Give an Arabic joining form's plain context rule no backtrack or lookahead (#189)

src/TTFontFile.php

@@ -2412,11 +2412,6 @@ function _getGSUBarray(array $Lookup, $lul, $scripttag)
 	{
 		$volt = [];
 
-		// A Type 5 Format 1 or 2 rule reads these without setting them, so hands whatever the last
-		// chained rule left in them to the Arabic shaper (#189).
-		$backtrackGlyphs = [];
-		$lookaheadGlyphs = [];
-
 		foreach ($lul as $i => $tag) {
 			$this->reportGSUBlookupStart($Lookup, $i, $tag);
 
@@ -2471,9 +2466,7 @@ function _getGSUBarray(array $Lookup, $lul, $scripttag)
 								[],
 								$inputGlyphs,
 								[],
-								count($inputGlyphs),
-								['', '', ''],
-								[$backtrackGlyphs, $lookaheadGlyphs]
+								count($inputGlyphs)
 							)));
 						}
 					}
@@ -2494,8 +2487,7 @@ function _getGSUBarray(array $Lookup, $lul, $scripttag)
 								$this->classInputGlyphs($subtable['InputClasses'], $inputClass, $rule),
 								[],
 								$rule['InputGlyphCount'],
-								$class0excl,
-								[$backtrackGlyphs, $lookaheadGlyphs]
+								$class0excl
 							)));
 						}
 					}
@@ -2517,15 +2509,12 @@ function _getGSUBarray(array $Lookup, $lul, $scripttag)
 							$inputGlyphs[0] = $firstInputGlyph;
 							ksort($inputGlyphs);
 
-							$backtrackGlyphs = $rule['BacktrackGlyphCount'] ? $rule['BacktrackGlyphs'] : [];
-							$lookaheadGlyphs = $rule['LookaheadGlyphCount'] ? $rule['LookaheadGlyphs'] : [];
-
 							$this->addTo($volt, $this->gsubContextRule($Lookup, $i, $c, $tag, $scripttag, $ignore, $this->contextRule(
 								$rctr,
 								$rule['SubstLookupRecord'],
-								$backtrackGlyphs,
+								$rule['BacktrackGlyphCount'] ? $rule['BacktrackGlyphs'] : [],
 								$inputGlyphs,
-								$lookaheadGlyphs,
+								$rule['LookaheadGlyphCount'] ? $rule['LookaheadGlyphs'] : [],
 								count($inputGlyphs)
 							)));
 						}
@@ -2590,17 +2579,11 @@ function _getGSUBarray(array $Lookup, $lul, $scripttag)
 	 * @param string[] $class0excl For a class-based rule, every glyph some class of the input,
 	 *                             backtrack and lookahead Class Definitions names, which is what each
 	 *                             one's class 0 excludes
-	 * @param array    $arabic     [backtrack, lookahead] for the entry an Arabic joining form's rule
-	 *                             becomes, where that is not the rule's own: see _getGSUBarray()
 	 *
 	 * @return array
 	 */
-	private function contextRule($index, array $records, array $backtrack, array $input, array $lookahead, $nInput, array $class0excl = ['', '', ''], $arabic = null)
+	private function contextRule($index, array $records, array $backtrack, array $input, array $lookahead, $nInput, array $class0excl = ['', '', ''])
 	{
-		if ($arabic === null) {
-			$arabic = [$backtrack, $lookahead];
-		}
-
 		return [
 			'index' => $index,
 			'records' => $records,
@@ -2609,8 +2592,6 @@ private function contextRule($index, array $records, array $backtrack, array $in
 			'lookahead' => $lookahead,
 			'nInput' => $nInput,
 			'class0excl' => $class0excl,
-			'prel' => $arabic[0],
-			'postl' => $arabic[1],
 		];
 	}
 
@@ -2767,7 +2748,7 @@ protected function gsubContextRule(array $Lookup, $i, $c, $tag, $scripttag, $ign
 					}
 
 					if (strpos("isol fina fin2 fin3 medi med2 init ", $tag) !== false && $scripttag == 'arab') {
-						$volt[] = ['match' => $lookupGlyphs[0], 'replace' => implode(" ", $luss['substitute']), 'tag' => $tag, 'prel' => $rule['prel'], 'postl' => $rule['postl'], 'ignore' => $ignore];
+						$volt[] = ['match' => $lookupGlyphs[0], 'replace' => implode(" ", $luss['substitute']), 'tag' => $tag, 'prel' => $rule['backtrack'], 'postl' => $rule['lookahead'], 'ignore' => $ignore];
 					} else {
 						$subRule['rules'][] = ['type' => $Lookup[$lup]['Type'], 'match' => $lookupGlyphs, 'replace' => $luss['substitute'], 'seqIndex' => $seqIndex, 'key' => $lookupGlyphs[0],];
 					}

tests/Mpdf/ContextRuleArabicFormTest.php

@@ -0,0 +1,81 @@
+<?php
+
+namespace Mpdf;
+
+/**
+ * A plain context rule (GSUB Type 5) of an Arabic joining form has no backtrack or lookahead, so the
+ * shaper gives the form wherever the joining calls for it, whatever chained rule came before (#189).
+ *
+ * No font in the corpus has such a rule, so NotoSansArabic-GSUB5Form-Synthetic is
+ * NotoSansArabic-Joining-Subset (Noto Sans Arabic 2.012, OFL 1.1) with its GSUB replaced by three
+ * lookups for beh, in this order: a 'fina' Type 6 Format 3 giving the dotless final form after beh and
+ * before low alef, an 'init' Type 5 Format 1 giving the dotless initial form, and a 'medi' Type 5
+ * Format 2 giving the dotless medial form. `hb-shape` draws the glyphs these tests expect.
+ */
+class ContextRuleArabicFormTest extends \Yoast\PHPUnitPolyfills\TestCases\TestCase
+{
+
+	const BEH = 0x0628;
+
+	const LOW_ALEF = 0x08AD;
+
+	/** The dotless forms have no codepoint, so they are mapped into the Private Use Area */
+	const DOTLESS_BEH_FINAL = 0xE002;
+
+	const DOTLESS_BEH_MEDIAL = 0xE003;
+
+	const DOTLESS_BEH_INITIAL = 0xE004;
+
+	/**
+	 * @param int[] $codepoints
+	 *
+	 * @return int[] the codepoints of the line as it is handed to the drawing code, in visual order
+	 */
+	private function drawn($codepoints)
+	{
+		$html = '';
+		foreach ($codepoints as $codepoint) {
+			$html .= sprintf('&#x%04X;', $codepoint);
+		}
+
+		$mpdf = new TextRecordingMpdf([
+			'mode' => 'utf-8',
+			'fontDir' => [__DIR__ . '/../data/ttf'],
+			'fontdata' => ['notosansarabicgsub5formsynthetic' => [
+				'R' => 'NotoSansArabic-GSUB5Form-Synthetic.ttf',
+				'useOTL' => 0xFF,
+			]],
+			'default_font' => 'notosansarabicgsub5formsynthetic',
+		]);
+		$mpdf->WriteHTML('<p>' . $html . '</p>');
+
+		return array_values(unpack('N*', mb_convert_encoding($mpdf->drawnText[0], 'UTF-32BE', 'UTF-8')));
+	}
+
+	public function dataRuns()
+	{
+		return [
+			'an initial form from Format 1, with nothing before it' => [
+				[self::BEH, self::BEH],
+				[self::BEH, self::DOTLESS_BEH_INITIAL],
+			],
+			'an initial form from Format 1, before the chained rule\'s final form' => [
+				[self::BEH, self::BEH, self::LOW_ALEF],
+				[self::LOW_ALEF, self::DOTLESS_BEH_FINAL, self::DOTLESS_BEH_INITIAL],
+			],
+			'a medial form from Format 2, with no low alef after it' => [
+				[self::BEH, self::BEH, self::BEH, self::LOW_ALEF],
+				[self::LOW_ALEF, self::DOTLESS_BEH_FINAL, self::DOTLESS_BEH_MEDIAL, self::DOTLESS_BEH_INITIAL],
+			],
+		];
+	}
+
+	/**
+	 * @dataProvider dataRuns
+	 */
+	public function testTheFormNeedsNoContext($codepoints, $expected)
+	{
+		$this->assertSame($expected, $this->drawn($codepoints));
+	}
+
+}

tests/Mpdf/Fonts/GsubArrayTest.php

@@ -54,10 +54,10 @@ public function testTheDumpReportsEachClassRuleWithItsOwnPositions()
 	}
 
 	/**
-	 * A Type 5 rule has no backtrack or lookahead, yet where it belongs to an Arabic joining form the
-	 * parser hands the shaper the sequences of the chained rule it read last (#189).
+	 * A Type 5 rule has no backtrack or lookahead, so where it belongs to an Arabic joining form the
+	 * shaper gets none, not those of the chained rule read before it (#189).
 	 */
-	public function testAPlainContextRuleOfAnArabicFormCarriesTheLastChainedRulesSequences()
+	public function testAPlainContextRuleOfAnArabicFormHasNoBacktrackOrLookahead()
 	{
 		$parser = $this->withGdef(new TTFontFile($this->cache(), 'win'));
 
@@ -78,7 +78,7 @@ public function testAPlainContextRuleOfAnArabicFormCarriesTheLastChainedRulesSeq
 
 		$volt = $parser->_getGSUBarray($Lookup, [0 => 'init', 2 => 'init'], 'arab');
 
-		$this->assertSame(['match' => '00041', 'replace' => '00044', 'tag' => 'init', 'prel' => ['00043'], 'postl' => [], 'ignore' => '()'], end($volt));
+		$this->assertSame(['match' => '00041', 'replace' => '00044', 'tag' => 'init', 'prel' => [], 'postl' => [], 'ignore' => '()'], end($volt));
 	}
 
 	private function chainedClassRules()