Accumulate discontiguous ranges in legacy Unicode blocks

Retain all Specials records, including U+FEFF, while preserving the overlaps present in the historical source files.

Assisted-By: GPT-6
Issue: #110

Author: JLHwung

scripts/parse-blocks-scripts-properties.mjs

@@ -305,10 +305,8 @@ const parseBlocks = async (version) => {
 				.split(';')
 				.map((field) => field.trim());
 			const canonicalBlockName = looseMatch('Block', blockName).value;
-			map[canonicalBlockName] = regenerate().addRange(
-				parseInt(start, 16),
-				parseInt(end, 16),
-			);
+			map[canonicalBlockName] ??= regenerate();
+			map[canonicalBlockName].addRange(parseInt(start, 16), parseInt(end, 16));
 		}
 	} else {
 		map = await parseBlocksScriptsProperties('blocks', version);

tests/parser-formats.mjs

@@ -147,6 +147,22 @@ test('Block names exclude inline comments in modern and legacy formats', async (
 	}
 });
 
+test('Repeated legacy block records accumulate without filling their gaps', async (t) => {
+	t.mock.method(
+		utils,
+		'readDataFile',
+		() => 'FEFF; FEFF; Specials\nFFF0; FFFD; Specials\n',
+	);
+	const blocks = await parseBlocks('3.0.1');
+	t.assert.deepEqual(blocks.Specials.toArray(), [
+		0xfeff,
+		...Array.from({ length: 14 }, (_, i) => 0xfff0 + i),
+	]);
+	t.assert.ok(!blocks.No_Block.contains(0xfeff));
+	t.assert.ok(!blocks.No_Block.contains(0xfffd));
+	t.assert.ok(blocks.No_Block.contains(0xff00));
+});
+
 test('Name aliases ignore inline comments without splitting alias groups', async (t) => {
 	t.mock.method(
 		utils,

tests/parsers.mjs

@@ -27,7 +27,8 @@ test('Historical titlecase digraphs retain their identity mapping', async (t) =>
 });
 
 test('Block values cover Unicode, including No_Block gaps', async (t) => {
-	for (const version of ['2.0.14', '3.1.0', '18.0.0']) {
+	// Earlier files overlap Arabic Presentation Forms-B and Specials at U+FEFF.
+	for (const version of ['2.1.9', '3.1.0', '18.0.0']) {
 		const blocks = await parseBlocks(version);
 		t.assert.ok(blocks.No_Block.contains(0x40000), version);
 		t.assert.ok(!blocks.No_Block.contains(0x41), version);
@@ -50,6 +51,31 @@ test('Block values cover Unicode, including No_Block gaps', async (t) => {
 	t.assert.equal(await parseBlocks('1.1.5'), undefined);
 });
 
+test('Legacy blocks retain every discontiguous Specials range', async (t) => {
+	for (const version of [
+		'2.0.14',
+		'2.1.2',
+		'2.1.5',
+		'2.1.8',
+		'2.1.9',
+		'3.0.0',
+		'3.0.1',
+	]) {
+		const blocks = await parseBlocks(version);
+		t.assert.ok(blocks.Specials.contains(0xfeff), version);
+		t.assert.ok(blocks.Specials.contains(0xfff0), version);
+		t.assert.ok(blocks.Specials.contains(0xfffd), version);
+		t.assert.ok(!blocks.Specials.contains(0xff00), version);
+		t.assert.ok(!blocks.No_Block.contains(0xfeff), version);
+		// Preserve the overlap explicitly recorded by the earliest source files.
+		t.assert.equal(
+			blocks.Arabic_Presentation_Forms_B.contains(0xfeff),
+			['2.0.14', '2.1.2', '2.1.5', '2.1.8'].includes(version),
+			version,
+		);
+	}
+});
+
 test('Hangul has pairwise canonical decomposition mappings', async (t) => {
 	for (const version of ['2.0.14', '3.2.0', '18.0.0']) {
 		const { Canonical } = await parseDecompositionMapping(version);