@kitn.ai/ui 0.28.0 → 0.29.0

package/src/state/parts.ts

@@ -1,295 +0,0 @@
-import type { MessagePart, RawOrigin } from '../elements/chat-types';
-import type { ToolPart } from '../components/tool-types';
-import type { CardEnvelope } from '../primitives/card-contract';
-import { classifyTool } from '../components/tool-classify';
-
-/** Stable structural fingerprint. Key order independent, so an identical snapshot
- *  arriving twice compares equal and can be skipped. */
-export function fingerprint(value: unknown): string {
-  const seen = new WeakSet<object>();
-  const walk = (v: unknown): unknown => {
-    if (v === null || typeof v !== 'object') return v;
-    if (seen.has(v as object)) return '[circular]';
-    seen.add(v as object);
-    const o = v as Record<string, unknown>;
-    const out = Array.isArray(v)
-      ? v.map(walk)
-      : Object.fromEntries(Object.keys(o).sort().map((k) => [k, walk(o[k])]));
-    seen.delete(v as object);
-    return out;
-  };
-  return JSON.stringify(walk(value)) ?? '';
-}
-
-/** Appends to the trailing text part, or OPENS A NEW ONE if the last part is not
- *  text. This is what stops a post-tool answer being glued onto the pre-tool text. */
-export function appendTextPart(parts: MessagePart[], delta: string): MessagePart[] {
-  const last = parts[parts.length - 1];
-  if (last?.type === 'text') {
-    return [...parts.slice(0, -1), { ...last, text: last.text + delta }];
-  }
-  return [...parts, { type: 'text', text: delta }];
-}
-
-export interface ReasoningOpts {
-  index?: number;
-  /** Namespaces `index` to one provider response stream. Producers that read more
-   *  than one stream into the SAME message must set it. See `appendReasoningPart`. */
-  streamId?: string;
-  label?: string;
-  signature?: string;
-  raw?: RawOrigin;
-}
-
-type ReasoningPart = Extract<MessagePart, { type: 'reasoning' }>;
-
-/** Keyed by `(streamId, index)` so parallel reasoning blocks stay distinct.
- *
- *  WHY THE KEY IS A PAIR. A block index alone is NOT unique inside one `parts`
- *  array. Anthropic numbers content blocks per MESSAGE and restarts at 0 on the
- *  next one, while a tool loop folds every round into a single assistant turn.
- *  Keyed on index alone, round 2's thinking block (index 0) merges into round 1's
- *  part: the text concatenates and `raw: opts.raw ?? cur.raw` OVERWRITES round 1's
- *  verbatim provider payload with round 2's. `toAnthropicMessages` then emits one
- *  thinking block where two belong, carrying round 2's signature in round 1's
- *  position -- a modified-and-filtered thinking block, which is exactly the 400
- *  the verbatim `raw` channel exists to prevent.
- *
- *  `streamId` is the namespace: one value per provider response stream, attached
- *  by `consumeModelStream`. Two rounds are two streams, so their index 0s are two
- *  parts. Producers that drive a sink from a single stream can omit it; `undefined`
- *  is its own namespace and behaves exactly as before.
- *
- *  Returns the SAME array reference when the merge produces an identical part,
- *  for the same reason `upsertToolPart` does: a new `parts` array is the
- *  re-render signal, so handing one back for a delta that changed nothing is a
- *  spurious render.
- *
- *  An EMPTY delta is not a no-op and must still reach here: it is how a redacted
- *  reasoning block, a `signature_delta` and an assembled `content_block_stop`
- *  block arrive, and how a format opens a block at the right position so block
- *  ORDER survives into `parts`. Those carry a new `raw`/`signature`/index and so
- *  compare unequal and DO rebuild. What the check absorbs is the other empty
- *  frame: one carrying nothing new, which a provider is free to send repeatedly.
- *
- *  `signature` and `raw` resolve with `??`, so an explicit `undefined` from a
- *  later delta never blanks a value an earlier one established. Pass a DEFINED
- *  value to replace either; there is no way to clear them. */
-export function appendReasoningPart(
-  parts: MessagePart[],
-  delta: string,
-  opts: ReasoningOpts = {},
-): MessagePart[] {
-  const index = opts.index ?? 0;
-  const streamId = opts.streamId;
-  const i = parts.findIndex(
-    (p) => p.type === 'reasoning' && (p.index ?? 0) === index && p.streamId === streamId,
-  );
-  if (i < 0) {
-    return [...parts, { type: 'reasoning', text: delta, index, streamId, label: opts.label, signature: opts.signature, raw: opts.raw }];
-  }
-  const cur = parts[i] as ReasoningPart;
-  const next: ReasoningPart = {
-    ...cur,
-    text: cur.text + delta,
-    label: opts.label ?? cur.label,
-    signature: opts.signature ?? cur.signature,
-    raw: opts.raw ?? cur.raw,
-  };
-  if (reasoningEqual(cur, next)) return parts;
-  return [...parts.slice(0, i), next, ...parts.slice(i + 1)];
-}
-
-/** Every `ReasoningPart` key, in the order `reasoningEqual` checks them. Same
- *  exhaustive-by-construction contract as `TOOL_KEYS` below: adding a field to
- *  the `reasoning` variant of `MessagePart` fails the build until it is listed
- *  here AND given a comparator, so a new field can never be silently ignored by
- *  the dedupe check and strand a stale array reference. */
-const REASONING_KEYS = [
-  'type', 'text', 'index', 'streamId', 'label', 'signature', 'raw',
-] as const satisfies readonly (keyof ReasoningPart)[];
-
-type _ReasoningKeysExhaustive = Exclude<
-  keyof ReasoningPart,
-  (typeof REASONING_KEYS)[number]
-> extends never
-  ? true
-  : ['MessagePart reasoning variant has a key missing from REASONING_KEYS', Exclude<keyof ReasoningPart, (typeof REASONING_KEYS)[number]>];
-const _reasoningKeysExhaustive: _ReasoningKeysExhaustive = true;
-void _reasoningKeysExhaustive;
-
-/** One comparator per key in `REASONING_KEYS`. The mapped type means a key added
- *  to that list with no comparator here is ALSO a compile error.
- *
- *  `raw` compares by REFERENCE, exactly as `TOOL_COMPARATORS.raw` does: it is the
- *  untranslated provider payload, a producer attaches it once, and
- *  `appendReasoningPart` itself carries it forward when a delta omits it, so
- *  reference equality holds on every real path. Hashing it would walk the whole
- *  payload on every empty keep-alive frame, which is the cost this check exists
- *  to avoid. Worst case is a producer handing over a fresh-but-equal `raw`: one
- *  extra re-render, never a wrong render. */
-const REASONING_COMPARATORS: {
-  [K in (typeof REASONING_KEYS)[number]]: (a: ReasoningPart, b: ReasoningPart) => boolean;
-} = {
-  type: (a, b) => a.type === b.type,
-  text: (a, b) => a.text === b.text,
-  index: (a, b) => a.index === b.index,
-  streamId: (a, b) => a.streamId === b.streamId,
-  label: (a, b) => a.label === b.label,
-  signature: (a, b) => a.signature === b.signature,
-  raw: (a, b) => a.raw === b.raw,
-};
-
-function reasoningEqual(a: ReasoningPart, b: ReasoningPart): boolean {
-  return REASONING_KEYS.every((key) => REASONING_COMPARATORS[key](a, b));
-}
-
-type CardPart = Extract<MessagePart, { type: 'card' }>;
-
-/** Creates or REPLACES a card part, keyed on `envelope.id`. Returns the SAME array
- *  reference when the incoming envelope is structurally identical to the current
- *  one, for the same reason `upsertToolPart` does: a new `parts` array is the
- *  re-render signal, so handing one back for a revision that changed nothing is a
- *  spurious render.
- *
- *  WHY THIS REPLACES WHERE `upsertToolPart` MERGES. A tool part is patched
- *  fragment-by-fragment as its arguments stream in, which is why that function
- *  needs carry-forward rules for `raw` and `kind` — a later patch that omits a
- *  field is not asserting the field is gone. A card envelope is the opposite: it
- *  arrives WHOLE, as one complete tool result, so an omitted field IS an
- *  assertion. Last-write-wins is both simpler and the only semantics under which
- *  a host can CLEAR `resolution` to re-open a dismissed card — a field-by-field
- *  merge can only ever set that field, never unset it, so `CardPolicy.onReopen`
- *  (see `primitives/card-contract.ts`) would have no way to express its result.
- *
- *  The PART-level `raw` is preserved across a revision. It is a different field
- *  from anything inside the envelope: the untranslated provider payload the part
- *  was built from, attached once by the producer, which a fresh envelope carries
- *  no opinion about.
- *
- *  Position is preserved: a revised card stays where it first appeared in the
- *  thread rather than jumping past the text that followed it. */
-export function upsertCardPart(parts: MessagePart[], envelope: CardEnvelope): MessagePart[] {
-  const i = parts.findIndex((p) => p.type === 'card' && p.envelope.id === envelope.id);
-  if (i < 0) return [...parts, { type: 'card', envelope }];
-  const cur = parts[i] as CardPart;
-  if (fingerprint(cur.envelope) === fingerprint(envelope)) return parts;
-  return [...parts.slice(0, i), { ...cur, envelope }, ...parts.slice(i + 1)];
-}
-
-/** Creates or merges a tool part. Returns the SAME array reference when the merge
- *  produces an identical tool, so repeated snapshots do not trigger a re-render.
- *
- *  Two fields do NOT follow plain spread semantics, because a streaming provider
- *  hands them over on one fragment and then keeps patching the rest:
- *  - `kind`: a value the consumer set is preserved across later patches instead
- *    of being reverted to `classifyTool(type)` (see `resolveKind`).
- *  - `raw`: an explicit `raw: undefined` never blanks a `raw` an earlier patch
- *    established. Pass a DEFINED `raw` to replace it; there is no way to clear it. */
-export function upsertToolPart(
-  parts: MessagePart[],
-  toolCallId: string,
-  patch: Partial<ToolPart>,
-): MessagePart[] {
-  const i = parts.findIndex((p) => p.type === 'tool' && p.tool.toolCallId === toolCallId);
-  if (i < 0) {
-    const tool: ToolPart = {
-      type: patch.type ?? 'unknown',
-      state: patch.state ?? 'input-streaming',
-      ...patch,
-      toolCallId,
-    };
-    tool.kind = tool.kind ?? classifyTool(tool.type);
-    return [...parts, { type: 'tool', tool }];
-  }
-  const cur = (parts[i] as Extract<MessagePart, { type: 'tool' }>).tool;
-  const merged: ToolPart = { ...cur, ...patch, toolCallId };
-  merged.kind = resolveKind(cur, patch, merged.type);
-  if (merged.raw === undefined && cur.raw !== undefined) merged.raw = cur.raw;
-  if (toolsEqual(cur, merged)) return parts;
-  return [...parts.slice(0, i), { type: 'tool', tool: merged }, ...parts.slice(i + 1)];
-}
-
-/** `kind` precedence on an update, in order:
- *  1. an explicit `patch.kind` always wins;
- *  2. a `kind` the consumer set that is NOT merely the auto-derivation of the
- *     current `type` is preserved. Create honors `patch.kind`, so an update must
- *     not silently revert a custom classification mid-stream (a tool created as
- *     `{ type: 'my_widget_tool', kind: 'mcp' }` stays `mcp` after a plain
- *     `{ state: 'output-available' }` patch);
- *  3. otherwise re-derive from the merged `type`, so a tool whose real name only
- *     arrives in a later fragment still classifies (the streaming path: created
- *     as `unknown`/`generic`, patched to `web_search` -> `search`).
- *
- *  A consumer `kind` that happens to equal `classifyTool(type)` is
- *  indistinguishable from the derived one, which is harmless: they agree until
- *  the `type` changes, and then re-deriving is the better answer anyway. */
-function resolveKind(cur: ToolPart, patch: Partial<ToolPart>, nextType: string): ToolPart['kind'] {
-  if (patch.kind !== undefined) return patch.kind;
-  const derived = cur.kind === undefined || cur.kind === classifyTool(cur.type);
-  return derived ? classifyTool(nextType) : cur.kind;
-}
-
-/** Every `ToolPart` key, in the order `toolsEqual` checks them (cheap identity
- *  checks first, the two fingerprint-based checks last). See `ToolPart` in
- *  `components/tool-types.ts`, which points back here.
- *
- *  This is not just documentation: `_toolKeysExhaustive` below fails to compile
- *  if `ToolPart` gains a key that is not listed here, and `TOOL_COMPARATORS`
- *  fails to compile if this list gains a key with no comparator. Adding a field
- *  to `ToolPart` without wiring it into both is a BUILD failure, not a silently
- *  stale array reference. */
-const TOOL_KEYS = [
-  'type', 'state', 'kind', 'toolCallId', 'errorText', 'rawInput', 'raw', 'input', 'output',
-] as const satisfies readonly (keyof ToolPart)[];
-
-// If `ToolPart` gains a key absent from `TOOL_KEYS`, `Exclude<...>` stops being
-// `never`, the assigned literal type no longer matches `true`, and `tsc` reports
-// the offending key name right here.
-type _ToolKeysExhaustive = Exclude<keyof ToolPart, (typeof TOOL_KEYS)[number]> extends never
-  ? true
-  : ['ToolPart has a key missing from TOOL_KEYS', Exclude<keyof ToolPart, (typeof TOOL_KEYS)[number]>];
-const _toolKeysExhaustive: _ToolKeysExhaustive = true;
-void _toolKeysExhaustive;
-
-/** One comparator per key in `TOOL_KEYS`. The mapped type over
- *  `(typeof TOOL_KEYS)[number]` means a key added to `TOOL_KEYS` with no
- *  comparator here is ALSO a compile error ("Property 'x' is missing"), so the
- *  guard reaches the actual comparison, not just the key list.
- *
- *  Semantics are unchanged from before this table existed:
- *  - `raw` compares by REFERENCE on purpose. It is the untranslated provider
- *    payload; a producer attaches it once and never rebuilds it (upsertToolPart
- *    itself carries `cur.raw` forward when a patch omits it), so reference
- *    equality holds on every real path. Hashing it would walk the accumulated
- *    argument string a second time, which is the cost this table exists to
- *    remove. The worst case is a producer handing over a fresh-but-equal `raw`,
- *    which costs one extra re-render and never a wrong render.
- *  - `input` and `output` are the two fields that are genuinely object-shaped:
- *    reference equality first, then `fingerprint()` as a structural fallback so
- *    a fresh-but-identical object still dedupes.
- *  - everything else, including `rawInput`, compares with `===`. `rawInput`
- *    against `!==` is exactly the test we want (did the accumulated text
- *    change?) and it is cheap: two strings of different lengths are unequal
- *    after the length check, so it never walks the full string.
- *
- *  This replaces `fingerprint(merged) === fingerprint(cur)`, which serialized
- *  the ENTIRE ToolPart on every patch. A streaming tool call is patched once
- *  per argument fragment while `rawInput` grows toward the full argument JSON,
- *  so hashing the whole part per fragment is quadratic in the argument size:
- *  fine at 4 KB, not at 200 KB. */
-const TOOL_COMPARATORS: { [K in (typeof TOOL_KEYS)[number]]: (a: ToolPart, b: ToolPart) => boolean } = {
-  type: (a, b) => a.type === b.type,
-  state: (a, b) => a.state === b.state,
-  kind: (a, b) => a.kind === b.kind,
-  toolCallId: (a, b) => a.toolCallId === b.toolCallId,
-  errorText: (a, b) => a.errorText === b.errorText,
-  rawInput: (a, b) => a.rawInput === b.rawInput,
-  raw: (a, b) => a.raw === b.raw,
-  input: (a, b) => a.input === b.input || fingerprint(a.input) === fingerprint(b.input),
-  output: (a, b) => a.output === b.output || fingerprint(a.output) === fingerprint(b.output),
-};
-
-function toolsEqual(a: ToolPart, b: ToolPart): boolean {
-  return TOOL_KEYS.every((key) => TOOL_COMPARATORS[key](a, b));
-}

package/src/state/persistence.ts

@@ -1,180 +0,0 @@
-// src/state/persistence.ts — the mechanical halves of the persistence story.
-//
-// The boundary (spec §1, CLAUDE.md): persistence POLICY — what is stored,
-// where, for how long, under what quota — is the app's. What lives here is
-// only the mechanics every such app re-derives by hand: validating a stored
-// thread back into `ChatMessage[]` (F-18), and the debounce/flush shape that
-// keeps a per-token stream from hitting storage per token. No storage call,
-// no fetch, no DOM; the save fn, the delay, and the reaction to a drop are
-// all the consumer's.
-import type { ChatMessage, FeedbackVote, MessagePart } from '../elements/chat-types';
-
-/** One record the parse refused, and why. Deciding loudly is the caller's
- *  verb (warn, reject the whole record, count it); this is the information
- *  that a quiet version would withhold. */
-export interface DroppedStored {
-  at: 'thread' | 'message' | 'part';
-  reason: string;
-  value: unknown;
-}
-
-export interface ParsedThread {
-  messages: ChatMessage[];
-  /** Empty on a clean parse. Never silently shorter output than input. */
-  dropped: DroppedStored[];
-}
-
-function isRecord(v: unknown): v is Record<string, unknown> {
-  return typeof v === 'object' && v !== null;
-}
-
-/**
- * One payload check per `MessagePart` variant. The key set is DERIVED from the
- * union in `src/elements/chat-types.ts` at compile time: `satisfies
- * Record<MessagePart['type'], …>` makes a missing key (a 7th variant landed)
- * and a stale key (a variant renamed away) each a type error in THIS file —
- * the same union `lint:silent-drops` and `verify:scaffold` read, so the lists
- * cannot disagree. The behavioral cross-check lives in ./persistence.test.ts,
- * which re-reads the union by a deliberately different method.
- *
- * These are shape checks, not a trust boundary for rendering: a rehydrated
- * part is model output twice over (it was model output before it was
- * hand-editable storage), and it stays safe to draw because the elements are
- * what draw it. What this buys is that a truncated or half-written record
- * reaches the element as a well-formed thread minus the broken piece, instead
- * of as `parts: undefined` taking the whole app down on first render.
- */
-const PART_VALIDATORS = {
-  text: (v) => typeof v.text === 'string',
-  reasoning: (v) => typeof v.text === 'string',
-  tool: (v) => isRecord(v.tool),
-  card: (v) => isRecord(v.envelope) && typeof v.envelope.type === 'string' && typeof v.envelope.id === 'string',
-  source: (v) => isRecord(v.source),
-  file: (v) => isRecord(v.attachment),
-} satisfies Record<MessagePart['type'], (v: Record<string, unknown>) => boolean>;
-
-function parsePart(v: unknown, dropped: DroppedStored[]): MessagePart | null {
-  if (!isRecord(v) || typeof v.type !== 'string') {
-    dropped.push({ at: 'part', reason: 'not a part record', value: v });
-    return null;
-  }
-  const validate = (PART_VALIDATORS as Record<string, (v: Record<string, unknown>) => boolean>)[v.type];
-  if (!validate) {
-    // An unknown `type` is dropped rather than passed on: every renderer and
-    // encoder switches on it and has no branch for a made-up one.
-    dropped.push({ at: 'part', reason: `unknown part type "${v.type}"`, value: v });
-    return null;
-  }
-  if (!validate(v)) {
-    dropped.push({ at: 'part', reason: `missing or invalid payload for "${v.type}" part`, value: v });
-    return null;
-  }
-  // Verbatim, not rebuilt: optional fields the variants carry (`raw`,
-  // `signature`, `index`, …) are load-bearing on the wire re-encode, so the
-  // validated record passes through whole.
-  return v as unknown as MessagePart;
-}
-
-function parseMessage(v: unknown, dropped: DroppedStored[]): ChatMessage | null {
-  if (!isRecord(v) || typeof v.id !== 'string' || (v.role !== 'user' && v.role !== 'assistant') || !Array.isArray(v.parts)) {
-    dropped.push({ at: 'message', reason: 'not a message record (id/role/parts)', value: v });
-    return null;
-  }
-  const parts = v.parts
-    .map((p) => parsePart(p, dropped))
-    .filter((p): p is MessagePart => p !== null);
-  if (parts.length === 0) {
-    // A message whose every part was unreadable is an empty bubble, and the
-    // per-part entries above already say what was lost.
-    dropped.push({ at: 'message', reason: 'no readable parts survived', value: v });
-    return null;
-  }
-  const message: ChatMessage = { id: v.id, role: v.role, parts };
-  if (v.feedback === 'like' || v.feedback === 'dislike') message.feedback = v.feedback as FeedbackVote;
-  if (isRecord(v.avatar)) message.avatar = v.avatar;
-  if (Array.isArray(v.actions)) message.actions = v.actions as ChatMessage['actions'];
-  return message;
-}
-
-/**
- * Validate a stored thread — the JSON-parsed value of one thread's messages
- * array — back into `ChatMessage[]`. Never throws: unreadable records are
- * DROPPED and reported in `dropped`, so a truncated write loses one record,
- * not every conversation. What to DO about a drop (warn, discard the whole
- * record, telemetry) is the caller's policy.
- */
-export function parseStoredThread(value: unknown): ParsedThread {
-  const dropped: DroppedStored[] = [];
-  if (!Array.isArray(value)) {
-    return { messages: [], dropped: [{ at: 'thread', reason: 'stored thread is not an array', value }] };
-  }
-  const messages = value
-    .map((m) => parseMessage(m, dropped))
-    .filter((m): m is ChatMessage => m !== null);
-  return { messages, dropped };
-}
-
-export interface SaveSchedulerOptions {
-  /** The quiet period before a scheduled snapshot is written. This is policy
-   *  and therefore REQUIRED — the kit ships no default debounce. */
-  delayMs: number;
-  /** The scheduler the seam runs on. Defaults to the global timers; inject
-   *  your own for tests or a non-timer scheduler. */
-  setTimeout?: (fn: () => void, ms: number) => unknown;
-  clearTimeout?: (handle: unknown) => void;
-}
-
-export interface SaveScheduler<T> {
-  /** Remember `snapshot` as the latest state and (re)start the quiet-period
-   *  timer. Call it on every state change; only the last snapshot inside the
-   *  quiet period is written. */
-  schedule(snapshot: T): void;
-  /** Write the pending snapshot NOW and disarm the timer — the
-   *  `beforeunload`/`visibilitychange` story, so a tab closed mid-stream still
-   *  persists. No-op when nothing is pending: the last write already happened. */
-  flush(): void;
-  /** Drop the pending write unsaved (e.g. unmount-without-persist). */
-  cancel(): void;
-  pending(): boolean;
-}
-
-/**
- * The debounce/flush seam: a streaming reply lands a state update per token,
- * and without this shape so would the consumer's storage backend. The kit owns
- * only the timer dance; `save` (WHAT and WHERE, including its own try/catch
- * policy for quota errors) and `delayMs` (HOW LONG) are the consumer's.
- */
-export function createSaveScheduler<T>(save: (snapshot: T) => void, opts: SaveSchedulerOptions): SaveScheduler<T> {
-  const setTimer = opts.setTimeout ?? ((fn: () => void, ms: number) => setTimeout(fn, ms));
-  const clearTimer = opts.clearTimeout ?? ((handle: unknown) => clearTimeout(handle as ReturnType<typeof setTimeout>));
-
-  let latest: T | undefined;
-  let armed = false;
-  let handle: unknown;
-
-  const disarm = () => {
-    if (!armed) return;
-    clearTimer(handle);
-    armed = false;
-  };
-  const write = () => {
-    armed = false;
-    save(latest as T);
-  };
-
-  return {
-    schedule(snapshot) {
-      latest = snapshot;
-      disarm();
-      handle = setTimer(write, opts.delayMs);
-      armed = true;
-    },
-    flush() {
-      if (!armed) return;
-      disarm();
-      save(latest as T);
-    },
-    cancel: disarm,
-    pending: () => armed,
-  };
-}