@baiyibai-antora/extensions 1.12.0 → 1.14.0

package/dist/extensions/trademark-it/index.js

@@ -83,16 +83,19 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.formatList = formatList;
 exports.parseTrademarks = parseTrademarks;
 exports.mergeTrademarks = mergeTrademarks;
+exports.stripComments = stripComments;
 exports.stripNonContent = stripNonContent;
 exports.extractSvgText = extractSvgText;
 exports.findTrademarks = findTrademarks;
 exports.generateTrademarkPartial = generateTrademarkPartial;
-exports.parseIncludeTarget = parseIncludeTarget;
 exports.resolveRelativeIncludePath = resolveRelativeIncludePath;
 exports.applyIncludeFilters = applyIncludeFilters;
 exports.collectIncludeText = collectIncludeText;
 exports.register = register;
+exports.getComponentVersions = getComponentVersions;
 const log_time_1 = require("../../lib/log-time");
+const asciidoc_table_1 = require("../../lib/asciidoc-table");
+const asciidoc_conditionals_1 = require("../../lib/asciidoc-conditionals");
 const fs = __importStar(require("node:fs"));
 const path = __importStar(require("node:path"));
 // ---------------------------------------------------------------------------
@@ -160,6 +163,28 @@ function ensureStringArray(v) {
         return v.split(',').map((s) => s.trim()).filter(Boolean);
     return [];
 }
+/**
+ * Locates this extension's own entry in `playbook.antora.extensions` by
+ * whether it carries `trademark_it`/`trademarkIt`/`trademarkit` as an own
+ * key, not by matching a substring of its `require` string. A require-
+ * string match breaks the moment a consumer vendors or re-exports this
+ * package under a different name or path: the require string then never
+ * contains "trademark-it"/"trademark_it" at all, `.find()` returns
+ * nothing, and every configured setting silently falls back to hardcoded
+ * defaults, no warning, no error (confirmed as a real report, not a
+ * hypothetical, and the same bug in cite-it/autoterm). Checking for the
+ * config key itself identifies the right entry regardless of how it was
+ * required; the require-string check remains only as a fallback for the
+ * case where no entry carries the key at all.
+ */
+function findOwnExtEntry(entries) {
+    return entries.find((ext) => 'trademark_it' in ext || 'trademarkIt' in ext || 'trademarkit' in ext)
+        ?? entries.find((ext) => {
+            const req = String(ext.require ?? '');
+            return req.includes('trademark-it') || req.includes('trademark_it');
+        })
+        ?? {};
+}
 /**
  * Build a case-insensitive membership test for an ignore list.  Trademark
  * scanning is case-insensitive, so ignore matching must be too; otherwise an
@@ -205,68 +230,19 @@ function superscriptSymbols(s) {
     return s.replace(/(®)/g, '^$1^');
 }
 /**
- * Parse an AsciiDoc table into an array of string-cell rows.
- * Handles the `|cell` style used in the trademark table.
- *
- * The table format is:
- *   |===
- *   |col0
- *   |col1
- *   ...
- *   |===
+ * Parse an AsciiDoc table into an array of string-cell rows, header row
+ * included as rows[0]. Handles the `|cell` style used in the trademark
+ * table: header cells inline on one line, data cells one per line.
  *
- * Rows are separated by blank lines between cell groups, OR by having
- * each cell on its own line (the style used in trademark-table.adoc).
- * We collect all `|…` tokens after the header row and group them by
- * the column count declared in the [cols=…] attribute.
+ * Delegates to the real Asciidoctor parser (see
+ * agent-memory/autoterm-trademark-it-api-rewrite-scope.md, Phase 1) rather
+ * than regexing the `|===` delimiters and `[cols=…]` attribute by hand.
  */
 function parseTrademarkTable(asciidocText) {
-    const rows = [];
-    // Find the table body between |=== delimiters.
-    const tableMatch = asciidocText.match(/\|={3,}([\s\S]*?)\|={3,}/);
-    if (!tableMatch)
-        return rows;
-    const tableBody = tableMatch[1];
-    // Determine column count from [cols="…"] attribute if present.
-    // Default to 5 (matching the trademark-table.adoc format).
-    let colCount = 5;
-    const colsMatch = asciidocText.match(/\[cols\s*=\s*["']([^"']+)["']/);
-    if (colsMatch) {
-        // Count comma-separated entries; handle repeat syntax like "3*1"
-        colCount = colsMatch[1].split(',').length;
-    }
-    // Extract all cell values.  The trademark table uses two styles:
-    //   • Header row — all cells inline on one line, each prefixed with |:
-    //       |Term |Statement |Date |Verified By |URL
-    //   • Data rows — one cell per line, each on its own line:
-    //       |Apple®
-    //       |Apple is a registered trademark…
-    //
-    // We detect inline multi-cell lines by splitting on | and checking whether
-    // more than one non-empty segment results.
-    const cells = [];
-    for (const raw of tableBody.split('\n')) {
-        const line = raw.trim();
-        if (!line || line === '|===')
-            continue;
-        if (!line.startsWith('|'))
-            continue;
-        // Split on | — the first segment before the leading | will be empty.
-        const parts = line.split('|').map((p) => p.trim()).filter((p) => p !== '');
-        if (parts.length > 1) {
-            // Inline multi-cell row (e.g. the header).
-            cells.push(...parts);
-        }
-        else {
-            // Single-cell-per-line row (all data rows).
-            cells.push(line.slice(1).trim());
-        }
-    }
-    // Group cells into rows of colCount.
-    for (let i = 0; i + colCount - 1 < cells.length; i += colCount) {
-        rows.push(cells.slice(i, i + colCount));
-    }
-    return rows;
+    const table = (0, asciidoc_table_1.parseAdocTable)(asciidocText);
+    if (!table)
+        return [];
+    return [...table.head, ...table.body];
 }
 /**
  * Parse the trademark definition AsciiDoc file and return a map of
@@ -430,12 +406,6 @@ function escapeRegex(s) {
         .replace(/[-\u2011]|\\\{nbhy\}/g, '(?:-|\u2011|\\{nbhy\\})')
         .replace(/[ \u00A0]/g, '(?:[ \u00A0])');
 }
-/**
- * Strip AsciiDoc content that should not be scanned for trademarks:
- *  - Lines beginning with // (single-line comments)
- *  - ifdef::/ifndef::/ifeval:: conditional directive lines (attribute names only,
- *    not the body content between the delimiters)
- */
 /**
  * Convert a glob pattern to a RegExp.
  * Supports * (any chars except /), ** (any chars including /), ? (single char).
@@ -453,79 +423,72 @@ function globToRegex(pattern) {
 function matchesAnyGlob(path, patterns) {
     return patterns.some((p) => globToRegex(p).test(path));
 }
-function stripNonContent(text, ignoreConditionals = []) {
-    const result = [];
+/**
+ * Strip AsciiDoc content that should not be scanned for trademarks: single-
+ * line comments, and unsatisfied ifdef::/ifndef:: conditional blocks
+ * evaluated against real attribute state.
+ *
+ * Delegates conditional evaluation to the shared, real-Asciidoctor-backed
+ * evaluator in src/lib/asciidoc-conditionals.ts (see
+ * agent-memory/trademark-it-project.md) rather than the ignore-list-only
+ * model this function used before it: an ifdef:: block now survives only
+ * when its attribute expression is genuinely satisfied, an ifndef:: block
+ * only when it genuinely is not — previously both were kept unconditionally
+ * regardless of real state, and only an ifdef:: block could be force-
+ * suppressed via ignoreConditionals (that force-list still works the same
+ * way; it now layers on top of real evaluation instead of being the only
+ * mechanism). ifeval:: content is kept, since expressions are not
+ * evaluated, matching the shared evaluator's own documented contract.
+ * Block comments (////) are now also handled by the shared evaluator's own
+ * real reader, which recognizes them natively; this function no longer
+ * tracks them itself.
+ *
+ * `isDefined` defaults to "nothing is defined": calling this without real
+ * attribute state to offer is conservative (an ifdef:: block is hidden, an
+ * ifndef:: block is shown) rather than reproducing the old "always keep"
+ * default, which was never actually correct.
+ *
+ * Comments, single-line and block, are stripped by this function itself,
+ * not the shared evaluator: verified empirically (not assumed) that
+ * Asciidoctor's raw preprocessor reader — `parse: false`, what the shared
+ * evaluator uses — does not recognize a `////` block comment as a unit at
+ * all. Block-comment recognition turned out to be a parser-level concern,
+ * only correctly handled during a full parse (confirmed separately for the
+ * real converter autoterm's indexterm scanner uses, which is unaffected).
+ * A directive-like line sitting inside an unstripped block comment would
+ * otherwise still be evaluated for real by the shared evaluator, corrupting
+ * content that is supposed to be entirely inert — caught by this file's own
+ * test suite, not assumed correct from Phase 2's autoterm work.
+ */
+/**
+ * Blanks `//` and `////`-delimited comment lines, preserving line count.
+ * Exported for cite-it, a second consumer that needs the same stripping
+ * before scanning page text for `citeit:[key]` usage.
+ */
+function stripComments(text) {
+    const out = [];
     let inBlockComment = false;
-    // Stack of suppressed conditional attribute names (for block-form ifdef/ifndef).
-    // Each entry is the attribute name that opened the suppressed block.
-    const suppressStack = [];
     for (const line of text.split('\n')) {
         const trimmed = line.trimStart();
-        // Block comment delimiter ////
         if (trimmed.startsWith('////')) {
             inBlockComment = !inBlockComment;
-            result.push('');
+            out.push('');
             continue;
         }
         if (inBlockComment) {
-            result.push('');
+            out.push('');
             continue;
         }
-        // Single-line comments
         if (trimmed.startsWith('//')) {
-            result.push('');
-            continue;
-        }
-        // endif:: — close the innermost suppressed conditional block (if any).
-        if (/^endif::/.test(trimmed)) {
-            if (suppressStack.length > 0)
-                suppressStack.pop();
-            result.push('');
-            continue;
-        }
-        // Conditional directive lines.
-        // Inline form: ifdef::attr[content] — keep content unless attr is ignored.
-        // Block form:  ifdef::attr[]        — open a block; suppress if attr is ignored.
-        const condMatch = trimmed.match(/^(i(?:fdef|fndef|feval))::([^\[]*)\[([^\]]*)\]/);
-        if (condMatch) {
-            const directive = condMatch[1]; // ifdef / ifndef / ifeval
-            const attrExpr = condMatch[2]; // attribute name(s), may contain + or ,
-            const body = condMatch[3]; // inline body (empty for block form)
-            if (body === '') {
-                // Block-form directive: push onto stack if any listed attr is in ignoreConditionals.
-                const attrs = attrExpr.split(/[+,]/).map((a) => a.trim()).filter(Boolean);
-                const shouldSuppress = ignoreConditionals.length > 0 &&
-                    (directive === 'ifdef'
-                        ? attrs.some((a) => ignoreConditionals.includes(a))
-                        : false); // only suppress ifdef:: blocks, not ifndef:: / ifeval::
-                if (shouldSuppress)
-                    suppressStack.push(attrExpr);
-                result.push('');
-            }
-            else {
-                // Inline form: keep body unless inside a suppressed block OR the
-                // directive's own attribute is in ignoreConditionals.
-                const inlineAttrs = attrExpr.split(/[+,]/).map((a) => a.trim()).filter(Boolean);
-                const inlineSuppressed = ignoreConditionals.length > 0 &&
-                    directive === 'ifdef' &&
-                    inlineAttrs.some((a) => ignoreConditionals.includes(a));
-                result.push((suppressStack.length === 0 && !inlineSuppressed) ? body : '');
-            }
-            continue;
-        }
-        // Bare block-form directive without brackets on this line.
-        if (/^i(?:fdef|fndef|feval)::/.test(trimmed)) {
-            result.push('');
-            continue;
-        }
-        // Suppress content inside a matched conditional block.
-        if (suppressStack.length > 0) {
-            result.push('');
+            out.push('');
             continue;
         }
-        result.push(line);
+        out.push(line);
     }
-    return result.join('\n');
+    return out.join('\n');
+}
+function stripNonContent(text, isDefined = () => false, ignoreConditionals = []) {
+    return (0, asciidoc_conditionals_1.suppressUnsetConditionals)(stripComments(text), isDefined, ignoreConditionals);
 }
 /**
  * Extract human-readable text content from an SVG file for trademark scanning.
@@ -536,6 +499,21 @@ function stripNonContent(text, ignoreConditionals = []) {
  * Attribute values (e.g. aria-label, inkscape:label) that commonly carry
  * descriptive text are also extracted.
  *
+ * <metadata>...</metadata> blocks are removed before extraction. The SVG
+ * spec defines <metadata> as a container for non-visual, non-rendered
+ * descriptive data (typically RDF/Dublin Core), never diagram content, so a
+ * false trademark hit inside one is never a real occurrence — confirmed
+ * against the spec (https://www.w3.org/TR/SVG2/struct.html#MetadataElement)
+ * rather than assumed. sodipodi:docname and content are excluded from the
+ * scanned attribute list for the same reason under a different guise:
+ * sodipodi:docname is Inkscape's saved filename, not diagram content, and
+ * content is not a real SVG attribute at all — draw.io/diagrams.net uses it
+ * to carry that tool's entire private editor-state document as one
+ * attribute value. Both are exactly the kind of export-tool metadata a
+ * reader never sees, reported as a false trademark 'hit' in practice
+ * (a saved filename like screenshot-chrome-windows.svg matching "Chrome"
+ * and "Windows" is what prompted this fix).
+ *
  * All XML tags are stripped; entity references (&amp; &lt; &gt; &quot;)
  * are decoded so that trademark terms are not split by encoding artefacts.
  * Returns a plain text string safe to pass to findTrademarks.
@@ -559,24 +537,59 @@ function extractSvgText(svgSource) {
     const source = /&lt;\s*\/?[a-zA-Z]/.test(svgSource)
         ? decodeXmlEntities(svgSource)
         : svgSource;
+    // Remove XML comments and non-visual containers before any extraction
+    // runs. Comments are never rendered, but the element/attribute regexes
+    // below have no concept of comment boundaries: prose inside a comment
+    // that happens to mention a real tag name (e.g. a fixture's own
+    // "- <title> element" documentation of itself) can extend a later
+    // element's lazy match all the way back through the comment, capturing
+    // everything in between — reproduced against this package's own SVG test
+    // fixture. <metadata> is excluded per the SVG spec's own definition of it
+    // as a non-rendered descriptive container (RDF/Dublin Core is the common
+    // real-world content) — https://www.w3.org/TR/SVG2/struct.html#MetadataElement.
+    // <script> and <style> are excluded because neither is ever visible text
+    // by definition — script is behavior, style is presentation — and both
+    // commonly carry literal browser/OS names as real content rather than
+    // incidental noise: vendor-prefixed CSS, OS-specific font-family
+    // fallbacks ("Segoe UI" is Windows-specific), and browser-sniffing script
+    // (checking navigator.userAgent for "Chrome") are all standard,
+    // unremarkable SVG authoring practice, not something a reader is meant to
+    // read as prose. Their content is also the standard real-world use of a
+    // CDATA section in SVG, wrapped precisely because it contains characters
+    // XML would otherwise try to parse.
+    // <script> and <style> are stripped wholesale above, which removes their
+    // CDATA-wrapped content along with them — CDATA's standard real-world use
+    // in SVG. What's unwrapped here instead is CDATA appearing in genuinely
+    // visible content (e.g. <desc><![CDATA[Real text]]></desc>): the final
+    // "strip any remaining tag" step below would otherwise treat the whole
+    // `<![CDATA[...]]>` span as one tag (it has no `>` until its own closing
+    // `]]>`) and discard the real text inside it entirely, not just the
+    // marker syntax.
+    const cleaned = source
+        .replace(/<!--[\s\S]*?-->/g, ' ')
+        .replace(/<metadata[^>]*>[\s\S]*?<\/metadata>/gi, ' ')
+        .replace(/<script[^>]*>[\s\S]*?<\/script>/gi, ' ')
+        .replace(/<style[^>]*>[\s\S]*?<\/style>/gi, ' ')
+        .replace(/<!\[CDATA\[([\s\S]*?)\]\]>/g, '$1');
     const textElements = ['text', 'tspan', 'title', 'desc', 'textPath'];
-    const textAttrs = ['aria-label', 'inkscape:label', 'sodipodi:docname', 'content'];
+    const textAttrs = ['aria-label', 'inkscape:label'];
     const chunks = [];
     // Extract content of text-bearing elements.
     for (const tag of textElements) {
         const re = new RegExp(`<${tag}[^>]*>([\\s\\S]*?)<\\/${tag}>`, 'gi');
         let m;
-        while ((m = re.exec(source)) !== null) {
+        while ((m = re.exec(cleaned)) !== null) {
             chunks.push(m[1]);
         }
     }
     // Extract named attributes that carry descriptive text.
     for (const attr of textAttrs) {
-        // Leading boundary so short attribute names (e.g. "content") do not match
-        // the tail of longer ones (e.g. "data-content", "og:content").
+        // Leading boundary so a short attribute name does not match the tail of
+        // a longer one (e.g. a hypothetical "label" would otherwise also match
+        // "aria-label" and "inkscape:label" themselves).
         const re = new RegExp(`(?<![\\w:-])${attr.replace(':', '\\:')}\\s*=\\s*["']([^"']+)["']`, 'gi');
         let m;
-        while ((m = re.exec(source)) !== null) {
+        while ((m = re.exec(cleaned)) !== null) {
             chunks.push(m[1]);
         }
     }
@@ -589,9 +602,9 @@ function extractSvgText(svgSource) {
  * terms that were actually found.  Every member of a group is tested so that
  * buildable declarations can be assembled from only the terms present.
  */
-function findTrademarks(sourceText, trademarks, ignoreConditionals = []) {
+function findTrademarks(sourceText, trademarks, isDefined = () => false, ignoreConditionals = []) {
     const found = new Map();
-    const scanText = stripNonContent(sourceText, ignoreConditionals);
+    const scanText = stripNonContent(sourceText, isDefined, ignoreConditionals);
     for (const entry of trademarks.values()) {
         // Check the canonical base term and every alias.
         const termsToCheck = entry.combinedTerms;
@@ -616,9 +629,9 @@ function findTrademarks(sourceText, trademarks, ignoreConditionals = []) {
  * A file using several members of one group therefore reports each of them, not just the first.
  * Used for detailed debugging to track the origin of each found trademark.
  */
-function findTrademarksInFile(sourceText, trademarks, ignoreConditionals = []) {
+function findTrademarksInFile(sourceText, trademarks, isDefined = () => false, ignoreConditionals = []) {
     const results = [];
-    const scanText = stripNonContent(sourceText, ignoreConditionals);
+    const scanText = stripNonContent(sourceText, isDefined, ignoreConditionals);
     // Locate first-occurrence line numbers on the stripped lines (not raw) so
     // that line numbers point to real content, not to comment lines where the
     // term may appear before its actual use.  stripNonContent preserves line
@@ -855,88 +868,14 @@ function generateTrademarkPartial(foundBaseTerms, trademarks, cols = '22,78', le
 // scanning the including component.
 // ---------------------------------------------------------------------------
 /**
- * Parse an Antora resource ID from an include:: directive target.
- *
- * Supported formats:
- *   [version@][component:][module:]family$relative
- *   [version@][component:]module:relative        (implicit family, family = '')
- *
- * Missing parts fall back to the provided context defaults.  An empty family
- * tells the catalog lookup to match any family.
- * Returns null for plain filesystem paths (no '$' and no ':' present).
- */
-function parseIncludeTarget(target, defaultComponent, defaultVersion, defaultModule) {
-    let ref = target.trim();
-    // version@ prefix
-    let version = defaultVersion;
-    const atIdx = ref.indexOf('@');
-    if (atIdx >= 0) {
-        version = ref.slice(0, atIdx);
-        ref = ref.slice(atIdx + 1);
-    }
-    // '$' is the family/relative separator — present in standard Antora resource IDs.
-    const dollarIdx = ref.lastIndexOf('$');
-    if (dollarIdx >= 0) {
-        // Standard Antora resource ID with explicit family: [component:][module:]family$relative
-        const relative = ref.slice(dollarIdx + 1);
-        const prefixAndFamily = ref.slice(0, dollarIdx);
-        const lastColon = prefixAndFamily.lastIndexOf(':');
-        const family = lastColon >= 0 ? prefixAndFamily.slice(lastColon + 1) : prefixAndFamily;
-        const prefix = lastColon >= 0 ? prefixAndFamily.slice(0, lastColon) : '';
-        // prefix = '' | 'module' | 'component:module'
-        const parts = prefix ? prefix.split(':') : [];
-        let component = defaultComponent;
-        let module = defaultModule;
-        if (parts.length >= 2) {
-            component = parts[0];
-            module = parts[1];
-        }
-        else if (parts.length === 1) {
-            module = parts[0];
-        }
-        return { component, version, module, family, relative };
-    }
-    // No '$': a plain filesystem path has no colons; a colon-separated string is
-    // an Antora resource ID with an implicit family (e.g. the page shorthand
-    // 'topazu:overview:overview.adoc' used by delegating page stubs).
-    if (!ref.includes(':'))
-        return null;
-    // Parse as [component:]module:relative with family left empty so the catalog
-    // lookup in collectIncludeText can match any family.
-    const colonParts = ref.split(':');
-    let component = defaultComponent;
-    let module = defaultModule;
-    const relative = colonParts[colonParts.length - 1];
-    if (colonParts.length >= 3) {
-        component = colonParts[0];
-        module = colonParts[1];
-    }
-    else if (colonParts.length === 2) {
-        module = colonParts[0];
-    }
-    // family: '' signals the caller to match any family in the catalog.
-    return { component, version, module, family: '', relative };
-}
-/**
- * Normalize a slash-separated path, resolving '.' and '..' segments.
- * Returns null when '..' climbs above the root, which marks the target as
- * outside the content tree.
+ * A target that looks like an Antora resource ID rather than a plain
+ * filesystem path: it carries a version@ prefix, a component/module
+ * qualifier, or a family$relative separator. Matches the exact detection
+ * `@antora/asciidoc-loader` itself uses to choose between resolveResource
+ * and getByPath, kept here as one small regex rather than a dependency, per
+ * agent-memory/autoterm-trademark-it-api-rewrite-scope.md (Phase 3).
  */
-function normalizePathSegments(path) {
-    const out = [];
-    for (const seg of path.split('/')) {
-        if (seg === '' || seg === '.')
-            continue;
-        if (seg === '..') {
-            if (out.length === 0)
-                return null;
-            out.pop();
-            continue;
-        }
-        out.push(seg);
-    }
-    return out.join('/');
-}
+const RESOURCE_ID_RE = /[$:@]/;
 /**
  * Resolve a plain relative include target against the including file's
  * catalog path, mirroring Asciidoctor's resolution of filesystem-style
@@ -949,7 +888,8 @@ function resolveRelativeIncludePath(target, currentPath) {
     if (!currentPath)
         return null;
     const dir = currentPath.split('/').slice(0, -1).join('/');
-    return normalizePathSegments(`${dir}/${target.trim()}`);
+    const candidate = path.posix.join(dir, target.trim());
+    return candidate.startsWith('..') ? null : candidate;
 }
 /**
  * Apply the tag=, tags=, and lines= attributes of an include:: directive to
@@ -1069,12 +1009,22 @@ function applyIncludeFilters(text, attrlist) {
  * Recursively collect the text of all files referenced via include:: directives
  * in `sourceText`, looking each up in the content catalog.
  *
+ * File resolution delegates to Antora's own ContentCatalog#resolveResource
+ * (resource-ID targets: version@, component:module:family$relative, and the
+ * familyless page-shorthand form) and #getByPath (plain relative targets,
+ * resolved against `currentPath` first). See
+ * agent-memory/autoterm-trademark-it-api-rewrite-scope.md (Phase 3) for why:
+ * both are Antora's real, public, documented resolution logic, not a
+ * reimplementation of its resource-ID grammar. Everything else here —
+ * recursion, cycle/dedup tracking via `visited`, tag=/tags=/lines= filtering
+ * via applyIncludeFilters — is unchanged; only file lookup moved.
+ *
  * `visited` prevents revisiting files already being scanned or already resolved
  * in a previous recursion step.  Callers should pre-populate it with the keys
  * of all files already in the main scan set.
  *
- * Cross-component includes without an explicit version@ prefix are resolved by
- * matching any version in the catalog, mirroring Antora's "latest" resolution.
+ * Cross-component includes without an explicit version@ prefix resolve to
+ * that component's latest version, per resolveResource's own semantics.
  *
  * `scanExclude` glob patterns (same as the top-level scan_exclude config) are
  * applied to each resolved file's path.  Matched files are skipped and added to
@@ -1098,25 +1048,30 @@ function collectIncludeText(sourceText, contentCatalog, defaultComponent, defaul
         const attrlist = match[2];
         if (target.includes('{'))
             continue;
-        const parsed = parseIncludeTarget(target, defaultComponent, defaultVersion, defaultModule);
         let file;
-        if (parsed) {
-            const crossComponent = parsed.component !== defaultComponent;
-            file = contentCatalog.getFiles().find((f) => f.src.component === parsed.component &&
-                (crossComponent ? true : f.src.version === parsed.version) &&
-                f.src.module === parsed.module &&
-                (parsed.family === '' || f.src.family === parsed.family) &&
-                f.src.relative === parsed.relative);
+        if (RESOURCE_ID_RE.test(target) && contentCatalog.resolveResource) {
+            // Antora's own resource-ID resolver: version@, component:module:
+            // qualifiers, family$relative, and the familyless page-shorthand form
+            // are all its concern, not this function's. A malformed spec throws
+            // rather than returning undefined; treated the same as unresolved.
+            try {
+                file = contentCatalog.resolveResource(target, { component: defaultComponent, version: defaultVersion, module: defaultModule }, 'page');
+            }
+            catch {
+                file = undefined;
+            }
         }
-        else {
+        else if (!RESOURCE_ID_RE.test(target)) {
             // A plain filesystem-style target resolves relative to the including
             // file's own path within the same component and version.
             const candidate = resolveRelativeIncludePath(target, currentPath);
             if (!candidate)
                 continue;
-            file = contentCatalog.getFiles().find((f) => f.src.component === defaultComponent &&
-                f.src.version === defaultVersion &&
-                f.src.path === candidate);
+            file = contentCatalog.getByPath
+                ? contentCatalog.getByPath({ component: defaultComponent, version: defaultVersion, path: candidate })
+                : contentCatalog.getFiles().find((f) => f.src.component === defaultComponent &&
+                    f.src.version === defaultVersion &&
+                    f.src.path === candidate);
         }
         if (!file)
             continue;
@@ -1174,10 +1129,7 @@ function register(context) {
         }
     });
     ctx.once('playbookBuilt', async ({ playbook }) => {
-        const extEntry = (playbook?.antora?.extensions ?? []).find((ext) => {
-            const req = String(ext.require ?? '');
-            return req.includes('trademark-it') || req.includes('trademark_it');
-        }) ?? {};
+        const extEntry = findOwnExtEntry(playbook?.antora?.extensions ?? []);
         const raw = extEntry.trademark_it ??
             extEntry.trademarkIt ??
             extEntry.trademarkit ??
@@ -1310,6 +1262,24 @@ function register(context) {
                     (compScanExclude.length === 0 || !matchesAnyGlob(f.src.path ?? f.src.relative ?? '', compScanExclude)));
                 if (scanFiles.length === 0)
                     continue;
+                // Effective attribute state for conditional evaluation: the
+                // component-version's compiled asciidoc attributes, overlaid per
+                // scanned file with its own header entries — the same real-state
+                // model autoterm's scan loop uses, see agent-memory/
+                // trademark-it-project.md. compIgnoreConditionals still
+                // force-suppresses ifdef:: blocks on top of real evaluation.
+                const componentAttrs = (contentCatalog.getComponents?.() ?? [])
+                    .find((c) => c.name === component)?.versions
+                    .find((v) => (v.version ?? v.name) === version)?.asciidoc?.attributes ?? {};
+                const makeIsDefined = (fileContent, family) => {
+                    const headerAttrs = family === 'page' ? (0, asciidoc_conditionals_1.pageHeaderAttributes)(fileContent) : new Map();
+                    return (attr) => {
+                        if (headerAttrs.has(attr))
+                            return headerAttrs.get(attr) === true;
+                        const v = componentAttrs[attr];
+                        return v !== undefined && v !== null && v !== false;
+                    };
+                };
                 // Pre-populate the visited set with all files already being scanned so
                 // that include resolution does not produce duplicate text.
                 const includeVisited = new Set();
@@ -1322,28 +1292,50 @@ function register(context) {
                 // Resolve include:: directives per file so that trademark terms appearing
                 // only in cross-component (or cross-module) included files are detected.
                 const includeChunks = [];
+                // Each scanned file's own conditional state is evaluated against
+                // ITS OWN attributes (component attrs overlaid with that file's own
+                // header entries) before joining into one combined scan blob below —
+                // a single shared isDefined applied to the whole concatenation would
+                // let one file's page-header override incorrectly leak onto every
+                // other file's conditionals too. Resolved include text is stripped
+                // with the SAME including file's isDefined, mirroring autoterm's own
+                // scan loop exactly: a delegating page stub that includes an entire
+                // other page (a real, deliberately-supported pattern, see
+                // trademark-it-include-test.js) pulls in that page's raw text via
+                // collectIncludeText, and any ifdef::/ifndef:: guards inside it need
+                // evaluating too — reproduced directly against the real demo site,
+                // not assumed: its own docs component includes the whole
+                // trademark-it.adoc reference page from extensions, and leaving the
+                // included copy's ifdef::removed[Jira...] example unstripped let it
+                // leak into the real generated table.
+                const strippedFileChunks = [];
                 for (const f of scanFiles) {
                     const fileContent = f.contents.toString('utf8');
+                    const isDefined = makeIsDefined(fileContent, f.src.family);
+                    strippedFileChunks.push(stripNonContent(fileContent, isDefined, compIgnoreConditionals));
                     // Log per-file trademark detection for debugging, split into
                     // added vs ignored entries and prefixed with the component name.
                     const scanFilePath = `${component}:${f.src.path ?? f.src.relative ?? ''}`;
-                    const fileMatches = findTrademarksInFile(fileContent, compTrademarks, compIgnoreConditionals);
+                    const fileMatches = findTrademarksInFile(fileContent, compTrademarks, isDefined, compIgnoreConditionals);
                     logFileMatches(logger, scanFilePath, fileMatches, isCompIgnored);
                     const onIncludedFile = quietMode ? undefined : (chain, includedContent) => {
-                        const includeMatches = findTrademarksInFile(includedContent, compTrademarks, compIgnoreConditionals);
+                        const includeMatches = findTrademarksInFile(includedContent, compTrademarks, isDefined, compIgnoreConditionals);
                         logFileMatches(logger, [scanFilePath, ...chain].join(' → '), includeMatches, isCompIgnored);
                     };
                     const extra = collectIncludeText(fileContent, contentCatalog, component, version, f.src.module ?? 'ROOT', includeVisited, compScanExclude, onIncludedFile, [], f.src.path ?? '');
                     if (extra)
-                        includeChunks.push(extra);
+                        includeChunks.push(stripNonContent(extra, isDefined, compIgnoreConditionals));
                 }
                 // Concatenate all source text for this component-version, including any
-                // text resolved from cross-component include:: directives.
+                // text resolved from cross-component include:: directives. Each scan
+                // file's own text is already stripped above; findTrademarks' own
+                // stripNonContent pass over the combined text is therefore a no-op
+                // for that portion (no directives or comments remain to act on).
                 const combinedText = [
-                    ...scanFiles.map((f) => f.contents.toString('utf8')),
+                    ...strippedFileChunks,
                     ...includeChunks,
                 ].join('\n');
-                const found = findTrademarks(combinedText, compTrademarks, compIgnoreConditionals);
+                const found = findTrademarks(combinedText, compTrademarks);
                 // Optionally scan SVG image files for text-bearing elements.
                 if (compScanSvgs) {
                     const svgFiles = contentCatalog.getFiles().filter((f) => f.src.component === component &&
@@ -1353,7 +1345,10 @@ function register(context) {
                         (compScanExclude.length === 0 || !matchesAnyGlob(f.src.path ?? f.src.relative ?? '', compScanExclude)));
                     for (const f of svgFiles) {
                         const svgText = extractSvgText(f.contents.toString('utf8'));
-                        const svgFound = findTrademarks(svgText, compTrademarks, []);
+                        // extracted SVG text carries no AsciiDoc conditionals or comments
+                        // to evaluate; isDefined and ignoreConditionals stay at their
+                        // defaults.
+                        const svgFound = findTrademarks(svgText, compTrademarks);
                         const svgMatches = [];
                         for (const [term, members] of svgFound) {
                             const existing = found.get(term) ?? new Set();
@@ -1519,7 +1514,9 @@ function resolveTrademarkFile(trademarkFileRef, contentCatalog, playbook, logger
 }
 /**
  * Derive the set of unique component-version pairs from the catalog.
- * Returns a Map<component, Set<version>>.
+ * Returns a Map<component, Set<version>>. Exported for cite-it, a third
+ * consumer of this same enumeration (autoterm still builds its own inline,
+ * a known small duplication; see root CLAUDE.md, "Content Extraction").
  */
 function getComponentVersions(contentCatalog) {
     const result = new Map();