Compare commits
2 Commits
| Author | SHA1 | Date | |
|---|---|---|---|
 agent_coder
|
4c1ee50dc9 | ||
|
agent_coder
|
bfcee6dddc |
@@ -294,7 +294,7 @@ Vite SPA. Code is organized by feature under `apps/client/src/features/*` (mirro
|
||||
- **Errors must never be swallowed or shown as generic messages.** Every caught error MUST (1) be logged in full to the console/logger — error name, message, stack, `cause`, and (for HTTP/provider failures) the status code and response body — and (2) be surfaced to the user with a *specific, human-readable explanation of what actually went wrong*, never a bare generic string like "Something went wrong" / "Could not start recording" / "Transcription failed". Include the real reason (the underlying error/provider message) in the user-facing text. On the server, wrap third-party/provider failures with `describeProviderError` (or equivalent) and rethrow as a meaningful HTTP status + message — never let them collapse into an opaque 500. On the client, `console.error(<context>, err)` the raw error AND show the extracted reason (e.g. `err.response?.data?.message`, or the error `name: message`) in the notification.
|
||||
- The version string shown in the UI comes from `APP_VERSION` (CI/Docker) or `git describe --tags --always` (local), resolved in `vite.config.ts` — not from `package.json`.
|
||||
- Server TS config is permissive (`noImplicitAny: false`, `strictNullChecks: false`, `no-explicit-any` lint disabled). Follow the existing relaxed style rather than tightening types broadly.
|
||||
- Dependency versions are heavily pinned via `pnpm.overrides` and `pnpm.patchedDependencies` (`scimmy`, `yjs`, `ai`) in the root `package.json`. Don't bump pinned/patched deps casually; the patches and overrides exist for compatibility/security reasons. The `ai@6.0.134` patch disables the SDK's O(n²) cumulative `partialOutput` accumulation when no output strategy is requested (server heap OOM on long agent runs, #184; tripwire test: `apps/server/src/integrations/ai/ai-sdk-partial-output.patch.spec.ts`) — it MUST be re-created via `pnpm patch` when bumping `ai`.
|
||||
- Dependency versions are heavily pinned via `pnpm.overrides` and `pnpm.patchedDependencies` (`scimmy`, `yjs`) in the root `package.json`. Don't bump pinned/patched deps casually; the patches and overrides exist for compatibility/security reasons.
|
||||
- **Adding/renaming/removing an MCP tool requires updating `SERVER_INSTRUCTIONS`** in `packages/mcp/src/index.ts` — the intent-routing guide MCP clients receive on initialize. This applies both to inline `server.registerTool(...)` calls in `index.ts` and to specs in `packages/mcp/src/tool-specs.ts`. Enforced by `packages/mcp/test/unit/server-instructions.test.mjs`, which fails when a registered tool is not mentioned in the guide (deliberate opt-outs go into its `EXCEPTIONS` list). `packages/mcp/build/` is gitignored and rebuilt in CI/Docker via `pnpm build` (same convention as `git-sync`/`prosemirror-markdown`) — never commit it; rebuild locally after editing to run the tests.
|
||||
|
||||
## CI / release
|
||||
|
||||
@@ -169,14 +169,6 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
|
||||
|
||||
### Fixed
|
||||
|
||||
- **The server no longer runs out of heap during long autonomous agent runs.** A
|
||||
new pnpm patch on `ai@6.0.134` stops the SDK from building a cumulative
|
||||
snapshot of the ENTIRE turn text on every streamed text-delta when no output
|
||||
strategy was requested (our server never requests one). Unpatched, those
|
||||
O(n²) `partialOutput` snapshots piled up in a never-consumed internal
|
||||
`tee()` branch of the stream result — a ~20-step, ~28k-chunk agent run
|
||||
retained ~1.7 GB and OOM'd the 2 GB JS heap. Streaming granularity is
|
||||
unchanged; the patch must be re-created if `ai` is ever bumped. (#184)
|
||||
- **Internal links in exported Markdown no longer lose their visible text.** A
|
||||
link whose target page name had no file extension (e.g. a bare title) was
|
||||
collapsed to empty text during export, producing an unclickable, label-less
|
||||
|
||||
@@ -1,92 +0,0 @@
|
||||
import { readFileSync } from 'fs';
|
||||
import { streamText } from 'ai';
|
||||
import { MockLanguageModelV3, simulateReadableStream } from 'ai/test';
|
||||
|
||||
/**
|
||||
* Regression tests for patches/ai@6.0.134.patch (server heap OOM on long
|
||||
* autonomous agent runs, #184).
|
||||
*
|
||||
* Unpatched ai@6.0.134 substitutes the default text() output strategy even
|
||||
* when the caller passes NO `output` option. Its createOutputTransformStream
|
||||
* then accumulates the ENTIRE turn text and, on EVERY text-delta, enqueues a
|
||||
* flat snapshot of all text so far as `partialOutput` (O(n^2) memory). Those
|
||||
* snapshots pile up in the never-consumed leftover tee() branch of
|
||||
* DefaultStreamTextResult.baseStream, which is what OOM'd production during a
|
||||
* ~28k-chunk agent turn. The pnpm patch skips partialOutput production
|
||||
* entirely when no output strategy was requested, while keeping per-delta
|
||||
* streaming granularity.
|
||||
*/
|
||||
describe('ai@6.0.134 pnpm patch: no partialOutput accumulation without an output strategy', () => {
|
||||
const makeModel = () =>
|
||||
new MockLanguageModelV3({
|
||||
doStream: async () => ({
|
||||
stream: simulateReadableStream({
|
||||
chunks: [
|
||||
{ type: 'stream-start' as const, warnings: [] },
|
||||
{ type: 'text-start' as const, id: '1' },
|
||||
{ type: 'text-delta' as const, id: '1', delta: 'Hello' },
|
||||
{ type: 'text-delta' as const, id: '1', delta: ', ' },
|
||||
{ type: 'text-delta' as const, id: '1', delta: 'world!' },
|
||||
{ type: 'text-end' as const, id: '1' },
|
||||
{
|
||||
type: 'finish' as const,
|
||||
finishReason: { unified: 'stop' as const, raw: 'stop' },
|
||||
usage: {
|
||||
inputTokens: {
|
||||
total: 1,
|
||||
noCache: undefined,
|
||||
cacheRead: undefined,
|
||||
cacheWrite: undefined,
|
||||
},
|
||||
outputTokens: { total: 1, text: 1, reasoning: undefined },
|
||||
},
|
||||
},
|
||||
],
|
||||
}),
|
||||
}),
|
||||
});
|
||||
|
||||
it('preserves per-delta streaming granularity in textStream', async () => {
|
||||
const result = streamText({ model: makeModel(), prompt: 'hi' });
|
||||
|
||||
const deltas: string[] = [];
|
||||
for await (const delta of result.textStream) {
|
||||
deltas.push(delta);
|
||||
}
|
||||
|
||||
// The patch must NOT coalesce or drop deltas: three model deltas arrive
|
||||
// as three separate textStream chunks.
|
||||
expect(deltas).toEqual(['Hello', ', ', 'world!']);
|
||||
});
|
||||
|
||||
it('emits NO partialOutput values when the caller did not request an output strategy', async () => {
|
||||
const result = streamText({ model: makeModel(), prompt: 'hi' });
|
||||
|
||||
// Fully consume the primary stream first (mirrors production usage).
|
||||
for await (const _ of result.textStream) {
|
||||
// drain
|
||||
}
|
||||
|
||||
const partials: unknown[] = [];
|
||||
for await (const partial of result.experimental_partialOutputStream) {
|
||||
partials.push(partial);
|
||||
}
|
||||
|
||||
// TRIPWIRE: on unpatched ai@6.0.134 the default text() output strategy
|
||||
// yields one cumulative partial per text-delta here (['Hello', 'Hello, ',
|
||||
// 'Hello, world!']). An empty stream proves the patch is applied and no
|
||||
// cumulative snapshots are being produced (and thus none can pile up in
|
||||
// the leftover internal tee branch).
|
||||
expect(partials).toEqual([]);
|
||||
});
|
||||
|
||||
it('both installed dist builds (CJS and ESM) carry the patch marker', () => {
|
||||
// Secondary guard: pins the patch to BOTH bundles the SDK ships, since
|
||||
// the NestJS server consumes CJS while other tooling may load ESM.
|
||||
const cjsPath = require.resolve('ai');
|
||||
const mjsPath = cjsPath.replace(/index\.js$/, 'index.mjs');
|
||||
expect(cjsPath).toMatch(/index\.js$/);
|
||||
expect(readFileSync(cjsPath, 'utf8')).toContain('PATCH(docmost');
|
||||
expect(readFileSync(mjsPath, 'utf8')).toContain('PATCH(docmost');
|
||||
});
|
||||
});
|
||||
+1
-2
@@ -96,8 +96,7 @@
|
||||
"pnpm": {
|
||||
"patchedDependencies": {
|
||||
"scimmy@1.3.5": "patches/scimmy@1.3.5.patch",
|
||||
"yjs@13.6.30": "patches/yjs@13.6.30.patch",
|
||||
"ai@6.0.134": "patches/ai@6.0.134.patch"
|
||||
"yjs@13.6.30": "patches/yjs@13.6.30.patch"
|
||||
},
|
||||
"overrides": {
|
||||
"prosemirror-changeset": "2.4.0",
|
||||
|
||||
Vendored
+16
@@ -0,0 +1,16 @@
|
||||
{
|
||||
"_bug": "BUG #351: a `column` whose `width` is a percentage string (e.g. \"50%\") is NOT byte-stable across export->import->export (violates P2). The `column` schema's parseHTML does `parseFloat(getAttribute('data-width'))`, which silently drops the '%' unit and returns the NUMBER 50. So the first export emits data-width=\"50%\" but the re-import stores width=50, and the second export emits data-width=\"50\": md2 !== md1, a permanent GS-EDIT-REVERT churn (every git-sync pull rewrites the column width). The editor authors column widths as percentages, so this is a real data/round-trip defect. Fix belongs in src/lib/docmost-schema.ts column.width parseHTML (preserve the unit / keep the string), which is OUT OF SCOPE for this test-only PR and must be a separate, maintainer-approved change. This flat generator therefore keeps `column.width` frozen (never generates a non-default width).",
|
||||
"doc": {
|
||||
"type": "doc",
|
||||
"content": [
|
||||
{
|
||||
"type": "columns",
|
||||
"attrs": { "layout": "two_equal", "widthMode": "normal" },
|
||||
"content": [
|
||||
{ "type": "column", "attrs": { "width": "50%" }, "content": [{ "type": "paragraph", "content": [{ "type": "text", "text": "L" }] }] },
|
||||
{ "type": "column", "attrs": { "width": "50%" }, "content": [{ "type": "paragraph", "content": [{ "type": "text", "text": "R" }] }] }
|
||||
]
|
||||
}
|
||||
]
|
||||
}
|
||||
}
|
||||
+19
@@ -0,0 +1,19 @@
|
||||
{
|
||||
"doc": {
|
||||
"type": "doc",
|
||||
"content": [
|
||||
{
|
||||
"type": "orderedList",
|
||||
"attrs": { "type": null, "start": 5 },
|
||||
"content": [
|
||||
{
|
||||
"type": "listItem",
|
||||
"content": [
|
||||
{ "type": "paragraph", "content": [{ "type": "text", "text": "alpha" }] }
|
||||
]
|
||||
}
|
||||
]
|
||||
}
|
||||
]
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,325 @@
|
||||
/**
|
||||
* Schema-DERIVED attribute-state fast-check arbitraries (#351, PR 1).
|
||||
*
|
||||
* This GENERALIZES the #350 stability-matrix helper (roundtrip-stability.helper.ts)
|
||||
* to fast-check. Where that helper sweeps a HAND-WRITTEN 2-state matrix for one
|
||||
* node spec, this module reads the attribute list straight from
|
||||
* `schema.nodes[type].spec.attrs` (never a hand list) and, per attribute,
|
||||
* generates over the FOUR states the issue calls for:
|
||||
*
|
||||
* - `absent` : the attribute is OMITTED entirely (the empty-string-vs-
|
||||
* absent churn class the #350 fix targets).
|
||||
* - `default` : the schema default value, authored explicitly.
|
||||
* - `nonDefault` : a representative legal non-default value.
|
||||
* - `degenerate` : `""` for strings, `0`/negative for numbers, the flipped
|
||||
* value for booleans.
|
||||
*
|
||||
* ── Why a per-attribute override table ──────────────────────────────────────
|
||||
* Everything that CAN be derived generically from the default's runtime type is
|
||||
* (booleans flip; the degenerate value follows the runtime type). But two facts
|
||||
* force a small, DOCUMENTED override table:
|
||||
*
|
||||
* 1. CONSTRAINED domains the schema does not encode. `image.align ∈
|
||||
* {left,center,right}`, `heading.level ∈ 1..6`, `callout.type ∈
|
||||
* {info,success,warning,danger}`, `columns.layout`, table-cell `align`,
|
||||
* `status.color`, `orderedList.start ≥ 1`, etc. A generic "default + 1"
|
||||
* would emit an ILLEGAL value, so these get an explicit legal domain.
|
||||
* 2. ROUND-TRIP-safety, established EMPIRICALLY by probing the live converter
|
||||
* (the classification captured in flat-roundtrip.property.test.ts). A frozen
|
||||
* attribute falls into ONE of TWO explicitly-distinguished classes — never a
|
||||
* silent "it just doesn't round-trip":
|
||||
*
|
||||
* (a) ACCEPTED LIMITATION — the attribute has NO markdown representation,
|
||||
* so the loss is inherent to targeting markdown, not a converter
|
||||
* defect. These: `paragraph`/`heading` `indent`, `callout.icon`,
|
||||
* `orderedList.type` (a/A/i markers), table `colwidth` /
|
||||
* `backgroundColor(Name)` (dropped by the raw-<table> fallback). Each is
|
||||
* tagged `// ACCEPTED:` inline. Freezing them is correct — there is
|
||||
* nothing to preserve in the target format.
|
||||
*
|
||||
* (b) PINNED BUG — the attribute IS representable in markdown but the
|
||||
* converter drops it anyway (a real defect). These are NOT silently
|
||||
* frozen: each is captured as a LOUD `it.fails` counterexample in
|
||||
* test/fixtures/counterexamples/ + counterexamples.test.ts, and the
|
||||
* freeze here only keeps the P1/P2 union green until a MAINTAINER rules
|
||||
* on accept-vs-fix (the epic guardrail reserves that call). These:
|
||||
* `column.width` (parseFloat drops `%`), `orderedList.start` (non-1
|
||||
* start renders as `1.`). Tagged `// PINNED-BUG:` inline.
|
||||
*
|
||||
* (c) DEFERRED-BUG — representable AND round-trips, frozen only because the
|
||||
* flat generator can't yet build a valid instance. Table
|
||||
* `colspan`/`rowspan` round-trip via the raw-<table> fallback, but a
|
||||
* geometrically-valid spanned table is PR-2 structural work; the flat
|
||||
* generator hardcodes span = 1. Tagged `// DEFERRED-BUG:` inline so a
|
||||
* maintainer does not read them as an inherent limitation.
|
||||
* - Several non-null-default attrs are MATERIALIZED on import but are not
|
||||
* in canonicalize's KNOWN_DEFAULTS (`callout.type`, `status.color`,
|
||||
* table `colspan`/`rowspan`, `columns.layout`/`widthMode`,
|
||||
* `embed.width`/`height`, `heading.level`, `taskItem.checked`,
|
||||
* `details.open`, `subpages.recursive`, `orderedList.start`). If left
|
||||
* `absent` they re-materialize as a non-canonical default and diverge
|
||||
* under P1. We mark them `always` so they are authored explicitly.
|
||||
* - The documented numeric→string coercion set (`width height size
|
||||
* aspectRatio`) is generated as STRINGS for the media family (a stored
|
||||
* number re-parses as a string), EXCEPT `embed.width/height` which the
|
||||
* embed schema keeps numeric — handled per-attr.
|
||||
*
|
||||
* Both PINNED-BUG attrs (`column.width` P2 churn, `orderedList.start` P1 loss)
|
||||
* are captured as committed `it.fails` counterexamples — NOT hidden here.
|
||||
*/
|
||||
import fc from 'fast-check';
|
||||
import { getSchema } from '@tiptap/core';
|
||||
import { docmostExtensions } from '../../src/lib/index.js';
|
||||
import { phraseArb, letterPhraseArb, urlArb } from './text-arbitraries.js';
|
||||
|
||||
/** The exact ProseMirror schema the converter targets. */
|
||||
export const schema = getSchema(docmostExtensions as any);
|
||||
|
||||
/** Sentinel: this attribute is OMITTED (the `absent` state). */
|
||||
export const ABSENT = Symbol('ABSENT');
|
||||
|
||||
/** The documented numeric→string coercion set (issue + roundtrip-stability.helper). */
|
||||
export const NUMERIC_STRING_ATTRS = ['width', 'height', 'size', 'aspectRatio'];
|
||||
|
||||
/** Read the schema default for every attribute of a node type. */
|
||||
export function schemaAttrDefaults(type: string): Record<string, unknown> {
|
||||
const specAttrs = (schema.nodes[type]?.spec?.attrs ?? {}) as Record<
|
||||
string,
|
||||
{ default: unknown }
|
||||
>;
|
||||
const out: Record<string, unknown> = {};
|
||||
for (const [k, v] of Object.entries(specAttrs)) out[k] = v.default;
|
||||
return out;
|
||||
}
|
||||
|
||||
/** Attribute names for a node type, straight from the schema (never hand-listed). */
|
||||
export function schemaAttrNames(type: string): string[] {
|
||||
return Object.keys((schema.nodes[type]?.spec?.attrs ?? {}) as object);
|
||||
}
|
||||
|
||||
/**
|
||||
* Per-attribute policy. Everything unlisted falls back to a generic policy:
|
||||
* - a BOOLEAN default is fuzzable (its non-default is the flipped value);
|
||||
* - any other default is `frozen` (only `absent`/`default` are generated) so
|
||||
* we never invent an unverified non-default that might not round-trip.
|
||||
* Listed attrs override this with a legal `arb` domain and/or flags.
|
||||
*/
|
||||
interface AttrPolicy {
|
||||
/** Arbitrary for the `nonDefault` state's value. */
|
||||
arb?: fc.Arbitrary<unknown>;
|
||||
/** Value for the `degenerate` state (fuzz mode only). Omit to skip degenerate. */
|
||||
degen?: unknown;
|
||||
/** Never emit `absent` — the attr must be authored (materialized default class). */
|
||||
always?: boolean;
|
||||
/** Never emit the schema default value (required-ish attrs like `src`). Implies always. */
|
||||
noDefault?: boolean;
|
||||
/** Never emit non-default/degenerate — attr has no md representation or churns. */
|
||||
frozen?: boolean;
|
||||
}
|
||||
|
||||
const num = (...xs: number[]) => fc.constantFrom(...xs);
|
||||
const str = (...xs: string[]) => fc.constantFrom(...xs);
|
||||
const widthStr = str('120', '320', '640');
|
||||
|
||||
// The documented override table, keyed `type.attr`. Every entry is grounded in
|
||||
// the empirical converter probe (see flat-roundtrip.property.test.ts header).
|
||||
const OVERRIDES: Record<string, AttrPolicy> = {
|
||||
// ── block text containers ────────────────────────────────────────────────
|
||||
// 'left' is the IMPLICIT default alignment: the converter drops it on export
|
||||
// (empirically confirmed), so it never round-trips. Only center/right/justify
|
||||
// carry through the `<!--attrs {textAlign}-->` comment.
|
||||
'paragraph.textAlign': { arb: str('center', 'right', 'justify') },
|
||||
'paragraph.indent': { frozen: true }, // ACCEPTED: no md representation
|
||||
'heading.level': { always: true, arb: num(2, 3, 4, 5, 6) },
|
||||
'heading.textAlign': { arb: str('center', 'right', 'justify') },
|
||||
'heading.indent': { frozen: true }, // ACCEPTED: no md representation
|
||||
// ── lists ────────────────────────────────────────────────────────────────
|
||||
// PINNED-BUG: markdown CAN express a non-1 start ("5."), but the converter
|
||||
// renders "1." and drops it -> P1 loss. See counterexamples.test.ts
|
||||
// (ordered-list-start.json). Frozen only until the maintainer rules accept-vs-fix.
|
||||
'orderedList.start': { always: true, frozen: true },
|
||||
'orderedList.type': { frozen: true }, // ACCEPTED: a/A/i markers not expressible in GFM
|
||||
'taskItem.checked': { always: true, arb: fc.constant(true) }, // boolean, default false
|
||||
// ── codeBlock ────────────────────────────────────────────────────────────
|
||||
'codeBlock.language': { arb: str('js', 'ts', 'python', 'go', 'rust', 'bash') },
|
||||
// ── image / media (numeric→string width family) ──────────────────────────
|
||||
'image.src': { noDefault: true, arb: urlArb, degen: '' },
|
||||
'image.align': { arb: str('left', 'right') },
|
||||
'image.alt': { arb: letterPhraseArb, degen: '' },
|
||||
'image.title': { arb: letterPhraseArb },
|
||||
'image.width': { arb: widthStr, degen: '' },
|
||||
'image.height': { arb: widthStr, degen: '' },
|
||||
'video.src': { noDefault: true, arb: urlArb, degen: '' },
|
||||
'video.alt': { arb: letterPhraseArb },
|
||||
'video.width': { arb: widthStr },
|
||||
'video.height': { arb: widthStr },
|
||||
'audio.src': { noDefault: true, arb: urlArb, degen: '' },
|
||||
'youtube.src': { noDefault: true, arb: urlArb },
|
||||
'pdf.src': { noDefault: true, arb: urlArb },
|
||||
'pdf.name': { arb: phraseArb },
|
||||
'drawio.src': { noDefault: true, arb: urlArb },
|
||||
'excalidraw.src': { noDefault: true, arb: urlArb },
|
||||
'attachment.url': { noDefault: true, arb: urlArb },
|
||||
'attachment.name': { arb: phraseArb },
|
||||
// ── callout / status ─────────────────────────────────────────────────────
|
||||
'callout.type': { always: true, arb: str('success', 'warning', 'danger') },
|
||||
'callout.icon': { frozen: true }, // ACCEPTED: no md representation (dropped on export)
|
||||
'status.text': { noDefault: true, arb: phraseArb, degen: '' },
|
||||
'status.color': { always: true, arb: str('green', 'orange', 'red', 'blue', 'yellow', 'purple') },
|
||||
// ── table cells ────────────────────────────────────────────────────────────
|
||||
// DEFERRED-BUG (not ACCEPTED): colspan/rowspan ARE representable and round-trip
|
||||
// — a spanned cell makes the converter emit the whole table as a raw <table>
|
||||
// with colspan/rowspan attrs (markdown-converter.ts tableToHtml), which the
|
||||
// tiptap parser reads back. They are frozen only because generating a
|
||||
// geometrically-valid spanned table is deferred STRUCTURAL work (the flat
|
||||
// generator hardcodes colspan/rowspan = 1), NOT a markdown limitation.
|
||||
'tableCell.colspan': { always: true, frozen: true },
|
||||
'tableCell.rowspan': { always: true, frozen: true },
|
||||
// ACCEPTED: colwidth / backgroundColor(Name) have no representation — the
|
||||
// raw-<table> fallback (tableToHtml) drops them, so there is nothing to preserve.
|
||||
'tableCell.colwidth': { frozen: true },
|
||||
'tableCell.backgroundColor': { frozen: true },
|
||||
'tableCell.backgroundColorName': { frozen: true },
|
||||
'tableCell.align': { arb: str('left', 'center', 'right') },
|
||||
'tableHeader.colspan': { always: true, frozen: true }, // DEFERRED-BUG (see tableCell.colspan)
|
||||
'tableHeader.rowspan': { always: true, frozen: true }, // DEFERRED-BUG (see tableCell.rowspan)
|
||||
'tableHeader.colwidth': { frozen: true }, // ACCEPTED: no representation
|
||||
'tableHeader.backgroundColor': { frozen: true }, // ACCEPTED: no representation
|
||||
'tableHeader.backgroundColorName': { frozen: true }, // ACCEPTED: no representation
|
||||
'tableHeader.align': { arb: str('left', 'center', 'right') },
|
||||
// ── details ──────────────────────────────────────────────────────────────
|
||||
'details.open': { always: true, arb: fc.constant(true) }, // boolean, default false
|
||||
// ── columns ──────────────────────────────────────────────────────────────
|
||||
'columns.layout': { always: true, arb: str('three_equal', 'left_sidebar', 'right_sidebar') },
|
||||
// widthMode round-trips via the `data-width-mode` attribute (verified P1+P2),
|
||||
// so it is fuzzed, not frozen.
|
||||
'columns.widthMode': { always: true, arb: str('custom') },
|
||||
// PINNED-BUG: parseFloat import drops the `%` unit -> P2 churn. See
|
||||
// counterexamples.test.ts (columns-column-width-percent.json).
|
||||
'column.width': { frozen: true },
|
||||
// ── embed (schema keeps width/height NUMERIC, not string-coerced) ─────────
|
||||
'embed.src': { noDefault: true, arb: urlArb, degen: '' },
|
||||
'embed.provider': { noDefault: true, arb: str('iframe', 'youtube', 'vimeo') },
|
||||
'embed.width': { always: true, frozen: true },
|
||||
'embed.height': { always: true, frozen: true },
|
||||
// ── subpages / math / htmlEmbed ──────────────────────────────────────────
|
||||
'subpages.recursive': { always: true, arb: fc.constant(true) }, // boolean, default false
|
||||
'mathBlock.text': { noDefault: true, arb: str('x^2', 'a < b', '\\frac{1}{2}'), degen: '' },
|
||||
'mathInline.text': { noDefault: true, arb: str('x^2', 'a < b', '\\frac{1}{2}'), degen: '' },
|
||||
'htmlEmbed.source': { noDefault: true, arb: str('<b>hi</b>', '<i>x</i>', '<span>y</span>'), degen: '' },
|
||||
'htmlEmbed.height': { arb: num(200, 300, 400) },
|
||||
// ── footnotes / transclusion / pageEmbed / mention ───────────────────────
|
||||
'footnoteDefinition.id': { noDefault: true, arb: str('fn1', 'fn2', 'note') },
|
||||
'footnoteReference.id': { noDefault: true, arb: str('fn1', 'fn2', 'note') },
|
||||
'pageEmbed.sourcePageId': { noDefault: true, arb: fc.uuid() },
|
||||
'transclusionSource.id': { noDefault: true, arb: str('src1', 'src2') },
|
||||
'transclusionReference.sourcePageId': { noDefault: true, arb: fc.uuid() },
|
||||
'transclusionReference.transclusionId': { noDefault: true, arb: str('tr1', 'tr2') },
|
||||
'mention.id': { noDefault: true, arb: fc.uuid() },
|
||||
'mention.label': { noDefault: true, arb: phraseArb },
|
||||
'mention.entityType': { noDefault: true, arb: str('user') },
|
||||
'mention.entityId': { noDefault: true, arb: fc.uuid() },
|
||||
};
|
||||
|
||||
/** Resolve the effective policy for one attribute (override merged over generic). */
|
||||
function policyFor(type: string, attr: string, def: unknown): AttrPolicy {
|
||||
const override = OVERRIDES[`${type}.${attr}`];
|
||||
if (override) return override;
|
||||
// Generic: booleans are fuzzable via their flipped value; everything else is
|
||||
// frozen (only absent/default) so no unverified non-default is invented.
|
||||
if (typeof def === 'boolean') return { arb: fc.constant(!def) };
|
||||
return { frozen: true };
|
||||
}
|
||||
|
||||
/**
|
||||
* Whether an attribute is actually exercised at a NON-DEFAULT value (i.e. its
|
||||
* policy has an `arb`, which the generic fallback does not). Used by the
|
||||
* attribute-coverage snapshot test to make the generic-frozen space VISIBLE: any
|
||||
* string/number attr not in OVERRIDES is silently only tested at absent/default,
|
||||
* so the snapshot pins exactly which attrs are NOT value-fuzzed and forces a
|
||||
* reviewer to look when a new attr lands in that invisible bucket.
|
||||
*/
|
||||
export function attrIsValueFuzzed(type: string, attr: string): boolean {
|
||||
const def = schemaAttrDefaults(type)[attr];
|
||||
return !!policyFor(type, attr, def).arb;
|
||||
}
|
||||
|
||||
/** Every node `type.attr` in the schema (excluding the auto `id`), sorted. */
|
||||
export function allSchemaAttrKeys(): string[] {
|
||||
const keys: string[] = [];
|
||||
for (const type of Object.keys(schema.nodes)) {
|
||||
for (const attr of schemaAttrNames(type)) {
|
||||
if (attr === 'id') continue;
|
||||
keys.push(`${type}.${attr}`);
|
||||
}
|
||||
}
|
||||
return keys.sort();
|
||||
}
|
||||
|
||||
/**
|
||||
* Every MARK attribute in the schema, keyed `mark:<name>.<attr>`, sorted. Marks
|
||||
* are not driven by the node OVERRIDES table (they are fuzzed by the text
|
||||
* generator, text-arbitraries.ts), so their value-fuzz coverage is tracked with a
|
||||
* separate snapshot (see flat-roundtrip.property.test.ts) — without this the
|
||||
* "no invisible coverage hole" guarantee would hold for node attrs only, letting a
|
||||
* new mark attr slip through unfuzzed and unallowlisted.
|
||||
*/
|
||||
export function allSchemaMarkAttrKeys(): string[] {
|
||||
const keys: string[] = [];
|
||||
for (const [name, mark] of Object.entries(schema.marks)) {
|
||||
const attrs = (mark.spec?.attrs ?? {}) as Record<string, unknown>;
|
||||
for (const attr of Object.keys(attrs)) keys.push(`mark:${name}.${attr}`);
|
||||
}
|
||||
return keys.sort();
|
||||
}
|
||||
|
||||
export type AttrMode = 'p1' | 'fuzz';
|
||||
|
||||
/**
|
||||
* Build an arbitrary for ONE attribute's value (or the ABSENT sentinel) across
|
||||
* the states legal for `mode`:
|
||||
* - p1 : absent / default / nonDefault (the round-trip-safe space).
|
||||
* - fuzz : the above PLUS degenerate (P2 tolerates the one-time
|
||||
* normalization; P3 only needs totality).
|
||||
*/
|
||||
export function attrValueArb(
|
||||
type: string,
|
||||
attr: string,
|
||||
mode: AttrMode,
|
||||
): fc.Arbitrary<unknown | typeof ABSENT> {
|
||||
const def = schemaAttrDefaults(type)[attr];
|
||||
const p = policyFor(type, attr, def);
|
||||
|
||||
const states: fc.Arbitrary<unknown | typeof ABSENT>[] = [];
|
||||
if (!p.always && !p.noDefault) states.push(fc.constant(ABSENT));
|
||||
if (!p.noDefault) states.push(fc.constant(def));
|
||||
if (!p.frozen && p.arb) states.push(p.arb);
|
||||
if (mode === 'fuzz' && !p.frozen && p.degen !== undefined) {
|
||||
states.push(fc.constant(p.degen));
|
||||
}
|
||||
if (states.length === 0) states.push(fc.constant(def));
|
||||
return fc.oneof(...states);
|
||||
}
|
||||
|
||||
/**
|
||||
* Build an arbitrary for a node's full `attrs` object over all schema attrs.
|
||||
* `base` pins caller-required attrs (e.g. a concrete `src`) verbatim; any attr
|
||||
* present in `base` is NOT re-generated. Omitted (ABSENT) attrs are dropped.
|
||||
*/
|
||||
export function nodeAttrsArb(
|
||||
type: string,
|
||||
mode: AttrMode,
|
||||
base: Record<string, unknown> = {},
|
||||
): fc.Arbitrary<Record<string, unknown>> {
|
||||
const names = schemaAttrNames(type).filter((n) => !(n in base) && n !== 'id');
|
||||
if (names.length === 0) return fc.constant({ ...base });
|
||||
return fc
|
||||
.tuple(...names.map((n) => attrValueArb(type, n, mode)))
|
||||
.map((vals) => {
|
||||
const attrs: Record<string, unknown> = { ...base };
|
||||
names.forEach((n, i) => {
|
||||
if (vals[i] !== ABSENT) attrs[n] = vals[i];
|
||||
});
|
||||
return attrs;
|
||||
});
|
||||
}
|
||||
@@ -0,0 +1,74 @@
|
||||
import { describe, expect, it } from 'vitest';
|
||||
import { readFileSync } from 'node:fs';
|
||||
import { fileURLToPath } from 'node:url';
|
||||
import path from 'node:path';
|
||||
import { convertProseMirrorToMarkdown } from '../../src/lib/markdown-converter.js';
|
||||
import { markdownToProseMirror } from '../../src/lib/markdown-to-prosemirror.js';
|
||||
import { docsCanonicallyEqual } from '../../src/lib/canonicalize.js';
|
||||
|
||||
// ---------------------------------------------------------------------------
|
||||
// #351 committed counterexamples — REAL round-trip bugs surfaced by the flat
|
||||
// generative probing (attribute level). Each is pinned here as an `it.fails`
|
||||
// (vitest passes ONLY WHILE the assertion still fails), so that the day the
|
||||
// underlying src/ bug is fixed, the `it.fails` starts PASSING and vitest turns
|
||||
// this test RED — forcing us to delete the counterexample and (per the epic
|
||||
// guardrail) tighten the generator. A bare `it.fails` would ship silent
|
||||
// corruption, so every case below carries a loud `// BUG #351:` explanation.
|
||||
//
|
||||
// These bugs are NOT worked around by weakening any property: the offending
|
||||
// attribute is kept OUT of the P1/P2 generators (documented in
|
||||
// attr-arbitraries.ts), and the exact failing document lives here as the
|
||||
// regression pin. FIXING the bug is a separate, maintainer-approved src/ change.
|
||||
// ---------------------------------------------------------------------------
|
||||
|
||||
const here = path.dirname(fileURLToPath(import.meta.url));
|
||||
const fixtureDir = path.resolve(here, '../fixtures/counterexamples');
|
||||
|
||||
function loadDoc(file: string): any {
|
||||
return JSON.parse(readFileSync(path.join(fixtureDir, file), 'utf8')).doc;
|
||||
}
|
||||
|
||||
describe('#351 counterexamples (known round-trip bugs, pinned as it.fails)', () => {
|
||||
// BUG #351: a `column` with a PERCENTAGE width ("50%") is not byte-stable.
|
||||
// The column schema parses `data-width` with parseFloat, dropping the '%':
|
||||
// md1 = '...data-width="50%"...' (first export)
|
||||
// re-import stores width = 50 (number)
|
||||
// md2 = '...data-width="50"...' (second export) => md2 !== md1
|
||||
// A permanent GS-EDIT-REVERT churn on every git-sync pull. The editor stores
|
||||
// column widths as percentages, so this is a genuine defect. The fix is in
|
||||
// src/lib/docmost-schema.ts (column.width parseHTML must preserve the unit)
|
||||
// and is out of scope for this test-only PR.
|
||||
it.fails('column percentage width is byte-stable (P2)', async () => {
|
||||
const doc = loadDoc('columns-column-width-percent.json');
|
||||
const md1 = convertProseMirrorToMarkdown(doc);
|
||||
const doc2 = await markdownToProseMirror(md1);
|
||||
const md2 = convertProseMirrorToMarkdown(doc2);
|
||||
// This assertion currently FAILS (md2 drops the '%'), which is exactly what
|
||||
// `it.fails` expects. When the schema is fixed, it will PASS and flip this
|
||||
// test red — our cue to remove the pin.
|
||||
expect(md2).toBe(md1);
|
||||
});
|
||||
|
||||
// BUG #351: an `orderedList` with a non-1 `start` loses its start number.
|
||||
// CommonMark CAN express this ("5." starts the list at 5), but the converter
|
||||
// always emits "1." and ignores `attrs.start` (markdown-converter.ts renders
|
||||
// `${index + 1}.`; the <ol> HTML path also omits `start`):
|
||||
// doc.start = 5 -> md1 = "1. alpha" (start dropped on export)
|
||||
// re-import stores start = 1 => docsCanonicallyEqual(rt, doc) === false
|
||||
// This is a P1 (semantic round-trip) loss of the SAME class as column.width:
|
||||
// representable in markdown, silently dropped by the converter. It is pinned
|
||||
// here as the LOUD counterexample rather than being masked as an "accepted
|
||||
// normalization" in the generator — per the epic guardrail, deciding
|
||||
// accept-vs-fix for a markdown-representable loss is a MAINTAINER call, so this
|
||||
// stays a visible known-bug until the maintainer rules on it. The fix would be
|
||||
// in src/lib/markdown-converter.ts (emit the start number on the first item)
|
||||
// and is out of scope for this test-only PR.
|
||||
it.fails('ordered list start number is preserved (P1)', async () => {
|
||||
const doc = loadDoc('ordered-list-start.json');
|
||||
const md1 = convertProseMirrorToMarkdown(doc);
|
||||
const doc2 = await markdownToProseMirror(md1);
|
||||
// Currently FAILS: doc2.start === 1 while doc.start === 5. When the converter
|
||||
// preserves `start`, this PASSES and flips the test red — remove the pin then.
|
||||
expect(docsCanonicallyEqual(doc2, doc)).toBe(true);
|
||||
});
|
||||
});
|
||||
@@ -0,0 +1,284 @@
|
||||
import { describe, expect, it, vi } from 'vitest';
|
||||
import fc from 'fast-check';
|
||||
// Real converter, imported the same way the sibling property test does.
|
||||
import { convertProseMirrorToMarkdown } from '../../src/lib/markdown-converter.js';
|
||||
// Importing markdownToProseMirror mutates the global DOM via jsdom at module
|
||||
// load (expected, required for @tiptap/html's generateJSON under Node).
|
||||
import { markdownToProseMirror } from '../../src/lib/markdown-to-prosemirror.js';
|
||||
import { docsCanonicallyEqual, canonicalizeContent } from '../../src/lib/index.js';
|
||||
import { firstDivergence } from '../roundtrip-helpers.js';
|
||||
import {
|
||||
schema,
|
||||
allSchemaAttrKeys,
|
||||
allSchemaMarkAttrKeys,
|
||||
attrIsValueFuzzed,
|
||||
} from './attr-arbitraries.js';
|
||||
import {
|
||||
buildGenerators,
|
||||
coveredTypes,
|
||||
KNOWN_UNCOVERED,
|
||||
} from './node-generators.js';
|
||||
|
||||
// ── Attribute-value coverage allowlist ──────────────────────────────────────
|
||||
// The node/mark completeness contract guarantees every TYPE is generated, but
|
||||
// NOT that every attribute is exercised at a NON-DEFAULT value. An attribute
|
||||
// with no `arb` in attr-arbitraries.ts is only ever tested at absent/default —
|
||||
// an INVISIBLE coverage hole (the reviewer's concern). This allowlist makes that
|
||||
// hole EXPLICIT: it is the exact set of attrs deliberately not value-fuzzed, so
|
||||
// a NEW attribute (or a newly-frozen one) that lands in this bucket flips the
|
||||
// snapshot test red and forces a reviewer to classify it. Each belongs to one of:
|
||||
// - internal/opaque ids & placeholders (attachmentId, slugId, placeholder,
|
||||
// creatorId, anchorId) — no meaningful non-default to assert;
|
||||
// - dimensions/among the media family with no standalone md form here
|
||||
// (aspectRatio, size, caption, drawio/excalidraw/pdf/video/youtube w/h/align)
|
||||
// — round-trip candidates deferred to a later PR, not silently dropped;
|
||||
// - ACCEPTED limitations with no md representation (indent, callout.icon,
|
||||
// orderedList.type, table spans/bg/colwidth);
|
||||
// - PINNED bugs (column.width, orderedList.start) tracked in
|
||||
// counterexamples.test.ts.
|
||||
const ATTR_VALUE_FUZZ_ALLOWLIST = new Set<string>([
|
||||
'attachment.attachmentId', 'attachment.mime', 'attachment.placeholder', 'attachment.size',
|
||||
'audio.attachmentId', 'audio.placeholder', 'audio.size',
|
||||
'callout.icon', 'column.width',
|
||||
'drawio.align', 'drawio.alt', 'drawio.aspectRatio', 'drawio.attachmentId',
|
||||
'drawio.height', 'drawio.size', 'drawio.title', 'drawio.width',
|
||||
'embed.align', 'embed.height', 'embed.width',
|
||||
'excalidraw.align', 'excalidraw.alt', 'excalidraw.aspectRatio', 'excalidraw.attachmentId',
|
||||
'excalidraw.height', 'excalidraw.size', 'excalidraw.title', 'excalidraw.width',
|
||||
'heading.indent',
|
||||
'image.aspectRatio', 'image.attachmentId', 'image.caption', 'image.placeholder', 'image.size',
|
||||
'mention.anchorId', 'mention.creatorId', 'mention.slugId',
|
||||
'orderedList.start', 'orderedList.type', 'paragraph.indent',
|
||||
'pdf.attachmentId', 'pdf.height', 'pdf.placeholder', 'pdf.size', 'pdf.width',
|
||||
'tableCell.backgroundColor', 'tableCell.backgroundColorName', 'tableCell.colspan',
|
||||
'tableCell.colwidth', 'tableCell.rowspan',
|
||||
'tableHeader.backgroundColor', 'tableHeader.backgroundColorName', 'tableHeader.colspan',
|
||||
'tableHeader.colwidth', 'tableHeader.rowspan',
|
||||
'video.align', 'video.aspectRatio', 'video.attachmentId', 'video.placeholder', 'video.size',
|
||||
'youtube.align', 'youtube.height', 'youtube.width',
|
||||
]);
|
||||
|
||||
// ── MARK attribute-value coverage ───────────────────────────────────────────
|
||||
// Marks are fuzzed by the text generator (text-arbitraries.ts markedTextRunArb),
|
||||
// not the node OVERRIDES table, so their value-fuzz coverage is tracked with this
|
||||
// separate registry — otherwise the "no invisible coverage hole" guarantee would
|
||||
// hold for node attrs only, and a new mark attr (or a new attributed mark) would
|
||||
// silently escape the fuzz set. Every schema mark attr must be in exactly one of:
|
||||
// MARK_ATTR_FUZZED — actually driven at a non-default value by the generator;
|
||||
// MARK_ATTR_ALLOWLIST — deliberately not value-fuzzed, with a reason.
|
||||
const MARK_ATTR_FUZZED = new Set<string>([
|
||||
'mark:link.href', // markedTextRunArb sets a random webUrl href
|
||||
'mark:link.title', // ...and an optional letter-bearing title
|
||||
'mark:highlight.color', // highlight mark carries a generated color
|
||||
'mark:textStyle.color', // textStyle mark carries a generated color
|
||||
'mark:comment.commentId', // comment anchor id (alphanumeric token)
|
||||
'mark:comment.resolved', // comment resolved flag (rides only when true)
|
||||
]);
|
||||
const MARK_ATTR_ALLOWLIST = new Set<string>([
|
||||
// link presentational/routing attrs: not part of the markdown link surface the
|
||||
// converter emits (it round-trips href + title only), so there is no
|
||||
// non-default value to assert here — a deferred concern for a link-specific
|
||||
// fixture, not the flat generative pass.
|
||||
'mark:link.internal',
|
||||
'mark:link.target',
|
||||
'mark:link.rel',
|
||||
'mark:link.class',
|
||||
]);
|
||||
|
||||
// Each run does a real convert + marked + jsdom parse (~ms). Give ample headroom
|
||||
// so the suite is deterministic regardless of parallel worker load (like the
|
||||
// sibling property file).
|
||||
vi.setConfig({ testTimeout: 30000 });
|
||||
|
||||
// ---------------------------------------------------------------------------
|
||||
// #351 PR 1 — GENERATIVE (property-based) round-trip over FLAT (single-node)
|
||||
// documents at the ATTRIBUTE level.
|
||||
//
|
||||
// We assert three invariants for ANY generated valid flat document `d`
|
||||
// (pmToMd = convertProseMirrorToMarkdown, mdToPm = markdownToProseMirror):
|
||||
//
|
||||
// P1 — semantic round-trip (nothing lost):
|
||||
// docsCanonicallyEqual(await mdToPm(pmToMd(d)), d) === true
|
||||
// P2 — byte fixpoint (anti "GS-EDIT-REVERT" churn):
|
||||
// pmToMd(await mdToPm(pmToMd(d))) === pmToMd(d)
|
||||
// P3 — totality: neither converter throws; bounded.
|
||||
//
|
||||
// The generators are schema-DERIVED (attribute lists come from
|
||||
// schema.nodes[type].spec.attrs) and stay inside the round-trip-supported space
|
||||
// proven empirically by probing the live converter (see attr-arbitraries.ts and
|
||||
// text-arbitraries.ts). P1 runs over the safe attribute space; P2/P3 run over
|
||||
// the wider 'fuzz' space that also injects degenerate attribute states, which
|
||||
// P2 tolerates via a one-time first-pass normalization and P3 via totality only.
|
||||
// ---------------------------------------------------------------------------
|
||||
|
||||
// Fixed seed so every failure is reproducible; fast-check also prints the
|
||||
// shrunk counterexample. numRuns starts modest to keep CI under budget — the
|
||||
// issue's CI target is ~300-500 per property; the nightly / PR 3 will crank
|
||||
// this up further. Each property runs over the UNION (fc.oneof) of all flat
|
||||
// node generators, so the runs are shared across node types (one test per
|
||||
// property keeps the jsdom import cost and memory bounded — a per-generator ×
|
||||
// per-property matrix is ~200 heavy tests that OOMs the worker).
|
||||
const SEED = 20250705;
|
||||
const NUM_RUNS = 300;
|
||||
|
||||
const P1_GENERATORS = buildGenerators('p1');
|
||||
const FUZZ_GENERATORS = buildGenerators('fuzz');
|
||||
|
||||
// Union arbitraries: a single draw picks one node generator, then a document
|
||||
// from it. On failure fast-check prints the shrunk counterexample doc, which
|
||||
// names the offending node type directly.
|
||||
const p1Union = fc.oneof(...P1_GENERATORS.map((g) => g.arb));
|
||||
const fuzzUnion = fc.oneof(...FUZZ_GENERATORS.map((g) => g.arb));
|
||||
|
||||
async function roundTrip(doc: unknown): Promise<{ md1: string; md2: string; doc2: any }> {
|
||||
const md1 = convertProseMirrorToMarkdown(doc);
|
||||
const doc2 = await markdownToProseMirror(md1);
|
||||
const md2 = convertProseMirrorToMarkdown(doc2);
|
||||
return { md1, md2, doc2 };
|
||||
}
|
||||
|
||||
describe('#351 flat generative round-trip — completeness contract', () => {
|
||||
it('every schema node and mark is covered by a generator or explicitly allowlisted', () => {
|
||||
const covered = coveredTypes();
|
||||
const uncovered: string[] = [];
|
||||
|
||||
for (const nodeType of Object.keys(schema.nodes)) {
|
||||
if (covered.has(nodeType)) continue;
|
||||
if (nodeType in KNOWN_UNCOVERED) continue;
|
||||
uncovered.push(`node:${nodeType}`);
|
||||
}
|
||||
for (const markType of Object.keys(schema.marks)) {
|
||||
if (covered.has(`mark:${markType}`)) continue;
|
||||
if (markType in KNOWN_UNCOVERED) continue;
|
||||
uncovered.push(`mark:${markType}`);
|
||||
}
|
||||
|
||||
// A new node/mark added to the schema with no generator AND no allowlist
|
||||
// entry MUST turn this test red — that is the whole point (no silent blind
|
||||
// spots).
|
||||
expect(
|
||||
uncovered,
|
||||
`these schema types have no generator and no KNOWN_UNCOVERED reason:\n ${uncovered.join(
|
||||
'\n ',
|
||||
)}`,
|
||||
).toEqual([]);
|
||||
});
|
||||
|
||||
it('every KNOWN_UNCOVERED entry is a real schema type (no stale allowlist rows)', () => {
|
||||
const all = new Set([...Object.keys(schema.nodes), ...Object.keys(schema.marks)]);
|
||||
for (const t of Object.keys(KNOWN_UNCOVERED)) {
|
||||
expect(all.has(t), `stale KNOWN_UNCOVERED entry: ${t}`).toBe(true);
|
||||
}
|
||||
});
|
||||
|
||||
it('every attribute is value-fuzzed OR explicitly allowlisted (no invisible hole)', () => {
|
||||
// Makes the "generic-frozen" coverage hole VISIBLE: any schema attr not
|
||||
// exercised at a non-default value must be a KNOWN entry in the allowlist.
|
||||
// A new attr (or one that loses its `arb`) that falls into the not-fuzzed
|
||||
// bucket without an allowlist row turns this red — no silent blind spots.
|
||||
const unaccounted: string[] = [];
|
||||
for (const key of allSchemaAttrKeys()) {
|
||||
const i = key.indexOf('.');
|
||||
const fuzzed = attrIsValueFuzzed(key.slice(0, i), key.slice(i + 1));
|
||||
if (!fuzzed && !ATTR_VALUE_FUZZ_ALLOWLIST.has(key)) unaccounted.push(key);
|
||||
}
|
||||
expect(
|
||||
unaccounted,
|
||||
`these attrs are not value-fuzzed and not in ATTR_VALUE_FUZZ_ALLOWLIST:\n ${unaccounted.join(
|
||||
'\n ',
|
||||
)}`,
|
||||
).toEqual([]);
|
||||
});
|
||||
|
||||
it('the attribute allowlist has no stale rows (every entry is really not-fuzzed)', () => {
|
||||
const notFuzzed = new Set(
|
||||
allSchemaAttrKeys().filter((key) => {
|
||||
const i = key.indexOf('.');
|
||||
return !attrIsValueFuzzed(key.slice(0, i), key.slice(i + 1));
|
||||
}),
|
||||
);
|
||||
for (const key of ATTR_VALUE_FUZZ_ALLOWLIST) {
|
||||
expect(
|
||||
notFuzzed.has(key),
|
||||
`stale allowlist row (attr is now value-fuzzed, remove it): ${key}`,
|
||||
).toBe(true);
|
||||
}
|
||||
});
|
||||
|
||||
it('every MARK attribute is value-fuzzed OR allowlisted (no invisible hole)', () => {
|
||||
// The node guard above covers node attrs; marks are fuzzed by the text
|
||||
// generator, so their coverage is tracked separately. A new mark attr (or a
|
||||
// newly-attributed mark) that lands in neither set turns this red.
|
||||
const unaccounted: string[] = [];
|
||||
for (const key of allSchemaMarkAttrKeys()) {
|
||||
if (!MARK_ATTR_FUZZED.has(key) && !MARK_ATTR_ALLOWLIST.has(key)) {
|
||||
unaccounted.push(key);
|
||||
}
|
||||
}
|
||||
expect(
|
||||
unaccounted,
|
||||
`these mark attrs are neither in MARK_ATTR_FUZZED nor MARK_ATTR_ALLOWLIST:\n ${unaccounted.join(
|
||||
'\n ',
|
||||
)}`,
|
||||
).toEqual([]);
|
||||
});
|
||||
|
||||
it('the MARK fuzz/allowlist sets have no stale rows (every entry is a real schema mark attr)', () => {
|
||||
const all = new Set(allSchemaMarkAttrKeys());
|
||||
for (const key of [...MARK_ATTR_FUZZED, ...MARK_ATTR_ALLOWLIST]) {
|
||||
expect(all.has(key), `stale mark-attr registry row: ${key}`).toBe(true);
|
||||
}
|
||||
});
|
||||
});
|
||||
|
||||
describe('#351 flat generative round-trip — properties', () => {
|
||||
it('generator validity: every generated doc passes schema.check()', () => {
|
||||
// A generator that emits an invalid ProseMirror document is a GENERATOR bug.
|
||||
fc.assert(
|
||||
fc.property(fuzzUnion, (doc) => {
|
||||
schema.nodeFromJSON(doc).check(); // throws on an invalid doc
|
||||
return true;
|
||||
}),
|
||||
{ numRuns: NUM_RUNS, seed: SEED },
|
||||
);
|
||||
});
|
||||
|
||||
it('P1 — semantic round-trip: docsCanonicallyEqual(mdToPm(pmToMd(d)), d)', async () => {
|
||||
await fc.assert(
|
||||
fc.asyncProperty(p1Union, async (doc) => {
|
||||
const { doc2 } = await roundTrip(doc);
|
||||
if (!docsCanonicallyEqual(doc2, doc)) {
|
||||
// Surface the precise divergence in the failure message.
|
||||
const div = firstDivergence(
|
||||
JSON.parse(JSON.stringify(canonicalizeContent(doc2))),
|
||||
JSON.parse(JSON.stringify(canonicalizeContent(doc))),
|
||||
);
|
||||
throw new Error(
|
||||
`P1 divergence @ ${div?.path}: got=${JSON.stringify(div?.a)} want=${JSON.stringify(div?.b)}`,
|
||||
);
|
||||
}
|
||||
}),
|
||||
{ numRuns: NUM_RUNS, seed: SEED },
|
||||
);
|
||||
});
|
||||
|
||||
it('P2 — byte fixpoint: pmToMd(mdToPm(pmToMd(d))) === pmToMd(d)', async () => {
|
||||
await fc.assert(
|
||||
fc.asyncProperty(fuzzUnion, async (doc) => {
|
||||
const { md1, md2 } = await roundTrip(doc);
|
||||
expect(md2).toBe(md1);
|
||||
}),
|
||||
{ numRuns: NUM_RUNS, seed: SEED },
|
||||
);
|
||||
});
|
||||
|
||||
it('P3 — totality: neither converter throws', async () => {
|
||||
await fc.assert(
|
||||
fc.asyncProperty(fuzzUnion, async (doc) => {
|
||||
// Throwing here fails the property; fast-check shrinks to a minimal doc.
|
||||
await roundTrip(doc);
|
||||
}),
|
||||
{ numRuns: NUM_RUNS, seed: SEED },
|
||||
);
|
||||
});
|
||||
});
|
||||
@@ -0,0 +1,310 @@
|
||||
/**
|
||||
* Flat single-node document generators (#351, PR 1).
|
||||
*
|
||||
* For every schema node type that can stand alone, a fast-check arbitrary
|
||||
* producing `{ type:'doc', content:[ <the target node> ] }` with generated attrs
|
||||
* (via nodeAttrsArb) and the minimal REQUIRED immediate children the schema
|
||||
* demands (a heading's inline text, a listItem's one paragraph, a table's
|
||||
* minimal rows, details' summary+content, a callout's one paragraph). Kept
|
||||
* FLAT: a single target node, no deep nesting — nested structural generation is
|
||||
* PR 2.
|
||||
*
|
||||
* The `mode` threads through to the attribute arbitraries:
|
||||
* - 'p1' : the round-trip-safe attribute space (P1 semantic round-trip).
|
||||
* - 'fuzz' : adds degenerate attribute states (P2 byte-fixpoint tolerates the
|
||||
* one-time normalization; P3 only needs totality).
|
||||
*
|
||||
* A COMPLETENESS CONTRACT (see flat-roundtrip.property.test.ts) enumerates the
|
||||
* whole schema and asserts every node/mark is EITHER produced by a generator
|
||||
* here OR listed in KNOWN_UNCOVERED with a reason — so a new schema type with no
|
||||
* generator turns the suite RED.
|
||||
*/
|
||||
import fc from 'fast-check';
|
||||
import { type AttrMode, nodeAttrsArb } from './attr-arbitraries.js';
|
||||
import {
|
||||
inlineContentArb,
|
||||
headingInlineContentArb,
|
||||
plainInlineContentArb,
|
||||
phraseArb,
|
||||
markedTextRunArb,
|
||||
} from './text-arbitraries.js';
|
||||
|
||||
const doc = (node: any) => ({ type: 'doc', content: [node] });
|
||||
const para = (content: any[]) => ({ type: 'paragraph', content });
|
||||
|
||||
/** A named flat-document generator. */
|
||||
export interface NamedGen {
|
||||
name: string;
|
||||
arb: fc.Arbitrary<any>;
|
||||
}
|
||||
|
||||
// ---------------------------------------------------------------------------
|
||||
// Per-target generators, each a function of mode.
|
||||
// ---------------------------------------------------------------------------
|
||||
|
||||
const gen = {
|
||||
paragraph: (m: AttrMode) =>
|
||||
fc.tuple(nodeAttrsArb('paragraph', m), inlineContentArb).map(([attrs, content]) =>
|
||||
doc({ type: 'paragraph', attrs, content }),
|
||||
),
|
||||
|
||||
heading: (m: AttrMode) =>
|
||||
fc.tuple(nodeAttrsArb('heading', m), headingInlineContentArb).map(([attrs, content]) =>
|
||||
doc({ type: 'heading', attrs, content }),
|
||||
),
|
||||
|
||||
blockquote: (_m: AttrMode) =>
|
||||
inlineContentArb.map((content) => doc({ type: 'blockquote', content: [para(content)] })),
|
||||
|
||||
bulletList: (_m: AttrMode) =>
|
||||
fc
|
||||
.array(inlineContentArb, { minLength: 1, maxLength: 3 })
|
||||
.map((items) =>
|
||||
doc({
|
||||
type: 'bulletList',
|
||||
content: items.map((c) => ({ type: 'listItem', content: [para(c)] })),
|
||||
}),
|
||||
),
|
||||
|
||||
orderedList: (m: AttrMode) =>
|
||||
fc
|
||||
.tuple(nodeAttrsArb('orderedList', m), fc.array(inlineContentArb, { minLength: 1, maxLength: 3 }))
|
||||
.map(([attrs, items]) =>
|
||||
doc({
|
||||
type: 'orderedList',
|
||||
attrs,
|
||||
content: items.map((c) => ({ type: 'listItem', content: [para(c)] })),
|
||||
}),
|
||||
),
|
||||
|
||||
taskList: (m: AttrMode) =>
|
||||
fc
|
||||
.array(fc.tuple(nodeAttrsArb('taskItem', m), inlineContentArb), { minLength: 1, maxLength: 3 })
|
||||
.map((items) =>
|
||||
doc({
|
||||
type: 'taskList',
|
||||
content: items.map(([attrs, c]) => ({ type: 'taskItem', attrs, content: [para(c)] })),
|
||||
}),
|
||||
),
|
||||
|
||||
codeBlock: (m: AttrMode) =>
|
||||
fc
|
||||
.tuple(
|
||||
nodeAttrsArb('codeBlock', m),
|
||||
// A fenced code block always re-imports with a TRAILING NEWLINE in its
|
||||
// text (empirically confirmed). Author the newline so the doc is already
|
||||
// at the round-trip fixpoint (supported-space shaping, not a masked bug).
|
||||
fc.array(phraseArb, { minLength: 1, maxLength: 3 }).map((lines) => lines.join('\n') + '\n'),
|
||||
)
|
||||
.map(([attrs, code]) =>
|
||||
doc({ type: 'codeBlock', attrs, content: [{ type: 'text', text: code }] }),
|
||||
),
|
||||
|
||||
horizontalRule: (_m: AttrMode) => fc.constant(doc({ type: 'horizontalRule' })),
|
||||
|
||||
pageBreak: (_m: AttrMode) => fc.constant(doc({ type: 'pageBreak' })),
|
||||
|
||||
image: (m: AttrMode) => nodeAttrsArb('image', m).map((attrs) => doc({ type: 'image', attrs })),
|
||||
|
||||
callout: (m: AttrMode) =>
|
||||
fc.tuple(nodeAttrsArb('callout', m), inlineContentArb).map(([attrs, content]) =>
|
||||
doc({ type: 'callout', attrs, content: [para(content)] }),
|
||||
),
|
||||
|
||||
mathBlock: (m: AttrMode) =>
|
||||
nodeAttrsArb('mathBlock', m).map((attrs) => doc({ type: 'mathBlock', attrs })),
|
||||
|
||||
details: (m: AttrMode) =>
|
||||
fc
|
||||
.tuple(nodeAttrsArb('details', m), plainInlineContentArb, inlineContentArb)
|
||||
.map(([attrs, summary, body]) =>
|
||||
doc({
|
||||
type: 'details',
|
||||
attrs,
|
||||
content: [
|
||||
{ type: 'detailsSummary', content: summary },
|
||||
{ type: 'detailsContent', content: [para(body)] },
|
||||
],
|
||||
}),
|
||||
),
|
||||
|
||||
table: (_m: AttrMode) =>
|
||||
fc.integer({ min: 1, max: 3 }).chain((cols) => {
|
||||
// GFM alignment is column-wide (encoded in the header separator), so a
|
||||
// column's alignment must be identical on the header and every body cell,
|
||||
// else the second export re-aligns and churns. Pick ONE align per column.
|
||||
const alignsArb = fc.array(fc.constantFrom(undefined, 'left', 'center', 'right'), {
|
||||
minLength: cols,
|
||||
maxLength: cols,
|
||||
});
|
||||
const cell = (header: boolean, align?: string) =>
|
||||
phraseArb.map((t) => ({
|
||||
type: header ? 'tableHeader' : 'tableCell',
|
||||
// colspan/rowspan pinned to 1 (GFM cannot express spans); optional
|
||||
// column-consistent align.
|
||||
attrs: { colspan: 1, rowspan: 1, ...(align ? { align } : {}) },
|
||||
content: [para([{ type: 'text', text: t }])],
|
||||
}));
|
||||
return alignsArb.chain((aligns) => {
|
||||
const headerRow = fc
|
||||
.tuple(...aligns.map((a) => cell(true, a)))
|
||||
.map((cells) => ({ type: 'tableRow', content: cells }));
|
||||
const bodyRow = fc
|
||||
.tuple(...aligns.map((a) => cell(false, a)))
|
||||
.map((cells) => ({ type: 'tableRow', content: cells }));
|
||||
return fc
|
||||
.tuple(headerRow, fc.array(bodyRow, { minLength: 1, maxLength: 2 }))
|
||||
.map(([h, body]) => doc({ type: 'table', content: [h, ...body] }));
|
||||
});
|
||||
}),
|
||||
|
||||
columns: (m: AttrMode) =>
|
||||
// Couple the column count to the layout so the two stay consistent
|
||||
// (two_equal/left_sidebar/right_sidebar -> 2, three_equal -> 3).
|
||||
fc
|
||||
.constantFrom('two_equal', 'three_equal', 'left_sidebar', 'right_sidebar')
|
||||
.chain((layout) => {
|
||||
const count = layout === 'three_equal' ? 3 : 2;
|
||||
return fc
|
||||
.tuple(
|
||||
nodeAttrsArb('columns', m, { layout, widthMode: 'normal' }),
|
||||
fc.array(inlineContentArb, { minLength: count, maxLength: count }),
|
||||
)
|
||||
.map(([attrs, bodies]) =>
|
||||
doc({
|
||||
type: 'columns',
|
||||
attrs,
|
||||
content: bodies.map((c) => ({ type: 'column', content: [para(c)] })),
|
||||
}),
|
||||
);
|
||||
}),
|
||||
|
||||
subpages: (m: AttrMode) =>
|
||||
nodeAttrsArb('subpages', m).map((attrs) => doc({ type: 'subpages', attrs })),
|
||||
|
||||
audio: (m: AttrMode) => nodeAttrsArb('audio', m).map((attrs) => doc({ type: 'audio', attrs })),
|
||||
video: (m: AttrMode) => nodeAttrsArb('video', m).map((attrs) => doc({ type: 'video', attrs })),
|
||||
pdf: (m: AttrMode) => nodeAttrsArb('pdf', m).map((attrs) => doc({ type: 'pdf', attrs })),
|
||||
youtube: (m: AttrMode) => nodeAttrsArb('youtube', m).map((attrs) => doc({ type: 'youtube', attrs })),
|
||||
embed: (m: AttrMode) => nodeAttrsArb('embed', m).map((attrs) => doc({ type: 'embed', attrs })),
|
||||
drawio: (m: AttrMode) => nodeAttrsArb('drawio', m).map((attrs) => doc({ type: 'drawio', attrs })),
|
||||
excalidraw: (m: AttrMode) =>
|
||||
nodeAttrsArb('excalidraw', m).map((attrs) => doc({ type: 'excalidraw', attrs })),
|
||||
attachment: (m: AttrMode) =>
|
||||
nodeAttrsArb('attachment', m).map((attrs) => doc({ type: 'attachment', attrs })),
|
||||
htmlEmbed: (m: AttrMode) =>
|
||||
nodeAttrsArb('htmlEmbed', m).map((attrs) => doc({ type: 'htmlEmbed', attrs })),
|
||||
pageEmbed: (m: AttrMode) =>
|
||||
nodeAttrsArb('pageEmbed', m).map((attrs) => doc({ type: 'pageEmbed', attrs })),
|
||||
transclusionReference: (m: AttrMode) =>
|
||||
nodeAttrsArb('transclusionReference', m).map((attrs) =>
|
||||
doc({ type: 'transclusionReference', attrs }),
|
||||
),
|
||||
|
||||
transclusionSource: (m: AttrMode) =>
|
||||
fc.tuple(nodeAttrsArb('transclusionSource', m), inlineContentArb).map(([attrs, content]) =>
|
||||
doc({ type: 'transclusionSource', attrs, content: [para(content)] }),
|
||||
),
|
||||
|
||||
// A footnote reference PLUS its definition (the reference has no standalone
|
||||
// markdown form without its definition — see KNOWN_UNCOVERED note for the
|
||||
// bare reference). Both carry the same id. The definition body uses
|
||||
// headingInlineContentArb (NO hard breaks): a footnote is serialized inline as
|
||||
// `^[...]`, so a hard break inside it collapses to a single space on re-parse
|
||||
// (empirically confirmed) — that is the container's markdown limitation, not
|
||||
// an attribute-level concern. The reference-bearing paragraph is a NORMAL
|
||||
// paragraph and keeps the full inline corpus.
|
||||
footnotes: (m: AttrMode) =>
|
||||
fc.tuple(fc.constantFrom('fn1', 'fn2', 'note'), inlineContentArb, headingInlineContentArb).map(
|
||||
([id, refText, noteBody]) => ({
|
||||
type: 'doc',
|
||||
content: [
|
||||
para([...refText, { type: 'footnoteReference', attrs: { id } }]),
|
||||
{
|
||||
type: 'footnotesList',
|
||||
content: [{ type: 'footnoteDefinition', attrs: { id }, content: [para(noteBody)] }],
|
||||
},
|
||||
],
|
||||
}),
|
||||
),
|
||||
|
||||
// ── inline targets wrapped in a paragraph ────────────────────────────────
|
||||
mention: (m: AttrMode) =>
|
||||
nodeAttrsArb('mention', m).map((attrs) => doc(para([{ type: 'mention', attrs }]))),
|
||||
|
||||
mathInline: (m: AttrMode) =>
|
||||
fc.tuple(phraseArb, nodeAttrsArb('mathInline', m)).map(([t, attrs]) =>
|
||||
doc(para([{ type: 'text', text: t }, { type: 'mathInline', attrs }])),
|
||||
),
|
||||
|
||||
status: (m: AttrMode) =>
|
||||
nodeAttrsArb('status', m).map((attrs) => doc(para([{ type: 'status', attrs }]))),
|
||||
|
||||
hardBreak: (_m: AttrMode) =>
|
||||
fc.tuple(phraseArb, phraseArb).map(([a, b]) =>
|
||||
doc(para([{ type: 'text', text: a }, { type: 'hardBreak' }, { type: 'text', text: b }])),
|
||||
),
|
||||
|
||||
// ── marks: a paragraph of marked runs (covers every mark type) ───────────
|
||||
marksOnText: (_m: AttrMode) =>
|
||||
fc.array(markedTextRunArb, { minLength: 1, maxLength: 5 }).map((runs) => {
|
||||
// Merge adjacent same-mark runs (see text-arbitraries.normalizeInline).
|
||||
const out: any[] = [];
|
||||
for (const r of runs) {
|
||||
const prev = out[out.length - 1];
|
||||
if (prev && JSON.stringify(prev.marks ?? []) === JSON.stringify(r.marks ?? [])) {
|
||||
prev.text += r.text;
|
||||
} else out.push({ ...r });
|
||||
}
|
||||
return doc(para(out));
|
||||
}),
|
||||
};
|
||||
|
||||
/** Build the full list of named generators for a given mode. */
|
||||
export function buildGenerators(mode: AttrMode): NamedGen[] {
|
||||
return Object.entries(gen).map(([name, f]) => ({ name, arb: f(mode) }));
|
||||
}
|
||||
|
||||
// ---------------------------------------------------------------------------
|
||||
// Completeness contract support.
|
||||
// ---------------------------------------------------------------------------
|
||||
|
||||
/**
|
||||
* Schema node/mark types deliberately NOT covered by a P1/P2 generator, each
|
||||
* with a one-line reason. Excluding a type means it is kept OUT of the round-
|
||||
* trip generators — it does NOT weaken any property.
|
||||
*
|
||||
* NOTE (empirical): the candidates the issue flagged for review — pageEmbed,
|
||||
* subpages, transclusionSource/Reference, mention, status — were PROBED against
|
||||
* the live converter and DO round-trip P1/P2 with placeholder ids, so they are
|
||||
* COVERED by real generators rather than allowlisted here. The allowlist below
|
||||
* holds only types with no standalone flat generator by construction.
|
||||
*/
|
||||
export const KNOWN_UNCOVERED: Record<string, string> = {
|
||||
// The root node; it is the wrapper every generated doc already is, never a
|
||||
// "target" content node, so it has no standalone generator of its own.
|
||||
doc: 'the document root wrapper, not a content node with a standalone generator',
|
||||
};
|
||||
|
||||
/** Recursively collect every node type and `mark:<type>` under a tree. */
|
||||
export function collectTypes(node: any, seen = new Set<string>()): Set<string> {
|
||||
if (!node || typeof node !== 'object') return seen;
|
||||
if (node.type) seen.add(node.type);
|
||||
for (const m of node.marks ?? []) if (m?.type) seen.add(`mark:${m.type}`);
|
||||
for (const c of node.content ?? []) collectTypes(c, seen);
|
||||
return seen;
|
||||
}
|
||||
|
||||
/**
|
||||
* Sample every generator and return the union of node/mark types they produce.
|
||||
* Deterministic (fixed seed) so the completeness contract is stable.
|
||||
*/
|
||||
export function coveredTypes(seed = 12345, perGen = 60): Set<string> {
|
||||
const seen = new Set<string>();
|
||||
for (const { arb } of buildGenerators('p1')) {
|
||||
for (const sample of fc.sample(arb, { numRuns: perGen, seed })) {
|
||||
collectTypes(sample, seen);
|
||||
}
|
||||
}
|
||||
return seen;
|
||||
}
|
||||
@@ -0,0 +1,258 @@
|
||||
/**
|
||||
* Hostile inline-text corpus for the generative flat-document round-trip suite
|
||||
* (#351, PR 1).
|
||||
*
|
||||
* These arbitraries are a DIRECT PORT of the "supported space" guardrails that
|
||||
* `test/markdown-roundtrip.property.test.ts` proved empirically against the live
|
||||
* converter. That file's long header documents WHY each guardrail exists; rather
|
||||
* than re-derive them, we reuse the exact same shapes here so the attribute-level
|
||||
* generative suite inherits the same byte-stable text space. Each guardrail is
|
||||
* cited back to that file below.
|
||||
*
|
||||
* The corpus deliberately spans the CommonMark / canon hostile alphabet
|
||||
* (`* _ [ ] ( ) { } | < > & # ! ~ = + -`), unicode / emoji / RTL, and the legal
|
||||
* mark combinations on runs (including the `code` mark, which the schema's
|
||||
* `excludes: "_"` makes suppress every co-occurring mark — so it is never
|
||||
* combined with another mark in the byte-stable space).
|
||||
*/
|
||||
import fc from 'fast-check';
|
||||
|
||||
// ---------------------------------------------------------------------------
|
||||
// Words and the hostile special-character alphabet.
|
||||
// (Ported from markdown-roundtrip.property.test.ts, "Inline text arbitraries".)
|
||||
// ---------------------------------------------------------------------------
|
||||
|
||||
/** Alphanumeric "word" (no markdown-significant characters). Length 1..6. */
|
||||
export const wordArb = fc
|
||||
.stringMatching(/^[A-Za-z0-9]{1,6}$/)
|
||||
.filter((w) => w.length > 0);
|
||||
|
||||
/**
|
||||
* A SINGLE markdown-significant character, emitted only as an isolated,
|
||||
* space-flanked token. Every char the task calls out plus a few more; each was
|
||||
* verified byte-stable in this position by the sibling property test.
|
||||
*
|
||||
* NOTE: the backtick (`) is DELIBERATELY excluded from free-floating plain text
|
||||
* (it is a code-span delimiter that re-pairs globally). It is exercised only via
|
||||
* the `code` mark and code blocks — see markdown-roundtrip.property.test.ts.
|
||||
*/
|
||||
export const specialCharArb = fc.constantFrom(
|
||||
'*', '_', '[', ']', '(', ')', '{', '}', '|', '<', '>', '&', '#', '!', '~', '=', '+', '-',
|
||||
);
|
||||
|
||||
// A pinch of unicode / emoji / RTL, always word-like (no markdown specials) so
|
||||
// it stays inside the space-flanked corpus. Kept letter/emoji-bearing so it is
|
||||
// never coerced to a number (see letterPhraseArb rationale).
|
||||
export const unicodeWordArb = fc.constantFrom(
|
||||
'café', 'naïve', 'Zürich', 'Москва', 'こんにちは', '你好', '😀', '🚀x', 'مرحبا', 'שלום',
|
||||
);
|
||||
|
||||
/**
|
||||
* A "safe special" text string: a space-joined sequence of tokens that always
|
||||
* BEGINS and ENDS with an alphanumeric word, with any isolated special chars (or
|
||||
* unicode words) confined to the MIDDLE, each space-flanked by words.
|
||||
*
|
||||
* Both boundary guarantees matter (verbatim from the sibling test):
|
||||
* * Leading word: the line never opens with a block/inline trigger
|
||||
* (">", "*", "-", "#", "1." ...).
|
||||
* * Trailing word: adjacent text runs CONCATENATE with no separator, so a run
|
||||
* ending in a bare "<" beside a run starting with a letter would form a fake
|
||||
* HTML tag. Ending every run with a word keeps every special internal and
|
||||
* space-flanked even after concatenation.
|
||||
*/
|
||||
export const safeTextArb: fc.Arbitrary<string> = fc
|
||||
.tuple(
|
||||
wordArb,
|
||||
fc.array(fc.oneof(wordArb, specialCharArb, unicodeWordArb), {
|
||||
minLength: 0,
|
||||
maxLength: 3,
|
||||
}),
|
||||
wordArb,
|
||||
)
|
||||
.map(([first, middle, last]) => [first, ...middle, last].join(' '));
|
||||
|
||||
/**
|
||||
* A plain alphanumeric phrase (1..3 words) for places where even isolated
|
||||
* specials are not wanted (e.g. code-block language, mention labels, status
|
||||
* text, table cells rendered on the plain-markdown path).
|
||||
*/
|
||||
export const phraseArb: fc.Arbitrary<string> = fc
|
||||
.array(wordArb, { minLength: 1, maxLength: 3 })
|
||||
.map((ws) => ws.join(' '));
|
||||
|
||||
/**
|
||||
* A phrase guaranteed to contain at least one letter. Used for image/media alt
|
||||
* text and link titles: a PURELY numeric alt/title (e.g. "0") is parsed back as
|
||||
* a NUMBER and then dropped by the converter's `value || ""` coercion — not
|
||||
* byte-stable. A letter anywhere keeps it a string. (Ported verbatim.)
|
||||
*/
|
||||
export const letterPhraseArb: fc.Arbitrary<string> = fc
|
||||
.tuple(
|
||||
fc.stringMatching(/^[A-Za-z]{1,4}$/),
|
||||
fc.array(wordArb, { minLength: 0, maxLength: 2 }),
|
||||
)
|
||||
.map(([head, rest]) => [head, ...rest].join(' '));
|
||||
|
||||
/** A paren/space-free URL — safe inside markdown link/image `(...)` syntax. */
|
||||
export const urlArb: fc.Arbitrary<string> = fc
|
||||
.webUrl()
|
||||
.filter((u) => !/[()\s]/.test(u));
|
||||
|
||||
// ---------------------------------------------------------------------------
|
||||
// Marked inline runs.
|
||||
// (Ported from markdown-roundtrip.property.test.ts "markedTextRunArb".)
|
||||
// ---------------------------------------------------------------------------
|
||||
|
||||
/**
|
||||
* A text run with an OPTIONAL single non-code formatting mark (bold/italic/
|
||||
* strike/underline/superscript/subscript/spoiler), or a SOLE `code` mark, or a
|
||||
* link, or an inline comment anchor. `code` is NEVER combined with another mark
|
||||
* in the byte-stable space (that combination is a documented converter
|
||||
* limitation — the schema's `code` mark declares `excludes: "_"`). Marks wrap
|
||||
* `safeTextArb`, which stays stable even when it contains isolated specials.
|
||||
*
|
||||
* The mark set here is broadened past the sibling test's {bold,italic,strike}
|
||||
* to also cover underline / superscript / subscript / spoiler / textStyle /
|
||||
* highlight (all single, non-code marks), so the marks-on-text generator
|
||||
* exercises every mark the schema declares except the deliberately-excluded
|
||||
* `code`+other combination.
|
||||
*/
|
||||
export const markedTextRunArb: fc.Arbitrary<any> = fc.oneof(
|
||||
// Plain text.
|
||||
safeTextArb.map((t) => ({ type: 'text', text: t })),
|
||||
// Single formatting mark (attribute-free marks).
|
||||
fc
|
||||
.tuple(
|
||||
safeTextArb,
|
||||
fc.constantFrom('bold', 'italic', 'strike', 'underline', 'superscript', 'subscript', 'spoiler'),
|
||||
)
|
||||
.map(([t, m]) => ({ type: 'text', text: t, marks: [{ type: m }] })),
|
||||
// highlight with a color attr.
|
||||
fc
|
||||
.tuple(safeTextArb, fc.constantFrom('#ffcc00', '#a0e0ff', 'yellow'))
|
||||
.map(([t, color]) => ({ type: 'text', text: t, marks: [{ type: 'highlight', attrs: { color } }] })),
|
||||
// textStyle with a color attr.
|
||||
fc
|
||||
.tuple(safeTextArb, fc.constantFrom('#123456', '#ff0000', '#00aa88'))
|
||||
.map(([t, color]) => ({ type: 'text', text: t, marks: [{ type: 'textStyle', attrs: { color } }] })),
|
||||
// Sole code mark (backtick span). safeTextArb is backtick-free, so the span
|
||||
// content cannot contain an inner backtick.
|
||||
safeTextArb.map((t) => ({ type: 'text', text: t, marks: [{ type: 'code' }] })),
|
||||
// Link with safe text, a paren/space-free href, optionally a letter-bearing
|
||||
// title (a purely numeric title is coerced to a number and dropped).
|
||||
fc
|
||||
.tuple(phraseArb, urlArb, fc.option(letterPhraseArb, { nil: undefined }))
|
||||
.map(([t, href, title]) => ({
|
||||
type: 'text',
|
||||
text: t,
|
||||
marks: [{ type: 'link', attrs: title ? { href, title } : { href } }],
|
||||
})),
|
||||
// Inline comment anchor: a span[data-comment-id] that must survive byte-for-
|
||||
// byte. commentId is an alphanumeric token; `resolved` rides only when true.
|
||||
fc
|
||||
.tuple(safeTextArb, fc.stringMatching(/^[A-Za-z0-9]{4,10}$/), fc.boolean())
|
||||
.map(([t, commentId, resolved]) => ({
|
||||
type: 'text',
|
||||
text: t,
|
||||
marks: [
|
||||
{ type: 'comment', attrs: resolved ? { commentId, resolved: true } : { commentId } },
|
||||
],
|
||||
})),
|
||||
);
|
||||
|
||||
// ---------------------------------------------------------------------------
|
||||
// Inline atoms and inline-content assembly.
|
||||
// (Ported from markdown-roundtrip.property.test.ts.)
|
||||
// ---------------------------------------------------------------------------
|
||||
|
||||
/** Inline math node carrying LaTeX that includes the `a < b` the task asks for. */
|
||||
export const mathInlineArb: fc.Arbitrary<any> = fc
|
||||
.constantFrom('a < b', 'x^2 + y^2', 'a < b < c', '\\frac{1}{2}', 'E = mc^2')
|
||||
.map((text) => ({ type: 'mathInline', attrs: { text } }));
|
||||
|
||||
/** Mention node; label/id/entity are plain phrases / uuids. */
|
||||
export const mentionArb: fc.Arbitrary<any> = fc
|
||||
.tuple(phraseArb, fc.uuid(), fc.uuid())
|
||||
.map(([label, id, entityId]) => ({
|
||||
type: 'mention',
|
||||
attrs: { id, label, entityType: 'user', entityId },
|
||||
}));
|
||||
|
||||
export const hardBreakArb: fc.Arbitrary<any> = fc.constant({ type: 'hardBreak' });
|
||||
|
||||
const sameMarks = (a: any[] | undefined, b: any[] | undefined): boolean =>
|
||||
JSON.stringify(a ?? []) === JSON.stringify(b ?? []);
|
||||
|
||||
/**
|
||||
* Canonicalize a generated inline-content array the way ProseMirror stores it,
|
||||
* then trim the markdown-fragile edges. (Ported verbatim from
|
||||
* markdown-roundtrip.property.test.ts "normalizeInline":)
|
||||
* 1) MERGE adjacent text runs with IDENTICAL marks (the editor coalesces
|
||||
* them; split same-mark runs export to ambiguous "**a****b**").
|
||||
* 2) Collapse CONSECUTIVE hard breaks (two render a blank line marked eats).
|
||||
* 3) Drop a TRAILING hard break (removed by the converter's .trim()).
|
||||
*/
|
||||
export function normalizeInline(nodes: any[]): any[] {
|
||||
const out: any[] = [];
|
||||
for (const node of nodes) {
|
||||
const prev = out[out.length - 1];
|
||||
if (node.type === 'hardBreak' && prev && prev.type === 'hardBreak') continue;
|
||||
if (
|
||||
node.type === 'text' &&
|
||||
prev &&
|
||||
prev.type === 'text' &&
|
||||
sameMarks(prev.marks, node.marks)
|
||||
) {
|
||||
prev.text += node.text;
|
||||
continue;
|
||||
}
|
||||
out.push(node.type === 'text' ? { ...node } : node);
|
||||
}
|
||||
while (out.length > 1 && out[out.length - 1].type === 'hardBreak') out.pop();
|
||||
return out;
|
||||
}
|
||||
|
||||
/**
|
||||
* Inline content for a paragraph: at least one marked text run, optionally with
|
||||
* inline atoms (math/mention) and hard breaks interspersed. Always starts with a
|
||||
* text run so the paragraph never opens with a block trigger. (Ported.)
|
||||
*/
|
||||
export const inlineContentArb: fc.Arbitrary<any[]> = fc
|
||||
.tuple(
|
||||
markedTextRunArb,
|
||||
fc.array(
|
||||
fc.oneof(
|
||||
{ weight: 5, arbitrary: markedTextRunArb },
|
||||
{ weight: 1, arbitrary: mathInlineArb },
|
||||
{ weight: 1, arbitrary: mentionArb },
|
||||
{ weight: 1, arbitrary: hardBreakArb },
|
||||
),
|
||||
{ minLength: 0, maxLength: 4 },
|
||||
),
|
||||
)
|
||||
.map(([first, rest]) => normalizeInline([first, ...rest]));
|
||||
|
||||
/**
|
||||
* Inline content for a HEADING — identical to a paragraph's, but WITHOUT hard
|
||||
* breaks. A hard break inside an ATX heading is not byte-stable (marked splits
|
||||
* the heading). (Ported.)
|
||||
*/
|
||||
export const headingInlineContentArb: fc.Arbitrary<any[]> = fc
|
||||
.tuple(
|
||||
markedTextRunArb,
|
||||
fc.array(
|
||||
fc.oneof(
|
||||
{ weight: 5, arbitrary: markedTextRunArb },
|
||||
{ weight: 1, arbitrary: mathInlineArb },
|
||||
{ weight: 1, arbitrary: mentionArb },
|
||||
),
|
||||
{ minLength: 0, maxLength: 4 },
|
||||
),
|
||||
)
|
||||
.map(([first, rest]) => normalizeInline([first, ...rest]));
|
||||
|
||||
/** Simple plain-text inline content (single run) for containers rendered on the
|
||||
* raw-HTML path (table cells / column bodies) where fancy inline is undesirable. */
|
||||
export const plainInlineContentArb: fc.Arbitrary<any[]> = phraseArb.map((t) => [
|
||||
{ type: 'text', text: t },
|
||||
]);
|
||||
@@ -1,68 +0,0 @@
|
||||
diff --git a/dist/index.js b/dist/index.js
|
||||
index ae447a12f7823ec0a00837ee9f0eb809a610d5f8..a3402b2c2d021ef432cfa76e35d370073d525135 100644
|
||||
--- a/dist/index.js
|
||||
+++ b/dist/index.js
|
||||
@@ -6578,9 +6578,19 @@ function createOutputTransformStream(output) {
|
||||
controller.enqueue({ part: chunk, partialOutput: void 0 });
|
||||
return;
|
||||
}
|
||||
- text2 += chunk.text;
|
||||
textChunk += chunk.text;
|
||||
textProviderMetadata = (_a21 = chunk.providerMetadata) != null ? _a21 : textProviderMetadata;
|
||||
+ if (output == null) {
|
||||
+ // PATCH(docmost #OOM): no output strategy requested -> publish each
|
||||
+ // text-delta immediately and do NOT build cumulative partialOutput
|
||||
+ // snapshots. Unpatched, the default text() output snapshots the ENTIRE
|
||||
+ // accumulated turn text on every delta (O(n^2) memory) and those
|
||||
+ // snapshots pile up in the never-consumed leftover tee branch of
|
||||
+ // DefaultStreamTextResult.baseStream -> heap OOM on long agent turns.
|
||||
+ publishTextChunk({ controller });
|
||||
+ return;
|
||||
+ }
|
||||
+ text2 += chunk.text;
|
||||
const result = await output.parsePartialOutput({ text: text2 });
|
||||
if (result !== void 0) {
|
||||
const currentJson = JSON.stringify(result.partial);
|
||||
@@ -6959,7 +6969,7 @@ var DefaultStreamTextResult = class {
|
||||
})
|
||||
);
|
||||
}
|
||||
- this.baseStream = stream.pipeThrough(createOutputTransformStream(output != null ? output : text())).pipeThrough(eventProcessor);
|
||||
+ this.baseStream = stream.pipeThrough(createOutputTransformStream(output)).pipeThrough(eventProcessor);
|
||||
const { maxRetries, retry } = prepareRetries({
|
||||
maxRetries: maxRetriesArg,
|
||||
abortSignal
|
||||
diff --git a/dist/index.mjs b/dist/index.mjs
|
||||
index 663875332e3f9a9bd167c25583c515876f42951b..b840b0502c9894df983e0154805abb80e70e6331 100644
|
||||
--- a/dist/index.mjs
|
||||
+++ b/dist/index.mjs
|
||||
@@ -6501,9 +6501,19 @@ function createOutputTransformStream(output) {
|
||||
controller.enqueue({ part: chunk, partialOutput: void 0 });
|
||||
return;
|
||||
}
|
||||
- text2 += chunk.text;
|
||||
textChunk += chunk.text;
|
||||
textProviderMetadata = (_a21 = chunk.providerMetadata) != null ? _a21 : textProviderMetadata;
|
||||
+ if (output == null) {
|
||||
+ // PATCH(docmost #OOM): no output strategy requested -> publish each
|
||||
+ // text-delta immediately and do NOT build cumulative partialOutput
|
||||
+ // snapshots. Unpatched, the default text() output snapshots the ENTIRE
|
||||
+ // accumulated turn text on every delta (O(n^2) memory) and those
|
||||
+ // snapshots pile up in the never-consumed leftover tee branch of
|
||||
+ // DefaultStreamTextResult.baseStream -> heap OOM on long agent turns.
|
||||
+ publishTextChunk({ controller });
|
||||
+ return;
|
||||
+ }
|
||||
+ text2 += chunk.text;
|
||||
const result = await output.parsePartialOutput({ text: text2 });
|
||||
if (result !== void 0) {
|
||||
const currentJson = JSON.stringify(result.partial);
|
||||
@@ -6882,7 +6892,7 @@ var DefaultStreamTextResult = class {
|
||||
})
|
||||
);
|
||||
}
|
||||
- this.baseStream = stream.pipeThrough(createOutputTransformStream(output != null ? output : text())).pipeThrough(eventProcessor);
|
||||
+ this.baseStream = stream.pipeThrough(createOutputTransformStream(output)).pipeThrough(eventProcessor);
|
||||
const { maxRetries, retry } = prepareRetries({
|
||||
maxRetries: maxRetriesArg,
|
||||
abortSignal
|
||||
Generated
+5
-8
@@ -44,9 +44,6 @@ overrides:
|
||||
ip-address: 10.1.1
|
||||
|
||||
patchedDependencies:
|
||||
ai@6.0.134:
|
||||
hash: f60bfc3357e01e1f3978c6c40fdd65aeb33fefaad7179cde8676465b6c5ff4d9
|
||||
path: patches/ai@6.0.134.patch
|
||||
scimmy@1.3.5:
|
||||
hash: 775d80f86830b2c5dd1a250c9802c10f8fc3da3c7898373de5aa0c23993d1673
|
||||
path: patches/scimmy@1.3.5.patch
|
||||
@@ -626,10 +623,10 @@ importers:
|
||||
version: 8.3.0(socket.io-adapter@2.5.4)
|
||||
ai:
|
||||
specifier: ^6.0.134
|
||||
version: 6.0.134(patch_hash=f60bfc3357e01e1f3978c6c40fdd65aeb33fefaad7179cde8676465b6c5ff4d9)(zod@4.3.6)
|
||||
version: 6.0.134(zod@4.3.6)
|
||||
ai-sdk-ollama:
|
||||
specifier: ^3.8.1
|
||||
version: 3.8.1(ai@6.0.134(patch_hash=f60bfc3357e01e1f3978c6c40fdd65aeb33fefaad7179cde8676465b6c5ff4d9)(zod@4.3.6))(zod@4.3.6)
|
||||
version: 3.8.1(ai@6.0.134(zod@4.3.6))(zod@4.3.6)
|
||||
bcrypt:
|
||||
specifier: ^6.0.0
|
||||
version: 6.0.0
|
||||
@@ -16358,17 +16355,17 @@ snapshots:
|
||||
|
||||
agent-base@7.1.4: {}
|
||||
|
||||
ai-sdk-ollama@3.8.1(ai@6.0.134(patch_hash=f60bfc3357e01e1f3978c6c40fdd65aeb33fefaad7179cde8676465b6c5ff4d9)(zod@4.3.6))(zod@4.3.6):
|
||||
ai-sdk-ollama@3.8.1(ai@6.0.134(zod@4.3.6))(zod@4.3.6):
|
||||
dependencies:
|
||||
'@ai-sdk/provider': 3.0.8
|
||||
'@ai-sdk/provider-utils': 4.0.21(zod@4.3.6)
|
||||
ai: 6.0.134(patch_hash=f60bfc3357e01e1f3978c6c40fdd65aeb33fefaad7179cde8676465b6c5ff4d9)(zod@4.3.6)
|
||||
ai: 6.0.134(zod@4.3.6)
|
||||
jsonrepair: 3.13.3
|
||||
ollama: 0.6.3
|
||||
transitivePeerDependencies:
|
||||
- zod
|
||||
|
||||
ai@6.0.134(patch_hash=f60bfc3357e01e1f3978c6c40fdd65aeb33fefaad7179cde8676465b6c5ff4d9)(zod@4.3.6):
|
||||
ai@6.0.134(zod@4.3.6):
|
||||
dependencies:
|
||||
'@ai-sdk/gateway': 3.0.77(zod@4.3.6)
|
||||
'@ai-sdk/provider': 3.0.8
|
||||
|
||||
Reference in New Issue
Block a user