort after multiple fix attempts kept introducing fresh bugs in the modal-panel-primitive variant; the modal-panel + modal-content combination interacted badly with virtualScroll's ResizeObserver in ways that I couldn't fix without rebuilding the primitive's height semantics; the pragmatic answer is to NOT use the modal-overlay / modal-panel primitives on this particular modal; reverted to the exact v0.8.21 wrapper markup (fixed inset-0 bg-black/70 backdrop-blur + h-[85vh] panel) which was the published version the user remembers working; the inner buttons / fields / footer still use design-system primitives (btn--primary, btn--ghost, etc.) so the visual language is consistent with the rest of the app — only the OUTER chrome (overlay + panel sizing) reverts; scrollToSegmentIndex is back to its 2-line v0.8.21 form (no querySelector / scrollIntoView / nextTick / RAF / suppression flag gymnastics — all of which were workarounds for the broken-height problem and introduced their own bugs once the height was fixed); highlightSpeakerInTranscript reverted to its 'if a group is already visible, just set currentSpeakerGroupIndex and don't scroll; otherwise pick nearest from visible center and scroll to it' behavior — that was the published v0.8.21 logic and was the right semantics; the recent change I made to 'always scroll' was treating a symptom (the visible-range check returning wrong answers because containerHeight was unstable) not a cause
Murtaza Nasir committed 5 months ago
8b279fb
Speaker scroll: instant pre-position + smooth fine-tune (no scroll-event suppression) — the flickering came from the smooth-scroll animation fighting virtual scroll's per-frame visibleRange recompute; each animation frame fires a scroll event, recomputes the range, renders different segments, their real heights differ from the 52 px estimate, scrollHeight changes, smooth-scroll target snaps to a different position, viewport stutters; new approach uses a two-phase split that lets each phase do what it's good at; (1) virtualScroll.scrollToIndex(index-5, 'auto') for the big jump — INSTANT so only one scroll event fires and the range recomputes once, no animation chaos; offsetting by 5 rows means the smooth fine-tune below is a small downward movement, not a large one that would push the target out of the new visible range mid-animation; (2) two requestAnimationFrames to wait for Vue's reactive render + the browser's paint pass so the target DOM node exists with its measured size; (3) scrollIntoView({block:'center', behavior:'smooth'}) on the actual rendered element — because the target is already in the rendered window from step 1, the smooth animation just moves it to the centre of the viewport without changing which items are visible, so visibleRange stays put and no flicker; no scroll-event suppression flag this time — that interfered with the user's mouse-wheel scrolling and was the wrong abstraction; native scroll path stays as plain virtualScroll.onScroll
Murtaza Nasir committed 5 months ago
bc03e37
Revert overengineered scrollToSegmentIndex back to v0.8.21's one-liner — the user is right that I overengineered this; the original was literally; const scrollToSegmentIndex = (index) => { if (showSpeakerModal.value) speakerModalVirtualScroll.scrollToIndex(index, 'smooth'); else mainTranscriptVirtualScroll.scrollToIndex(index, 'smooth'); }; that worked perfectly in v0.8.21-alpha; the ONLY real regression was the modal-panel-primitive migration (commit 8e1c080) that swapped h-[85vh] fixed height for max-height: 88vh and unstable containerHeight; with the modal's height locked back via style=height: 90vh, the original one-liner works the same way it did then; everything I added (querySelector + scrollIntoView two-step, two-nextTick dance, 600 ms scroll-event suppression wrappers, isProgrammaticScroll flag) was solving the WRONG problem and introduced its own race conditions — the smooth-scroll animation fought my querySelector lookup, the suppression window swallowed the user's real mouse-wheel scrolls and made them appear to 'reset', the nextTick gymnastics found the wrong rendered DOM element when virtualScroll's range hadn't caught up yet; reverted the wrappers and the scrollToSegmentIndex back to the original two-line shape; mouse-wheel scroll handlers are back to plain virtualScroll.onScroll references (no suppression), so the user's scroll events flow straight through to virtualScroll's visibleRange recompute the same way they did pre-regression; net code change is roughly -60 lines from my fights with the wrong abstraction
Murtaza Nasir committed 5 months ago
8a9c892
Speaker modal scroll: restore virtual scrolling (preserves #157 perf), keep fixed height + two-step scroll + smooth animation + auto-scroll on input focus — answering the question 'does content-visibility protect against the #157 4500-segment perf regression': NO, because content-visibility only saves browser paint/layout cost, not Vue's reactivity overhead (4500 mounted components × 2 buttons × handlers × dropdown v-ifs = the 2-minute pauses on edits that #157 originally reported); proper fix preserves virtual scroll which keeps ~30 components mounted, AND solves the variable-height navigation problem AND restores the polished UX the user remembered; (1) speaker-modal.html keeps virtual-scroll wiring (spacers, speakerModalVisibleSegments, @scroll handler) — only the modal-panel gets style='height: 90vh' to stop the ResizeObserver oscillation that was the actual recent regression from commit 8e1c080; (2) styles.css .speaker-segment goes back to its plain form (content-visibility removed, was a workaround for the wrong problem); (3) scrollToSegmentIndex restores the two-step dance — virtualScroll.scrollToIndex(index-3, 'auto') gets target's DOM node into the rendered window, then on two nextTicks querySelector + scrollIntoView({block:'center', behavior:'smooth'}) centres it using MEASURED position (handles variable heights) with a polished animation; 600 ms scroll-event suppression covers the smooth-scroll duration (~200-400 ms) so the in-flight visibleRange doesn't recompute and shift the destination underneath us; (4) onScroll wrappers re-added that respect the suppression flag; (5) highlightSpeakerInTranscript now ALWAYS scrolls to the nearest group rather than skipping when something happened to be already visible — earlier 'is it visible' check fought user expectation that clicking into a speaker's textbox would visibly snap to that speaker; with smooth scroll the re-centring on already-visible targets reads as polish, not as a wasted scroll; net result: clicking SPEAKER_05 textbox = smooth jump to nearest SPEAKER_05 occurrence; Next/Prev = smooth jump to next/prev occurrence; mouse-wheel = normal native scroll; 4500-segment transcripts stay fast
Murtaza Nasir committed 5 months ago
8c069f5
Identify Speakers modal scroll: rip out fixed-height virtual scroll, use content-visibility + scrollIntoView, ALSO restore the fixed height that broke during modal-primitive migration — actual regression bisected to commit 8e1c080 'Migrate the remaining 8 modals'; the migration replaced 'class=...h-[85vh] sm:h-[85vh]' (FIXED height) with the new .modal-panel primitive which uses 'max-height: 88vh' (MAX height); with max-height the panel shrinks to its content, so the transcript pane's clientHeight wasn't stable anymore — every render change inside it shifted the modal's overall height, which the ResizeObserver caught and recomputed visibleRange on, which rendered different items, which changed the height again — the classic feedback-loop oscillation the user was reporting; two-part fix to make this truly robust; (1) restored a fixed height on the modal panel via inline style='height: 90vh' (slightly taller than the original 85vh because the new modal-header / modal-footer chrome eats more vertical space than the old hand-rolled headers); this alone would fix the oscillation, BUT virtual scrolling with fixed itemHeight: 52 against truly variable segment heights still mis-positions scrollIntoView targets when the natural items differ wildly from 52px (long podcast paragraphs vs single 'Yes.' lines), so; (2) replaced the speaker-modal virtual scroll with native rendering of all segments + 'content-visibility: auto; contain-intrinsic-size: auto 60px' on .speaker-segment in styles.css — content-visibility is a browser-native primitive that skips the layout/paint cost of off-screen elements while keeping them in the DOM so querySelector/scrollIntoView/Tab focus all work against measured positions; gives us the perf win of virtual scrolling AND eliminates the variable-height bug AND lets the user's natural mouse-wheel scroll never trigger any JS state changes (no @scroll handler at all in the new transcript container); scrollToSegmentIndex for the modal is now just querySelector + scrollIntoView({block:'center', behavior:'auto'}) — robust, non-fragile, uses measured positions; the main transcript panel keeps its virtual scroll wiring unchanged (the bug doesn't show up there because there's no Prev/Next nav with measurable target indices)
Murtaza Nasir committed 5 months ago
aa30913
Fix speaker-modal scroll landing in wrong place + the 'vibrating' on rapid Next clicks — regression introduced by virtual scrolling; the user is right that this worked before — the fix; root cause is that virtualScroll.scrollToIndex uses index * 52 (a FIXED 52 px estimate per row) but real segments range from ~40 px for a single 'Yes.' to 600+ px for a long podcast paragraph, so when you ask to scroll to segment 50 the math says scrollTop=2600 but the actual element 50 lives at scrollTop=8000 because long segments precede it — you land in the middle of an unrelated earlier speaker block, exactly what the screenshot shows (clicked SPEAKER_05 input, viewport stays on SPEAKER_07); the 'vibrating' on rapid clicks is the same bug amplified — each click sets a new wrong scrollTop, the native scroll event fires, the visibleRange recomputes, layout shifts, scrollTop settles at a different wrong spot, repeat; two-part fix; (1) scrollToSegmentIndex is now a two-step dance: first use virtual scroll's fixed-height math to ROUGHLY position so the target index enters the rendered window, then on two consecutive nextTick boundaries find the actual [data-segment-index=N] DOM element via querySelector and call scrollIntoView({block: 'center', behavior: 'auto'}) — scrollIntoView uses the element's MEASURED position so variable heights stop being a problem, and block:'center' gives context above and below the navigated row which is the whole point of Prev/Next; (2) added a 300 ms suppression window via utils.isProgrammaticScroll() — both onSpeakerModalScroll and onMainTranscriptScroll wrappers ignore native scroll events during that window so the programmatic scroll can't trigger a visibleRange recompute that re-shifts the position underneath us; this is what stops the oscillation when Next is clicked rapidly
Murtaza Nasir committed 5 months ago
da43e31
Identify Speakers modal: fix Next/Prev scroll jankiness + add collapsible video — two issues reported in the speaker modal; (1) jerky scroll + target speaker not visible after Next/Prev landed; root cause was scrollToSegmentIndex used 'smooth' behaviour against a virtual-scroll container, which is a fundamentally jerky combo because the browser animates scrollTop in 16ms steps and each step triggers a visibleRange recompute — rows in the destination region don't exist in the DOM yet so the user sees an empty band sliding past until the range catches up; ALSO no upward offset meant the navigated row landed flush at the viewport top edge (visually reading as 'cut off' even though it was technically visible); fix: switch to 'auto' (instant) which paints the new range in a single frame, AND offset upward by ~3 rows so the target lands a comfortable distance below the top with context segments above it; applies to both the speaker modal virtual scroller AND the main transcript virtual scroller (same root issue, just shows up most when speaker nav is the trigger); (2) video files take half the available vertical pane in the speaker modal so the transcript-with-context that Next/Prev relies on has almost no room — added speakerModalVideoCollapsed ref + toggleSpeakerModalVideo action, persisted via localStorage so the user only has to set it once; new button next to the speed control in the modal's audio-transport row (only shown when the recording is video) — fas-video / fas-video-slash icon toggle that flips the <video> element's class between 'w-full rounded-lg mb-2' and 'hidden'; element is always mounted so audio playback continues either way, only the visual rendering toggles, and the transcript pane gets back ~250 px of room when collapsed
Murtaza Nasir committed 5 months ago
1ed4930
Localization sweep round 2 — folder modal labels, editor save hint, dynamic modal titles — addresses user-reported hard-coded English in (a) Edit Folder modal where labels Folder Name / Color / Choose a color / tab names / Transcription Default Settings / Tag settings take priority / Default Language / Min Speakers / Max Speakers / Retention & Auto-Deletion / Retention Period (days) / Protect from Auto-Deletion / Custom Summary Prompt / Naming Template / Export Template / Group Assignment / Auto-share with all group members / Share with group admins only / Note about both enabled / sharing empty title / Save Folder were all hard-coded English (the tag modal had data-i18n attributes but the folder modal didn't); now every label has data-i18n attribute resolving to existing folderManagement.* keys or new ones added in the previous commit; (b) Editor save hint 'Editing — save from the tab bar above' rendered below the EasyMDE editor in both the desktop summary tab and desktop notes tab — new summary.editingSaveHint key in all 7 locales; (c) Dynamic modal titles set by JS via textContent — openTagModal's 'Edit Tag' / 'Create Tag' / 'Update Tag' / 'Save Tag' and openFolderModal's 'Edit Folder' / 'Create Folder' assignments now read from t() with English fallback; (d) tagSharingEmptyState title gets data-i18n attr matching the folder version. Sidebar TODAY / YESTERDAY / LAST WEEK / etc keys all exist in sidebar.{today,yesterday,thisWeek,lastWeek,thisMonth,lastMonth,older,upcoming} for all 7 locales and the code path t('sidebar.today') is correct, so if the user is still seeing English there it's likely a timing/cache issue rather than missing translation
Murtaza Nasir committed 5 months ago
ff6c92f
Tag/Folder modal Sharing empty-state + consistent height; alphabetical language dropdowns; localization sweep for Stats, Processing Queue, Reprocess modal — addresses three user-reported issues; (1) Tag/Folder edit modal Sharing tab now surfaces a role-appropriate empty state when internal sharing is on but the user has no admin groups (was just a blank tab) — site admins see a link to Admin → Groups, regular users see 'Ask your administrator to create a group and add you', applies to both tag and folder edit modals via a shared .modal-tabs-pane wrapper that also fixes the modal height jumping as the user switches between Transcription / Summary & Templates / Sharing tabs (min-height: 320px on the pane keeps the modal a stable size on desktop; min-height: 0 on mobile so it can still collapse), and the JS that toggles group selection now also toggles the new empty state for both modals; (2) language dropdowns (Interface, Transcription, Output language) are now sorted alphabetically in the current locale's collation — backward-compatible because option attributes are untouched, only display order changes; sort code uses .selected property instead of select.value setter to survive detach+re-append cleanly (a previous attempt using select.value caused the dropdown to lose its selection and show English when the actual UI locale was different); auto-detect / default-model sentinel rows pinned to top via data-sort-pin='top'; new syncUILanguageDropdownToActiveLocale() handles DB↔localStorage drift by pulling the dropdown into line with whatever locale is actually rendering, so clicking the displayed value isn't a no-op anymore; (3) localization sweep adds 60+ new keys across 7 locales (en/es/fr/de/ru/zh/pt-BR) for the summary.customizePrompt tooltip (was rendering raw key), the Stats tab headline metrics + per-speaker table headers + empty-state copy (desktop + mobile variants), the Processing Queue popup header / status verbs / Cancel / Failed / Completed / Done / empty-state + retry+delete tooltips, the Reprocess Modal subtitle / Created prefix / Custom Summarization Prompt section / radio labels / tag selector / preview / custom prompt placeholder / no-tags hint / Start Reprocessing button; Vue template fix: progressQueue.*Short keys used as two separate interpolations rather than a single t() call with an object-literal params arg, because Speakr's Vue config uses ${...} delimiters which can't see past nested {} — caught by a runtime SyntaxError after the first attempt
Murtaza Nasir committed 5 months ago
29d11fe
Version-sync infrastructure + bump VERSION to v0.9.0 — fixes the long-standing problem where VERSION at repo root drifted out of sync with the README version badge and the docs (the About page in the Speakr app reads VERSION via src/config/version.py's get_version(), so when VERSION lags behind a release commit the in-app About page silently shows a stale version even though the README and docs already advertise the new one); two-part fix: (1) bumps VERSION from v0.8.21-alpha to v0.9.0 to match the README badge, docs/index.md banner, release_notes_v0.9.0.md, and the v0.9.0 docs Wave 1+2 commits; (2) extends scripts/pre-commit (the hook is already symlinked into .git/hooks/pre-commit per the existing install line) to detect drift on every commit that touches README.md or VERSION — when either is staged the hook parses the version out of the shields.io badge URL in the README, compares it to the VERSION file contents, and if they differ rewrites VERSION + git-adds it so the commit captures the sync; the hook is intentionally non-blocking (no exit 1) since the auto-fix is safe, just prints a 'Syncing VERSION file: old → new' line so the author sees what happened; also upgrades scripts/update_version.py so the manual flow (python scripts/update_version.py v0.9.1) now updates BOTH the VERSION file AND the README badge in one shot, then optionally creates the git tag — previously the script only touched VERSION which is what got us into this drift situation; the README badge is the single source of truth in both flows so it doesn't matter whether you bump README and let the hook fix VERSION, or run the script which fixes both at once
Murtaza Nasir committed 5 months ago
587f6a4