takaebato commented on code in PR #497:
URL: https://github.com/apache/echarts-doc/pull/497#discussion_r3035183164
##########
build/build-llms.js:
##########
@@ -0,0 +1,350 @@
+/**
+ * Converts built part JSONs (HTML desc) to Markdown using turndown,
+ * and generates llms.txt + individual .md files.
+ *
+ * Mechanically converts documents/*-parts/*.json to llms-documents/ (.md
files).
+ * Root files (e.g. option.md) are placed at llms-documents/, while part files
+ * (e.g. option.title.md) are placed at llms-documents/*-parts/.
+ * Type information is extracted from documents/*.json (full schema) via
traverse.
+ *
+ * Prerequisites: JSON must be built first (node build.js --env dev)
+ * Usage: node build/build-llms.js --env dev
+ */
+const fs = require('fs');
+const fse = require('fs-extra');
+const path = require('path');
+const globby = require('globby');
+const TurndownService = require('turndown');
+const {gfm} = require('turndown-plugin-gfm');
+const {traverse} = require('../tool/schemaHelper');
+const {readConfigEnvFile} = require('./helper');
+
+// --- Constants ---
+
+const LANGUAGES = ['en', 'zh'];
+const OUTPUT_DIR_NAME = 'llms-documents';
+const MAX_HEADING_DEPTH = 6;
+
+const SECTION_LABELS = {
+ en: {'option-parts': 'Option', 'option-gl-parts': 'Option GL',
'api-parts': 'API', 'tutorial-parts': 'Tutorial'},
+ zh: {'option-parts': '配置项 (Option)', 'option-gl-parts': 'Option GL',
'api-parts': 'API', 'tutorial-parts': '教程 (Tutorial)'}
+};
+
+const LLMS_TXT_HEADER = [
+ '# Apache ECharts Documentation',
+ '',
+ '> Apache ECharts is a free, powerful charting and visualization library
offering easy ways to add intuitive, interactive, and highly customizable
charts to your commercial products.',
+ ''
+].join('\n');
+
+// --- Config ---
+
+const argv = require('yargs').argv;
+const envType = (argv.dev != null || argv.debug != null || argv.env === 'dev')
? 'dev' : argv.env;
+if (!envType) throw new Error('--env MUST be specified');
+const config = readConfigEnvFile(envType);
+
+// --- Turndown ---
+
+const td = new TurndownService({headingStyle: 'atx', codeBlockStyle:
'fenced'});
+td.use(gfm);
+td.addRule('iframe', {filter: 'iframe', replacement: () => ''});
+
+function htmlToMd(html) {
+ return html ? td.turndown(html).replace(/\n{3,}/g, '\n\n').trim() : '';
+}
+
+// --- Extract type info from full schema JSON ---
+
+/**
+ * Extract type and default value info from a full schema JSON by traversing
+ * the nested schema tree.
+ *
+ * @param {string} schemaJsonPath - path to schema JSON (e.g.
"documents/option.json")
+ * @param {string} docName - e.g. "option", "api"
+ * @returns {Object<string, {type: string|null, default: string|null}>}
+ * e.g. { "option.title.show": {type: "boolean", default: "true"} }
+ */
+function buildTypeMap(schemaJsonPath, docName) {
+ if (!fs.existsSync(schemaJsonPath)) return {};
+ const schema = JSON.parse(fs.readFileSync(schemaJsonPath, 'utf-8'));
+ const typeMap = {};
+ traverse(schema, docName, (schemaPath, node) => {
+ if (node.type || node.default != null) {
+ typeMap[schemaPath] = {
+ type: node.type ? (Array.isArray(node.type) ?
node.type.join('|') : node.type) : null,
+ default: node.default != null ? String(node.default) : null
+ };
+ }
+ });
+ return typeMap;
+}
+
+// --- Resolve links in HTML ---
+// Best-effort rewriting of <a href="#path"> and <a href="api.html#path"> in
HTML
+// so that turndown produces markdown links pointing to the correct .md files.
+// Some source links have non-standard formats (e.g. missing "#", no dot
separator)
+// that cannot be resolved; these are left as-is or linked to the root file.
+
+/**
+ * Split linkPath into a part key (first segment) and fragment (rest), matching
+ * the key against partKeys with case-insensitive and singular/plural fallback.
+ *
+ * @param {string} linkPath - e.g. "title.show", "echarts.init"
+ * @param {Set<string>} partKeys - e.g. Set{'title','series-bar','geo',...}
+ * @returns {{key: string, frag: string|null}|null}
+ * e.g. "title.show" -> {key: "title", frag: "show"}
+ * "angleAxis.axisLabel.interval" -> {key: "angleAxis", frag:
"axisLabel.interval"}
+ * "geo" -> {key: "geo", frag: null}
+ * "unknown" -> null
+ */
+function tryResolvePartKey(linkPath, partKeys) {
+ const [seg, ...rest] = linkPath.split('.');
+ const frag = rest.length > 0 ? rest.join('.') : null;
+
+ if (partKeys.has(seg)) return {key: seg, frag};
+
+ // Fallback: case-insensitive and singular/plural matching
+ const segL = seg.toLowerCase();
+ for (const k of partKeys) {
+ if (k.toLowerCase() === segL) return {key: k, frag};
+ }
+ for (const k of partKeys) {
+ const kl = k.toLowerCase();
+ if (kl === segL + 's' || kl + 's' === segL) return {key: k, frag};
+ }
+ return null;
+}
+
+/**
+ * Resolve a link path to an href pointing to the correct .md file.
+ * If partKeys contains a match, link to the individual part file;
+ * otherwise fall back to the root file.
+ *
+ * @param {string} linkPath - e.g. "title.show", "visualMap"
+ * @param {Set<string>} partKeys - keys of individual part files
+ * @param {string} pathPrefix - path prefix for part files
+ * same-doc: "option" -> "option.title.md"
+ * cross-doc: "../api-parts/api" -> "../api-parts/api.echarts.md"
+ * @param {string|null} rootPath - path prefix for root file fallback
+ * same-doc: "../option" -> "../option.md#visualMap"
+ * cross-doc: "../api" -> "../api.md#events"
+ * @returns {string|null} resolved href attribute string, or null
+ */
+function resolveLink(linkPath, partKeys, pathPrefix, rootPath) {
+ const resolved = tryResolvePartKey(linkPath, partKeys);
+ if (!resolved) {
+ if (rootPath) return `href="${rootPath}.md#${linkPath}"`;
+ return null;
+ }
+ return `href="${pathPrefix}.${resolved.key}.md${resolved.frag ? '#' +
resolved.frag : ''}"`;
+}
+
+/**
+ * Rewrite internal links in HTML so that turndown produces correct .md links.
+ * Handles two patterns:
+ * 1. Same-doc: href="#title.show" -> href="option.title.md#show"
+ * 2. Cross-doc: href="api.html#echarts.init" ->
href="../api-parts/api.echarts.md#init"
+ * Unresolvable links are left as-is or fall back to the root file.
+ *
+ * @param {string} html - HTML string containing <a href="..."> links
+ * @param {Object<string, Set<string>>} partKeysByDoc - part keys for all docs
+ * @param {string} docName - current doc name (e.g. "option")
+ * @returns {string} HTML with rewritten href attributes
+ */
+function tryResolveHtmlLinks(html, partKeysByDoc, docName) {
+ const partKeys = partKeysByDoc[docName];
+
+ // Same-doc links: href="#title.show" -> href="option.title.md#show"
+ const resolved = html.replace(/href="#([^"]+)"/g, (match, linkPath) =>
+ (partKeys && resolveLink(linkPath, partKeys, docName,
`../${docName}`)) || match
+ );
+
+ // Cross-doc links: href="api.html#echarts.init" ->
href="../api-parts/api.echarts.md#init"
+ return resolved.replace(
+ /href="(option-gl|option|api|tutorial)\.html#([^"]+)"/g,
+ (match, targetDoc, fragment) => {
+ const keys = partKeysByDoc[targetDoc];
+ if (!keys) return match;
+ return resolveLink(fragment, keys,
`../${targetDoc}-parts/${targetDoc}`, `../${targetDoc}`) || match;
+ }
+ );
+}
+
+// --- Convert part JSON to Markdown ---
+
+function formatPropertyEntry(key, entry, typeInfo, linkResolver) {
+ const heading = '#'.repeat(Math.min(key.split('.').length + 1,
MAX_HEADING_DEPTH)) + ' ' + key;
+ const meta = [
+ typeInfo && typeInfo.type && `- **Type**: \`${typeInfo.type}\``,
+ typeInfo && typeInfo.default != null && `- **Default**:
\`${typeInfo.default}\``
+ ].filter(Boolean);
+ const body = entry.desc ? htmlToMd(linkResolver(entry.desc)) : '';
+ return [heading, ...meta, ...(body ? ['', body] : []), ''];
+}
+
+function jsonToMd(data, typeMap, baseName, linkResolver) {
+ const lines = Object.entries(data).flatMap(([key, entry]) => {
+ const fullKey = baseName ? `${baseName}.${key}` : key;
+ return formatPropertyEntry(key, entry, typeMap[fullKey], linkResolver);
Review Comment:
This seems like a trade-off.
Since this is primarily LLM-facing documentation rather than
browser-rendered content, not adding explicit HTML anchors to keep the Markdown
clean seems better.
The heading text preserves the original property path (e.g. ###
textStyle.color), which should be sufficient for LLM consumption.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
