@antiphony/shared 0.4.0 → 0.5.1

package/dist/cjs/api-codecs.cjs

@@ -46,21 +46,109 @@ var FirestoreTimestampSchema = zod.z.union([
   return date;
 });
 var ProcessingStageStatusSchema = zod.z.enum(["pending", "ready", "failed", "skipped"]);
+var PROCESSING_STAGES = ["denoise", "trim", "transcribe", "waveform"];
+zod.z.enum(PROCESSING_STAGES);
 var ProcessingRequestSchema = zod.z.object({
   transcribe: zod.z.boolean().optional(),
-  denoise: zod.z.boolean().optional()
+  denoise: zod.z.boolean().optional(),
+  trim: zod.z.boolean().optional(),
+  waveform: zod.z.boolean().optional(),
+  /**
+   * Whether a completed byte-mutating stage should invalidate and recompute
+   * the derived artifacts that describe the old audio. Defaults to **true**
+   * — a transcript of superseded audio is wrong, not merely stale.
+   *
+   * `false` opts out, for an app that would rather keep the existing
+   * transcript than pay to regenerate it. It does NOT name a stage, so a
+   * request carrying only `reprocess` requests no work.
+   */
+  reprocess: zod.z.boolean().optional()
 });
-var ProcessingStateSchema = zod.z.object({
+var ProcessingStageMapSchema = zod.z.object({
   transcribe: ProcessingStageStatusSchema.optional(),
   denoise: ProcessingStageStatusSchema.optional(),
-  /** Content CID of the denoised audio variant, once `denoise === 'ready'`. */
-  denoisedBlobCid: zod.z.string().optional(),
-  updatedAt: FirestoreTimestampSchema
+  trim: ProcessingStageStatusSchema.optional(),
+  waveform: ProcessingStageStatusSchema.optional()
 });
-var ProcessingViewSchema = zod.z.object({
-  transcribe: ProcessingStageStatusSchema.optional(),
-  denoise: ProcessingStageStatusSchema.optional()
+var ResolvedProcessingSchema = ProcessingStageMapSchema.extend({
+  reprocess: zod.z.boolean().optional()
 });
+var ProcessingStateSchema = ResolvedProcessingSchema.extend({
+  /**
+   * Content CID of the processed audio variant — the composed output of every
+   * byte-mutating stage that has completed. The record's own
+   * `embed.audio.ref.$link` stays the ORIGINAL CID (immutable content
+   * address); only the read-time view swaps playback to this variant.
+   */
+  processedBlobCid: zod.z.string().optional(),
+  /**
+   * MIME type of the processed variant. Present because providers may
+   * TRANSCODE — the ElevenLabs Voice Isolator returns MP3 regardless of what
+   * it is given — so the variant's type cannot be assumed to match
+   * `embed.audio.mimeType`. Anything reading the variant's bytes must use
+   * this, not the embed's.
+   */
+  processedMimeType: zod.z.string().optional(),
+  /**
+   * Duration of the processed variant, when a byte-mutating stage changed it
+   * (i.e. trim). Absent when the variant's duration matches the original.
+   */
+  processedDurationMs: zod.z.number().int().min(0).optional(),
+  /**
+   * Which denoiser produced the variant's denoise contribution — provenance,
+   * the counterpart to a transcript record's `model`.
+   *
+   * Lives here because a cleaned variant, unlike a transcript, has no record
+   * of its own to carry it: it is a blob CID on this state. Without it,
+   * changing denoisers leaves no way to tell which variants predate the
+   * switch, so nothing can identify what to re-run.
+   *
+   * Named for the STAGE, not the variant (`processedModel`), because it
+   * describes one link of the byte-mutating chain rather than the composed
+   * artifact. Trim contributes to the same variant and has no model, and a
+   * later external link would want its own field rather than to overwrite
+   * this one.
+   *
+   * Written on every successful denoise, never cleared — it moves with
+   * `processedBlobCid`, which is only ever set, never reset. A denoise that
+   * FAILS leaves both alone, which is correct: the variant still holds the
+   * previous denoiser's output, so the previous model still describes it.
+   *
+   * Internal, like the other variant fields — `toProcessingView` projects
+   * stages only, so this never reaches a client.
+   */
+  denoiseModel: zod.z.string().optional(),
+  /**
+   * Peaks for the processed variant, once the `waveform` stage completes.
+   * Same normalization and bounds as `embed.waveform` (0–100, max 1000), so
+   * a view can never carry a larger payload than the record allows.
+   */
+  waveformPeaks: zod.z.array(zod.z.number().int().min(0).max(100)).max(1e3).optional(),
+  /**
+   * When the current runner's exclusive claim on this post expires.
+   *
+   * Queue delivery is at-least-once, so the same job can arrive twice and
+   * run CONCURRENTLY. `process()` is idempotent under sequential retry — it
+   * acts on `pending` and re-does nothing already settled — but two passes
+   * interleaved is a different failure: both read the same `pending` state,
+   * both bill the provider for the same stage, and both write
+   * `processedBlobCid`, so the surviving variant is whichever finished last
+   * and the other's blob is orphaned.
+   *
+   * A runner claims this field transactionally before doing any work and
+   * clears it when finished; a second runner finding it unexpired declines
+   * and returns. It is an EXPIRY, not a boolean lock, because the holder can
+   * die mid-run (instance recycled, process killed) with no chance to
+   * release — a plain flag would strand the post permanently, where a lapsed
+   * lease lets the next delivery pick it up.
+   *
+   * Internal, like the variant fields above: `toProcessingView` projects
+   * stages only, so this never reaches a client.
+   */
+  leaseUntil: FirestoreTimestampSchema.optional(),
+  updatedAt: FirestoreTimestampSchema
+});
+var ProcessingViewSchema = ProcessingStageMapSchema;
 
 // types/audio.ts
 var StrongRefSchema = zod.z.object({
@@ -96,20 +184,34 @@ var TimedTranscriptSchema = zod.z.object({
 });
 var AudioEmbedViewSchema = zod.z.object({
   $type: zod.z.literal("dev.antiphony.embed.audio#view"),
+  /**
+   * `url`, `durationMs` and `waveform` are RESOLVED, not copied: once
+   * processing has produced an audio variant they describe that variant
+   * rather than the bytes the client uploaded. They always agree with one
+   * another — a duration and a set of peaks are only meaningful against the
+   * audio `url` actually points at.
+   *
+   * A client that stored `durationMs` at upload time should therefore expect
+   * it to change (trim removes leading/trailing silence), and should render
+   * these three as a set rather than caching them independently. The record's
+   * originals are immutable and unaffected; this is a read-time resolution.
+   */
   url: zod.z.string().url(),
   durationMs: zod.z.number().int().min(0).optional(),
-  // `alt`/`waveform` are copied from the stored embed; keep the same bounds
-  // so a view can never carry a larger payload than the record allows.
+  // `alt` is copied from the stored embed; keep the same bounds so a view can
+  // never carry a larger payload than the record allows.
   alt: zod.z.string().max(1e4).optional(),
   waveform: zod.z.array(zod.z.number().int().min(0).max(100)).max(1e3).optional(),
   /** Lifted from the transcript enrichment record; absent until transcription completes. */
   transcript: TimedTranscriptSchema.optional(),
   /**
-   * Per-stage audio-processing status (transcribe / denoise), when the app
-   * opted into processing on create. Absent otherwise. A `pending` stage
-   * means the client should poll (or re-render) for the result. When
-   * `denoise === 'ready'`, `url` above already resolves to the cleaned
-   * audio variant. See `types/processing.ts`.
+   * Per-stage audio-processing status (denoise / trim / transcribe /
+   * waveform), when the app opted into processing. Absent otherwise. A
+   * `pending` stage means the client should poll (or re-render) for the
+   * result — including a stage that returns to `pending` after having been
+   * `ready`, which is how a recompute surfaces. Once a byte-mutating stage
+   * completes, `url`/`durationMs`/`waveform` above already resolve to the
+   * processed audio variant. See `types/processing.ts`.
    */
   processing: ProcessingViewSchema.optional()
 });
@@ -213,10 +315,15 @@ var ViewerStateSchema = zod.z.object({
   replyDisabledReason: zod.z.enum(["unauthenticated", "not_a_participant"]).optional()
 });
 var PostRecordPublicSchema = zod.z.object({
-  text: zod.z.string(),
-  title: zod.z.string().optional(),
+  // `text`, `title` and `langs` keep the record's bounds, for the same reason
+  // `AudioEmbedViewSchema.alt` does: a view must never be able to carry a
+  // larger payload than the record it projects. The write path already
+  // enforces these, so no stored post can exceed them — stating them here
+  // keeps the published contract honest rather than adding a new constraint.
+  text: zod.z.string().max(3e3),
+  title: zod.z.string().max(3e3).optional(),
   reply: ReplyRefSchema.optional(),
-  langs: zod.z.array(zod.z.string()).optional(),
+  langs: zod.z.array(zod.z.string()).max(3).optional(),
   selfLabels: zod.z.array(zod.z.string()).optional(),
   createdAt: FirestoreTimestampSchema
 });
@@ -251,9 +358,11 @@ var CreateAudioPostRequestSchema = zod.z.object({
   /** Author self-label values (content warnings). */
   selfLabels: zod.z.array(zod.z.string()).optional(),
   /**
-   * Opt-in audio processing for this post's audio (transcribe / denoise).
-   * Both default off. Stages the deployment can't provide come back marked
-   * `skipped` on the view rather than failing the create. See
+   * Opt-in audio processing for this post's audio (denoise / trim /
+   * transcribe / waveform). All default off. Stages the deployment can't
+   * provide come back marked `skipped` on the view rather than failing the
+   * create. A multi-stage request runs denoise → trim → (transcribe,
+   * waveform); request stages individually to override that order. See
    * `types/processing.ts`.
    */
   processing: ProcessingRequestSchema.optional()

package/dist/cjs/api-codecs.d.cts

@@ -53,9 +53,9 @@ declare const CreateAudioPostRequestSchema: z.ZodEffects<z.ZodEffects<z.ZodObjec
             mimeType: string;
             size: number;
         };
+        waveform?: number[] | undefined;
         durationMs?: number | undefined;
         alt?: string | undefined;
-        waveform?: number[] | undefined;
     }, {
         $type: "dev.antiphony.embed.audio";
         audio: {
@@ -66,9 +66,9 @@ declare const CreateAudioPostRequestSchema: z.ZodEffects<z.ZodEffects<z.ZodObjec
             mimeType: string;
             size: number;
         };
+        waveform?: number[] | undefined;
         durationMs?: number | undefined;
         alt?: string | undefined;
-        waveform?: number[] | undefined;
     }>>;
     /** Present ⇒ this is a reply (StrongRef root + parent). */
     reply: z.ZodOptional<z.ZodObject<{
@@ -116,26 +116,40 @@ declare const CreateAudioPostRequestSchema: z.ZodEffects<z.ZodEffects<z.ZodObjec
     /** Author self-label values (content warnings). */
     selfLabels: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
     /**
-     * Opt-in audio processing for this post's audio (transcribe / denoise).
-     * Both default off. Stages the deployment can't provide come back marked
-     * `skipped` on the view rather than failing the create. See
+     * Opt-in audio processing for this post's audio (denoise / trim /
+     * transcribe / waveform). All default off. Stages the deployment can't
+     * provide come back marked `skipped` on the view rather than failing the
+     * create. A multi-stage request runs denoise → trim → (transcribe,
+     * waveform); request stages individually to override that order. See
      * `types/processing.ts`.
      */
     processing: z.ZodOptional<z.ZodObject<{
         transcribe: z.ZodOptional<z.ZodBoolean>;
         denoise: z.ZodOptional<z.ZodBoolean>;
+        trim: z.ZodOptional<z.ZodBoolean>;
+        waveform: z.ZodOptional<z.ZodBoolean>;
+        reprocess: z.ZodOptional<z.ZodBoolean>;
     }, "strip", z.ZodTypeAny, {
-        transcribe?: boolean | undefined;
         denoise?: boolean | undefined;
-    }, {
+        trim?: boolean | undefined;
         transcribe?: boolean | undefined;
+        waveform?: boolean | undefined;
+        reprocess?: boolean | undefined;
+    }, {
         denoise?: boolean | undefined;
+        trim?: boolean | undefined;
+        transcribe?: boolean | undefined;
+        waveform?: boolean | undefined;
+        reprocess?: boolean | undefined;
     }>>;
 }, "strip", z.ZodTypeAny, {
     text: string;
     processing?: {
-        transcribe?: boolean | undefined;
         denoise?: boolean | undefined;
+        trim?: boolean | undefined;
+        transcribe?: boolean | undefined;
+        waveform?: boolean | undefined;
+        reprocess?: boolean | undefined;
     } | undefined;
     reply?: {
         root: {
@@ -158,17 +172,20 @@ declare const CreateAudioPostRequestSchema: z.ZodEffects<z.ZodEffects<z.ZodObjec
             mimeType: string;
             size: number;
         };
+        waveform?: number[] | undefined;
         durationMs?: number | undefined;
         alt?: string | undefined;
-        waveform?: number[] | undefined;
     } | undefined;
     langs?: string[] | undefined;
     selfLabels?: string[] | undefined;
 }, {
     text?: string | undefined;
     processing?: {
-        transcribe?: boolean | undefined;
         denoise?: boolean | undefined;
+        trim?: boolean | undefined;
+        transcribe?: boolean | undefined;
+        waveform?: boolean | undefined;
+        reprocess?: boolean | undefined;
     } | undefined;
     reply?: {
         root: {
@@ -191,17 +208,20 @@ declare const CreateAudioPostRequestSchema: z.ZodEffects<z.ZodEffects<z.ZodObjec
             mimeType: string;
             size: number;
         };
+        waveform?: number[] | undefined;
         durationMs?: number | undefined;
         alt?: string | undefined;
-        waveform?: number[] | undefined;
     } | undefined;
     langs?: string[] | undefined;
     selfLabels?: string[] | undefined;
 }>, {
     text: string;
     processing?: {
-        transcribe?: boolean | undefined;
         denoise?: boolean | undefined;
+        trim?: boolean | undefined;
+        transcribe?: boolean | undefined;
+        waveform?: boolean | undefined;
+        reprocess?: boolean | undefined;
     } | undefined;
     reply?: {
         root: {
@@ -224,17 +244,20 @@ declare const CreateAudioPostRequestSchema: z.ZodEffects<z.ZodEffects<z.ZodObjec
             mimeType: string;
             size: number;
         };
+        waveform?: number[] | undefined;
         durationMs?: number | undefined;
         alt?: string | undefined;
-        waveform?: number[] | undefined;
     } | undefined;
     langs?: string[] | undefined;
     selfLabels?: string[] | undefined;
 }, {
     text?: string | undefined;
     processing?: {
-        transcribe?: boolean | undefined;
         denoise?: boolean | undefined;
+        trim?: boolean | undefined;
+        transcribe?: boolean | undefined;
+        waveform?: boolean | undefined;
+        reprocess?: boolean | undefined;
     } | undefined;
     reply?: {
         root: {
@@ -257,17 +280,20 @@ declare const CreateAudioPostRequestSchema: z.ZodEffects<z.ZodEffects<z.ZodObjec
             mimeType: string;
             size: number;
         };
+        waveform?: number[] | undefined;
         durationMs?: number | undefined;
         alt?: string | undefined;
-        waveform?: number[] | undefined;
     } | undefined;
     langs?: string[] | undefined;
     selfLabels?: string[] | undefined;
 }>, {
     text: string;
     processing?: {
-        transcribe?: boolean | undefined;
         denoise?: boolean | undefined;
+        trim?: boolean | undefined;
+        transcribe?: boolean | undefined;
+        waveform?: boolean | undefined;
+        reprocess?: boolean | undefined;
     } | undefined;
     reply?: {
         root: {
@@ -290,17 +316,20 @@ declare const CreateAudioPostRequestSchema: z.ZodEffects<z.ZodEffects<z.ZodObjec
             mimeType: string;
             size: number;
         };
+        waveform?: number[] | undefined;
         durationMs?: number | undefined;
         alt?: string | undefined;
-        waveform?: number[] | undefined;
     } | undefined;
     langs?: string[] | undefined;
     selfLabels?: string[] | undefined;
 }, {
     text?: string | undefined;
     processing?: {
-        transcribe?: boolean | undefined;
         denoise?: boolean | undefined;
+        trim?: boolean | undefined;
+        transcribe?: boolean | undefined;
+        waveform?: boolean | undefined;
+        reprocess?: boolean | undefined;
     } | undefined;
     reply?: {
         root: {
@@ -323,9 +352,9 @@ declare const CreateAudioPostRequestSchema: z.ZodEffects<z.ZodEffects<z.ZodObjec
             mimeType: string;
             size: number;
         };
+        waveform?: number[] | undefined;
         durationMs?: number | undefined;
         alt?: string | undefined;
-        waveform?: number[] | undefined;
     } | undefined;
     langs?: string[] | undefined;
     selfLabels?: string[] | undefined;
@@ -346,22 +375,37 @@ declare const PatchAudioPostRequestSchema: z.ZodObject<{
     processing: z.ZodObject<{
         transcribe: z.ZodOptional<z.ZodBoolean>;
         denoise: z.ZodOptional<z.ZodBoolean>;
+        trim: z.ZodOptional<z.ZodBoolean>;
+        waveform: z.ZodOptional<z.ZodBoolean>;
+        reprocess: z.ZodOptional<z.ZodBoolean>;
     }, "strip", z.ZodTypeAny, {
-        transcribe?: boolean | undefined;
         denoise?: boolean | undefined;
-    }, {
+        trim?: boolean | undefined;
         transcribe?: boolean | undefined;
+        waveform?: boolean | undefined;
+        reprocess?: boolean | undefined;
+    }, {
         denoise?: boolean | undefined;
+        trim?: boolean | undefined;
+        transcribe?: boolean | undefined;
+        waveform?: boolean | undefined;
+        reprocess?: boolean | undefined;
     }>;
 }, "strip", z.ZodTypeAny, {
     processing: {
-        transcribe?: boolean | undefined;
         denoise?: boolean | undefined;
+        trim?: boolean | undefined;
+        transcribe?: boolean | undefined;
+        waveform?: boolean | undefined;
+        reprocess?: boolean | undefined;
     };
 }, {
     processing: {
-        transcribe?: boolean | undefined;
         denoise?: boolean | undefined;
+        trim?: boolean | undefined;
+        transcribe?: boolean | undefined;
+        waveform?: boolean | undefined;
+        reprocess?: boolean | undefined;
     };
 }>;
 type PatchAudioPostRequest = z.infer<typeof PatchAudioPostRequestSchema>;