Learn: per-lesson URLs, a :::data shared-dataset construct, and a linked-views deep dive

This commit is contained in:
2026-06-28 12:34:30 +03:00
parent fd5c51a761
commit 20e70bee0e
16 changed files with 1818 additions and 178 deletions
+81 -15
View File
@@ -22,11 +22,15 @@
*
* Free intro prose (markdown)…
*
* :::data
* { "ev": [ {…row…}, {…row…} ] }
* :::
*
* :::progression
* ## raw counts
* Note prose for this stage (markdown)…
* ```vega-lite
* { …spec… }
* { "data": { "name": "ev" }, …spec… }
* ```
* ## + bin
* …
@@ -42,10 +46,13 @@
* A callout (markdown)…
* :::
*
* A `:::progression` block holds `##`-delimited stages (heading → tab label, prose
* → note, fenced `vega-lite` → spec). A bare fenced `vega-lite` block is a
* standalone chart. Any other `:::name` block is a callout. Everything else is
* prose, emitted in the order it appears.
* A `:::data` block names datasets once for the whole lesson (body: a JSON object
* of `name → rows`); specs reference them by `{ "name": … }` and the rows are
* injected at render time, so they aren't repeated per stage. A `:::progression`
* block holds `##`-delimited stages (heading → tab label, prose → note, fenced
* `vega-lite` → spec). A bare fenced `vega-lite` block is a standalone chart. Any
* other `:::name` block is a callout. Everything else is prose, emitted in the
* order it appears.
*/
export interface ProgressionStage {
@@ -81,6 +88,12 @@ export interface ParsedLesson {
slug: string;
title: string;
tagline: string;
/**
* Lesson-level named datasets, collected from `:::data` blocks. Specs reference
* a dataset by `{ "data": { "name": … } }`; {@link injectDatasets} merges these
* in at render time so the rows aren't repeated in every fenced spec.
*/
datasets: Record<string, unknown[]>;
/** The lesson body, in document order. */
blocks: LessonBlock[];
}
@@ -100,14 +113,25 @@ export function parseLesson(source: string): ParsedLesson {
if (!fm) throw new Error('lesson: missing `---` frontmatter block');
const meta = parseFrontmatter(fm[1]);
const slug = required(meta, 'slug');
// The slug is the lesson's URL segment (`/learn/<slug>/`) and the route key; keep it
// to the charset the router and the page-shell generator both assume, so a typo fails
// loudly here rather than silently serving the wrong page.
if (!/^[a-z0-9-]+$/.test(slug)) {
throw new Error(`lesson "${slug}": slug must be lowercase letters, digits, and hyphens`);
}
const title = required(meta, 'title');
const tagline = required(meta, 'tagline');
return { slug, title, tagline, blocks: parseBlocks(text.slice(fm[0].length), slug) };
const { blocks, datasets } = parseBlocks(text.slice(fm[0].length), slug);
return { slug, title, tagline, datasets, blocks };
}
function parseBlocks(body: string, slug: string): LessonBlock[] {
function parseBlocks(
body: string,
slug: string,
): { blocks: LessonBlock[]; datasets: Record<string, unknown[]> } {
const blocks: LessonBlock[] = [];
const datasets: Record<string, unknown[]> = {};
const pushProse = (text: string): void => {
const trimmed = text.trim();
if (trimmed) blocks.push({ type: 'prose', markdown: trimmed });
@@ -121,13 +145,15 @@ function parseBlocks(body: string, slug: string): LessonBlock[] {
if (m[1] !== undefined) {
const variant = m[1];
const inner = m[2];
// TODO: if lessons multiply, add a `:::data` directive whose values stages
// inherit — today a lesson's inline data is repeated in every fenced spec.
blocks.push(
variant === 'progression'
? { type: 'progression', stages: parseStages(inner, slug) }
: { type: 'callout', variant, markdown: inner.trim() },
);
if (variant === 'progression') {
blocks.push({ type: 'progression', stages: parseStages(inner, slug) });
} else if (variant === 'data') {
// Lesson metadata, not a rendered block: fold the named rows into the
// lesson's dataset map for render-time injection.
Object.assign(datasets, parseDatasets(inner, slug));
} else {
blocks.push({ type: 'callout', variant, markdown: inner.trim() });
}
} else {
blocks.push({ type: 'chart', spec: parseSpec(m[3], slug, 'chart') });
}
@@ -136,7 +162,47 @@ function parseBlocks(body: string, slug: string): LessonBlock[] {
pushProse(body.slice(cursor));
if (blocks.length === 0) throw new Error(`lesson "${slug}": no content`);
return blocks;
return { blocks, datasets };
}
/** Parse a `:::data` body — a JSON object of `name → rows[]` — into named datasets. */
function parseDatasets(inner: string, slug: string): Record<string, unknown[]> {
let parsed: unknown;
try {
parsed = JSON.parse(inner.trim());
} catch (e) {
throw new Error(`lesson "${slug}", :::data: invalid JSON — ${(e as Error).message}`, {
cause: e,
});
}
if (typeof parsed !== 'object' || parsed === null || Array.isArray(parsed)) {
throw new Error(`lesson "${slug}", :::data: body must be a JSON object of { name: rows }`);
}
const out: Record<string, unknown[]> = {};
for (const [name, rows] of Object.entries(parsed)) {
if (!Array.isArray(rows)) {
throw new Error(`lesson "${slug}", :::data "${name}": value must be an array of rows`);
}
out[name] = rows;
}
return out;
}
/**
* Merge lesson-level named datasets into a spec for *rendering* — the authored
* spec (shown in the source pane and diffed) keeps its `{ "data": { "name": … } }`
* reference and never carries the rows. Returns a clone with the spec's own
* `datasets` winning on a name clash; mirrors the clone-before-embed rule the app's
* other render-time spec tweaks follow (never mutate the authored spec). A no-op
* (same reference) when there are no datasets to inject.
*/
export function injectDatasets(
spec: Record<string, unknown>,
datasets: Record<string, unknown[]>,
): Record<string, unknown> {
if (Object.keys(datasets).length === 0) return spec;
const own = (spec.datasets as Record<string, unknown> | undefined) ?? {};
return { ...spec, datasets: { ...datasets, ...own } };
}
function parseStages(inner: string, slug: string): ProgressionStage[] {