@lokascript/framework 2.7.2 → 2.9.0

package/dist/core/index.js

@@ -125,6 +125,9 @@ function isQuote(char) {
 function isDigit(char) {
   return /\d/.test(char);
 }
+function stripOptionalDiacritics(word) {
+  return word.replace(/[ً-ْٰ]/g, "");
+}
 function isAsciiLetter(char) {
   return /[a-zA-Z]/.test(char);
 }
@@ -915,7 +918,39 @@ var _BaseTokenizer = class _BaseTokenizer {
         pos++;
       }
     }
-    return new TokenStreamImpl(tokens, this.language);
+    return new TokenStreamImpl(this.mergeColonQualifiedNames(tokens), this.language);
+  }
+  /**
+   * Fuse `name` + `:qualifier` into ONE identifier (`draggable:start`).
+   *
+   * `:name` is hyperscript's local-variable sigil, but a colon IMMEDIATELY
+   * preceded by an identifier is a qualifier (custom event namespace), not a
+   * sigil. The English tokenizer already merges these inside
+   * EnglishKeywordExtractor; this post-pass gives the other 23 languages the
+   * same stream. Strict position adjacency is the discriminator: whitespace
+   * between the tokens (`trigger :start`) breaks `end === start`, so a spaced
+   * local-variable reference survives untouched.
+   *
+   * Self-gating for non-hyperscript tokenizers (domain DSLs): their extractor
+   * sets tokenize `:` as bare punctuation (length 1), which never matches
+   * COLON_QUALIFIER, so this pass is a no-op for them.
+   */
+  mergeColonQualifiedNames(tokens) {
+    const out = [];
+    for (const tok of tokens) {
+      const prev = out[out.length - 1];
+      if (prev && _BaseTokenizer.ASCII_WORD.test(prev.value) && _BaseTokenizer.COLON_QUALIFIER.test(tok.value) && prev.position.end === tok.position.start) {
+        const merged = prev.value + tok.value;
+        out[out.length - 1] = createToken(
+          merged,
+          this.classifyToken(merged),
+          createPosition(prev.position.start, tok.position.end)
+        );
+        continue;
+      }
+      out.push(tok);
+    }
+    return out;
   }
   /**
    * Classify an unknown character when no extractor matches.
@@ -1048,7 +1083,7 @@ var _BaseTokenizer = class _BaseTokenizer {
    * @returns Word without diacritics
    */
   removeDiacritics(word) {
-    return word.replace(/[\u064B-\u0652\u0670]/g, "");
+    return stripOptionalDiacritics(word);
   }
   /**
    * Try to match a keyword from profile at the current position.
@@ -1139,24 +1174,40 @@ var _BaseTokenizer = class _BaseTokenizer {
     });
   }
   /**
-   * Look up a keyword by native word (case-insensitive).
+   * Look up a keyword by native word (case-insensitive, diacritic-insensitive).
    * O(1) lookup using the keyword map.
    *
+   * The map is INDEXED both with and without diacritics (see
+   * `initializeKeywordsFromProfile`), so a stripped QUERY is the other half of
+   * that: it lets a surface form carrying harakat the profile does not happen to
+   * spell still find its entry. Only consulted after the exact lookup misses, so
+   * every previously-matching word resolves byte-identically.
+   *
+   * Half-implementing this — indexing stripped but querying exact — is what made
+   * diacritized `بَدِّل` (toggle) tokenize as `kind=particle normalized=with`:
+   * `isKeyword` returned false, so the guard in `ArabicProcliticExtractor` that
+   * exists to prevent exactly that handed the word on, and the single-char `ب`
+   * bi- proclitic claimed it. A wrong CONCEPT, not a failed parse.
+   *
    * @param native - Native word to look up
    * @returns KeywordEntry if found, undefined otherwise
    */
   lookupKeyword(native) {
-    return this.profileKeywordMap.get(native.toLowerCase());
+    const exact = this.profileKeywordMap.get(native.toLowerCase());
+    if (exact) return exact;
+    const stripped = this.removeDiacritics(native);
+    if (stripped === native) return void 0;
+    return this.profileKeywordMap.get(stripped.toLowerCase());
   }
   /**
-   * Check if a word is a known keyword (case-insensitive).
-   * O(1) lookup using the keyword map.
+   * Check if a word is a known keyword (case-insensitive, diacritic-insensitive).
+   * O(1) lookup using the keyword map. See {@link lookupKeyword}.
    *
    * @param native - Native word to check
    * @returns true if the word is a keyword
    */
   isKeyword(native) {
-    return this.profileKeywordMap.has(native.toLowerCase());
+    return this.lookupKeyword(native) !== void 0;
   }
   /**
    * Set the morphological normalizer for this tokenizer.
@@ -1421,6 +1472,14 @@ var _BaseTokenizer = class _BaseTokenizer {
     return null;
   }
 };
+/**
+ * ASCII word of the shape the English word-walker produces. Excludes `:`, so a
+ * token that already carries a qualifier never merges again — `a:b:c` yields
+ * `a:b` + `:c`, byte-matching the English extractor's single-segment merge.
+ */
+_BaseTokenizer.ASCII_WORD = /^[A-Za-z_][A-Za-z0-9_]*$/;
+/** `:name` — only a variable-ref-style extractor ever emits this token shape. */
+_BaseTokenizer.COLON_QUALIFIER = /^:[A-Za-z_][A-Za-z0-9_]*$/;
 /**
  * Configuration for native language time units.
  * Maps patterns to their standard suffix (ms, s, m, h).
@@ -2944,6 +3003,7 @@ export {
   noChange,
   normalized,
   patternMatcher,
+  stripOptionalDiacritics,
   validateValueType,
   withDefaultExtractors
 };