edfcore 0.4.229 → 0.4.237

package/dist/constants.d.ts

@@ -109,5 +109,5 @@ export declare const SIGNAL_FIELD_BLOCK_OFFSETS: {
     readonly reserved: 224;
 };
 /** Published package version. Kept in sync with package.json by a test. */
-export declare const VERSION = "0.4.229";
+export declare const VERSION = "0.4.237";
 //# sourceMappingURL=constants.d.ts.map

package/dist/constants.js

@@ -79,5 +79,5 @@ export const SIGNAL_FIELD_BLOCK_OFFSETS = {
     reserved: 224,
 };
 /** Published package version. Kept in sync with package.json by a test. */
-export const VERSION = '0.4.229';
+export const VERSION = '0.4.237';
 //# sourceMappingURL=constants.js.map

package/docs/CHANGELOG.md

@@ -6,6 +6,115 @@ alone does not tell you whether you were affected.
 edfcore is pre-1.0. Patch releases have carried behaviour changes where the old behaviour was a
 defect; those are called out below.
 
+## 0.4.237
+
+- **Fixed** `npm run check` failing on any machine without the website's dependencies installed,
+  which is every CI runner. 0.4.231 added a check comparing the docs reader's glob pattern against
+  the collection loader's, and read `website/src/content.config.ts` through `import.meta.glob`
+  with `?raw`. A raw glob still hands the path to vite's transform, which resolves that file's
+  nearest tsconfig — `website/tsconfig.json`, which extends `astro/tsconfigs/strict` out of
+  `website/node_modules`. The CI `check` job installs the root workspace only, so the run died
+  with `[TSCONFIG_ERROR] Failed to load tsconfig 'astro/tsconfigs/strict'` while the same command
+  passed locally, where the site's dependencies happen to be present. Both files are read with
+  `readFileSync` now: bytes, no transform, no tsconfig.
+
+  **0.4.231 through 0.4.236 were never released.** Each was tagged and each publish run failed at
+  the check above, so six numbers are holes on npm the way 0.2.29, 0.2.36, 0.2.59 and 0.4.176 are.
+  Nothing is lost: every change they carried is in this release. The 0.4.200 revert cannot reach
+  this case — the checks passed on the machine cutting the tag, and it was the *runner's*
+  environment that differed, which is the gap 0.4.233 had just moved `publint` into CI to narrow
+  from the other side.
+
+## 0.4.236
+
+- **Added** a check that every internal link on the site points at something that exists. The
+  documentation pages carry over a hundred `/docs/...` and `#anchor` links between them and
+  nothing checked one: `astro check` validates types and content collections, not hrefs, and a
+  static build turns a link to a renamed page into a 404 for the reader rather than an error for
+  the author. The 404 page exists because that happens — "the address may have moved when the docs
+  were reorganised" — which is a good page to have and a poor substitute for not shipping the
+  link. Anchors are the half that rots quietly, since one breaks when someone rewords a heading
+  three sections away and the link still looks right; 0.4.234 nearly shipped exactly that, a table
+  cell pointing at `#patient-identification` on a page whose redaction note has no heading. The
+  nine links hard-coded in `.astro` routes are swept too — the 404's three ways out and the
+  landing page's four are the ones a reader hits first. Nothing was broken today.
+
+## 0.4.235
+
+- **Added** documentation for the three selection types, taking the recorded undocumented list
+  from eleven to eight. `StreamSelection`, `EnvelopeSelection` and `TriggerSelection` are what
+  `streamRecords`, `readEnvelope` and `readTriggers` take, and every page showed an object literal
+  without ever naming the type — so a wrapper that accepts one had nothing to import.
+  `api-types.md` lists all three in the selections table and `api-helpers.md` names each in the
+  section that teaches its call, including the fact that the first two are a `WindowSelection`
+  plus one field.
+- **Fixed** the sentence under that table, which said `signalIndices` is required "on both
+  selections". There are five now, and `TriggerSelection` is the one with no channel field at all
+  — `readTriggers` finds the BioSemi Status signal itself, because a 24-bit EEG sample decoded as
+  a trigger word yields plausible events out of ordinary data.
+
+## 0.4.234
+
+- **Added** documentation for the three formatter options types, and took them off the recorded
+  list of undocumented ones. `FormatHeaderOptions`, `FormatAnnotationsOptions` and
+  `FormatReportOptions` are exported and every field of each was described somewhere in prose, but
+  none of the three was named on a page — so writing a wrapper that accepts one, or building an
+  options object ahead of the call, meant reading the `.d.ts`. `api-helpers.md` now gives each a
+  field table with its default, and says why `includePatientId` defaults off while
+  `diagnosticsHint` defaults on: the cost of forgetting the first is a person's name in an issue
+  tracker, and the cost of forgetting the second is one redundant line. `UNDOCUMENTED_TYPES` is
+  down from fourteen to eleven, which is the direction 0.4.220 built it to move in.
+
+## 0.4.233
+
+- **Changed** the packaging checks to run on every push instead of only at publish time.
+  `publint --strict` and `@arethetypeswrong/cli` are the two things `npm run check` cannot do —
+  they read the manifest against the files npm would actually pack, and resolve each subpath the
+  way a consumer's TypeScript would — and they lived in `publish.yml`, which runs after the tag is
+  pushed. That is the one window `scripts/release.mjs` cannot undo, so a packaging mistake found
+  there could only be fixed by cutting another version. CI now has a `package` job, and both
+  workflows call the same `npm run verify:package` so the two cannot drift. It stays out of
+  `npm run check`: that one downloads nothing, and `git clone && npm test` being green offline is
+  a property worth keeping.
+
+## 0.4.232
+
+- **Fixed** the site's version sweep reading only `.astro` files. `website/src/pages/` also holds
+  seven `.ts` routes, and they emit prose exactly the way a component does: `llms.txt` is the map
+  an agent is handed, `[...slug].md.ts` is the markdown twin of every documentation page, and
+  `robots.txt` and `api.json` are served verbatim. A stale version in one of those reaches a
+  reader the same way the footer's "MIT licensed. Version 0.1.0." did for three minor series.
+  Widening it needed one more thing first: comments are now stripped before the scan, because
+  `api.json.ts` quotes that footer defect as the reason it counts the surface rather than stating
+  it, and a whole-file match finds the quotation rather than a claim. What a file emits is the
+  claim; what it says about the past is history. HTML comments are left in — those ship.
+
+## 0.4.231
+
+- **Fixed** four documentation sweeps reading a narrower set of pages than the site publishes.
+  `docs-coverage.test.ts`, `diagnostic-docs.test.ts` and `readme-status.test.ts` — the last of
+  them twice — each wrote its own reader, three globbing `content/docs/*.md` and one calling
+  `readdirSync(...).filter(name => name.endsWith('.md'))`. The collection loads
+  `**/*.{md,mdx}` and `astro.config.mjs` registers the MDX integration, so a page in a
+  subdirectory or written as `.mdx` was published and unswept. It fails in the unhelpful
+  direction too: a name documented only on an unseen page reports as undocumented, and a type on
+  the recorded `UNDOCUMENTED_TYPES` list stays there after the page documenting it is written.
+  All four now read `tests/support/docs-pages.ts`, and a new check compares that reader's glob
+  call against the loader's pattern, so narrowing either side fails rather than going quiet.
+
+## 0.4.230
+
+- **Added** a check on the three packaging promises in the README's compatibility list — zero
+  runtime dependencies, three entry points with no environment conditions in the exports map, and
+  ESM only. What the package *contains* is well covered: `api-surface.test.ts` counts the exports
+  and `public-api.test.ts` walks the module graph. Nothing read the manifest that says how it is
+  shipped. Adding a dependency is one `npm install --save` away and the tree stays green — nothing
+  imports it, so no test fails, and "zero-dependency" stops being true in the one place a reader
+  looks before installing. An environment condition next to `default` is the same shape of edit.
+  Every export and `bin` target is also checked to exist in the build; `publint` and
+  `@arethetypeswrong/cli` cover a stronger version of that, but they run in `publish.yml` only —
+  after the tag is pushed, in the window 0.4.226 is about.
+
 ## 0.4.229
 
 - **Added** a check that the Node floor is one number. It is written down eleven times — the

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "edfcore",
-  "version": "0.4.229",
+  "version": "0.4.237",
   "description": "Modern, typed, zero-dependency reader for EDF, EDF+, BDF and BDF+ biosignal files. Works in browsers and Node with true random access.",
   "keywords": [
     "edf",
@@ -70,6 +70,7 @@
     "lint": "biome check --error-on-warnings .",
     "format": "biome check --write --linter-enabled=false .",
     "check": "npm run lint && npm run typecheck && npm run build && npm run test",
+    "verify:package": "npx --yes publint@latest --strict && npx --yes @arethetypeswrong/cli@latest --pack . --profile esm-only",
     "corpus:fetch": "node scripts/fetch-corpus.mjs",
     "release": "node scripts/release.mjs",
     "prepublishOnly": "npm run check && npm run build"

package/src/constants.ts

@@ -93,4 +93,4 @@ export const SIGNAL_FIELD_BLOCK_OFFSETS = {
 } as const;
 
 /** Published package version. Kept in sync with package.json by a test. */
-export const VERSION = '0.4.229';
+export const VERSION = '0.4.237';