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

[jakejackson1]

Closes #189

What

A GSUB Type 5 rule has an input sequence and nothing else. When the rule belongs to an Arabic joining form (isol/fina/fin2/fin3/medi/med2/init, script arab), TTFontFile::_getGSUBarray() turns each nested substitution into an entry for Shaper\Arabic, with a prel and a postl. Format 3 gave both as []. Formats 1 and 2 passed contextRule() the method's $backtrackGlyphs and $lookaheadGlyphs instead. Type 5 never sets either, so these rules got the sequences of the last chained rule (Type 6 or Type 8) read for the script. The shaper then gave the form only where the text also held that unrelated rule's context.

The fix is the one the issue names: the 5.1 and 5.2 calls no longer pass [$backtrackGlyphs, $lookaheadGlyphs]. That argument was the only use of contextRule()'s $arabic parameter, so:

• the parameter is removed, along with the prel/postl keys contextRule() built from it
• gsubContextRule() gives the Arabic entry the rule's own backtrack/lookahead, which are [] for every Type 5 rule
• the walk's $backtrackGlyphs = []; $lookaheadGlyphs = []; initialisation and its The parser hands an Arabic form's plain context rule the backtrack and lookahead of the last chained rule it read #189 comment are removed. Every branch that reads the two variables now sets them first (Type 8, Type 6 Formats 1–3).

With the Type 5 branches no longer reading them, Type 6 Format 1's own sequences go straight into its contextRule() call (a /simplify cleanup), so $backtrackGlyphs/$lookaheadGlyphs are only set from coverageSequences()/classSequences() right before they're used.

/simplify also suggested one helper for the Arabic joining-form entry that gsubContextRule() and gsubReverseChainRule() both build. I skipped it: the duplication was there before this PR.

Test font and shaping test

NotoSansArabic-GSUB5Form-Synthetic.ttf is NotoSansArabic-Joining-Subset.ttf (Noto Sans Arabic 2.012, OFL 1.1), changed in fontTools 4.59.2. Its GSUB is replaced by three lookups for DFLT and arab, in this order, and name IDs 1, 4 and 6 are renamed:
0. fina Type 6 Format 3: beh → uni066E.fina after beh and before U+08AD LOW ALEF (which doesn't join)

1. init Type 5 Format 1: beh → uni066E.init
2. medi Type 5 Format 2: beh → uni066E.medi

Before the fix, both Type 5 rules inherited lookup 0's prel [beh] and postl [lowalef].

ContextRuleArabicFormTest records the drawn glyphs with TextRecordingMpdf. The output is in visual order, and PUA E002/E003/E004 = uni066E.fina/.medi/.init:

Run (logical)	Before	After	hb-shape 14.3.1
beh beh	0628 0628	0628 E004	uni0628 uni066E.init ✓
beh beh lowalef	08AD E002 0628	08AD E002 E004	uni08AD uni066E.fina uni066E.init ✓
beh beh beh lowalef	08AD E002 0628 0628	08AD E002 E003 E004	uni08AD uni066E.fina uni066E.medi uni066E.init ✓

HarfBuzz agrees with mPDF after the fix on all three runs. The third run covers the 5.2 call separately from the 5.1 call: the medial beh is followed by a beh, not a low alef.

GsubArrayTest::testAPlainContextRuleOfAnArabicFormCarriesTheLastChainedRulesSequences is renamed to testAPlainContextRuleOfAnArabicFormHasNoBacktrackOrLookahead and expects 'prel' => [].

Fixtures

Regenerated from cold: I deleted tests/Mpdf/tmp/mpdf, tmp/mpdf and tmp/ttfontdata, then ran fontcache:update, otldump:update, shaping:update and subset:update all. git status -- tests/data showed only the new font's five files. No existing fixture moved, which matches the issue's measurement.

In the new font's fixtures:

• fontcache GSUB.arab.DFLT.json: rtlSUB → 00628 has forms 1/2/3 = 0E002/0E004/0E003, with prel/postl only on form 1, the chained fina rule.
• shaping: the Arabic run's leading beh (before a tatweel) is drawn as E004, the initial form. It would be 0628 before the fix.

Mutation check

With the argument put back (the pre-fix src/TTFontFile.php) and caches cleared, 6 tests fail:

• all three data sets of ContextRuleArabicFormTest::testTheFormNeedsNoContext
• GsubArrayTest::testAPlainContextRuleOfAnArabicFormHasNoBacktrackOrLookahead
• ParserGoldenMasterTest for NotoSansArabic-GSUB5Form-Synthetic
• ShapingGoldenMasterTest for NotoSansArabic-GSUB5Form-Synthetic

Mirroring upstream

Upstream has no contextRule(). In mpdf/mpdf development (744f75a), TTFontFile::_getGSUBarray()'s LookupType 5 branches build the Arabic entry inline:

• Format 1 (around line 2513)
• Format 2 (around line 2581)

$volt[] = ['match' => $lookupGlyphs[0], 'replace' => $REPL, 'tag' => $tag, 'prel' => $backtrackGlyphs, 'postl' => $lookaheadGlyphs, 'ignore' => $ignore];

In both, the mirror changes 'prel' => $backtrackGlyphs, 'postl' => $lookaheadGlyphs to 'prel' => [], 'postl' => []. Upstream never initialises the two variables, so a 5.1/5.2 Arabic rule read before any chained rule also raises "Undefined variable" there, and this change removes that warning too. Upstream's OtlDump only reports these rules and has no prel/postl, so it doesn't change. The upstream PR also needs a CHANGELOG.md entry. The test font and ContextRuleArabicFormTest can be taken over as they are.

Base

Stacked on #195 (fix/170-class-rule-positions) → #190 → #188 → #186 → gravitypdf. #177 and #178 were squash-merged into gravitypdf, and the stack below was rebased onto it. I rebased this branch with git rebase --onto fix/170-class-rule-positions <old #195 head 6680d8b> and force-pushed it with a lease. There were no conflicts. #186, #188, #190 and #195 are still open, so the base stays fix/170-class-rule-positions.

Verification

• vendor/bin/phpunit (after the rebase): 2009 tests, 5801 assertions, 1 skipped (Match each class-based chained rule against its own backtrack and lookahead positions (#170) #195's branch: 1999 / 5781 / 1).
• Golden masters regenerated from cold again after the rebase: git status -- tests/data is clean.
• vendor/bin/phpunit --group=snapshot: OK, 79 tests, 110 assertions.
• composer cs: clean.
• phpstan: 33 errors, the same as the baseline.

🤖 Generated with Claude Code

[jakejackson1]

Rebased: #190 was squash-merged into gravitypdf (f9b3c0e), so this stack still carried its pre-squash commits. This branch is rebased onto its rebased parent with --onto; no conflicts and the base is unchanged. Suite, snapshots and cs pass on the new head, with no fixture movement.

[jakejackson1]

This PR merged into a base branch that had already been squash-merged, so its change never reached gravitypdf. It's carried to gravitypdf by #215.