Fraction typesetter — style override, cfrac struts, thinspace, numerator alignment

[kostub]

Summary

Wires the styleOverride, isContinuedFraction, and numeratorAlignment fields (added in PR 1) into the typesetter. This is PR 2 of 4 in the AMSMath fractions + multi-integral stack.

Plan: docs/plans/2026-05-25-amsmath-fractions-and-multi-integrals.md
LLD: docs/lld/2026-05-25-amsmath-fractions-and-multi-integrals.md

Goal

Render \dfrac/\tfrac/\dbinom/\tbinom at their forced styles; render \cfrac with AMSMath strut equivalents (0.85em/0.35em) and 3mu thinspace padding; honor \cfrac[l|c|r] numerator alignment. Existing \frac / \binom rendering remains pixel-identical.

Commits

• [item 10] Add -lineStyleForFractionStyle: helper to MTTypesetter
• [item 11] Add numeratorAlignment to MTFractionDisplay positioning
• [item 12] Honor MTFraction.styleOverride and numeratorAlignment in typesetter
• [item 13] Apply \cfrac strut floors (0.85em / 0.35em) to operand displays
• [item 14] Wrap \cfrac with 3mu thinspace padding via MTMathListDisplay
• [item 15] Typesetter test for \cfrac[l]/[r] numerator alignment

Stack

• PR 1 (base): Fraction data model + parser #202 — Fraction data model + parser

Test plan

• 217 tests pass (verified by item 15 full-suite run)
• testDfracInlineStylePicksDisplayMetrics — \dfrac in text mode produces display-style bar/gap metrics
• testCfracStrutAppliedToOperands — cfrac operand ascent/descent floored at 0.85em/0.35em; plain frac is not
• testCfracThinspaceWrap — cfrac top display is MTMathListDisplay wrapper with width = inner + 2×thinspace
• testCfracLeftAlignmentNumeratorOffset — \cfrac[l]/\cfrac[r]/\cfrac numerator x-offsets correct
• All existing typesetter tests pass (67 tests before new additions)

[kostub]

Code Review — PR #203: Fraction typesetter (style override, cfrac struts, thinspace, numerator alignment)

Overall this is a clean, well-scoped implementation. The three main additions (style override, cfrac struts, thinspace wrapper) are logically separated and easy to follow. A few items worth addressing before merge:

---

Bug: initWithDisplays:range: runs recomputeDimensions before position is set, then the code silently overrides the result

In MTTypesetter.m, the thinspace-wrapping block:

result.position = CGPointMake(thinspace, 0);   // ← set BEFORE init
MTMathListDisplay* wrapped = [[MTMathListDisplay alloc] initWithDisplays:@[result]
                                                                   range:frac.indexRange];
// recomputeDimensions runs inside initWithDisplays and computes width = result.width + thinspace
// (only one side of padding, not two). Then:
wrapped.width = result.width + 2.0 * thinspace;  // ← manual override

initWithDisplays:range: calls recomputeDimensions, which computes width = atom.width + atom.position.x = result.width + thinspace. The code then immediately overrides wrapped.width to result.width + 2*thinspace, so the final value is correct — but it relies on a fragile ordering where recomputeDimensions's width output is always wrong and always discarded.

A cleaner approach: set result.position after initWithDisplays:, so the wrapper's recomputeDimensions starts from {0,0} (natural position), and then set the position and override width in a single explicit pass. Or, simpler: just leave result.position = CGPointZero until after init:

MTMathListDisplay* wrapped = [[MTMathListDisplay alloc] initWithDisplays:@[result]
                                                                   range:frac.indexRange];
result.position = CGPointMake(thinspace, 0);   // set after init
wrapped.position = _currentPosition;
wrapped.ascent  = result.ascent;
wrapped.descent = result.descent;
wrapped.width   = result.width + 2.0 * thinspace;

---

Nit: didOverrideStyle guard slightly over-narrow — style can be the same value but still benefit from explicit save/restore

The condition if (overrideStyle != _style) skips setting didOverrideStyle = YES when the override happens to equal the current style. This is correct for avoiding a no-op setStyle: call, but it also suppresses the setStyle:savedStyle restore, which doesn't matter since nothing changed. The logic is sound; this is just worth a brief comment so future readers don't wonder whether the != _style guard is load-bearing:

// Only mutate _style (and _styleFont) when the override actually differs.
// When they're equal, the frame metrics are already correct and no restore is needed.
if (overrideStyle != _style) {

---

Test issue: testCfracThinspaceWrap computes muUnit from the wrong font size

// "Get the muUnit for the *display* style (cfrac forces Display)."
MTFont* styleFont = [font copyFontWithSize:font.fontSize];   // ← same as font
CGFloat thinspace = 3.0 * styleFont.mathTable.muUnit;

The comment says "display style" but [font copyFontWithSize:font.fontSize] is identical to font — no style scaling is applied. This works accidentally because kMTLineStyleDisplay uses font.fontSize unchanged (see +getStyleSize:font:), so the muUnit is the same. The code is correct for the current default font size, but the comment is misleading. Consider removing the intermediate styleFont variable and just writing:

CGFloat thinspace = 3.0 * font.mathTable.muUnit;

---

Minor: \cfrac + delimiter combination not defensively guarded

The isContinuedFraction thinspace-wrapping block runs regardless of whether addDelimitersToFractionDisplay: was already called (i.e., \cfrac with explicit leftDelimiter/rightDelimiter). Since the parser never produces that combination, this is not a current bug, but the code doesn't assert it. A brief NSAssert(!frac.leftDelimiter && !frac.rightDelimiter, @"cfrac should not have delimiters") inside the isContinuedFraction branch would make the assumption explicit.

---

Style observation: strut is applied before gap-adjustment, which is intentional but worth a comment

The strut floors on numeratorDisplay.ascent/descent and denominatorDisplay.ascent/descent are applied at lines 1260–1263, before numeratorShiftUp / denominatorShiftDown are computed and gap-adjusted. This means the strut inflates the operands and the gap logic then re-expands the shift values to maintain the required clearance from the bar. This matches AMSMath's behavior (the strut is a floor on the operand box, not on the shift), but it's non-obvious. A one-line comment would help:

// Apply cfrac strut floors to operand boxes *before* gap adjustment so that
// the bar-clearance logic in the next block sees the inflated extents.

---

All four new tests are well-structured and cover the key invariants. The testDfracInlineStylePicksDisplayMetrics cross-comparison approach (render at the forced style, compare metrics) is particularly clean.

[kostub]

Test comment to verify posting works

[gemini-code-assist]

Code Review

This pull request implements support for fraction alignment and continued fractions (\cfrac). Key changes include adding a numeratorAlignment property to MTMathListDisplay, updating the typesetter to handle style overrides, and implementing struts and thinspace wrapping for continued fractions. A comprehensive suite of unit tests was added to verify these new features. Review feedback identifies that the \cfrac implementation should force DisplayStyle and avoid cramping the denominator to correctly match AMSMath behavior.

[gemini-code-assist]

AMSMath's \cfrac implementation requires that both the numerator and denominator are typeset in \displaystyle (regardless of the surrounding style) and that the denominator is not cramped. The current implementation uses the standard fraction style increment (which results in TextStyle for a DisplayStyle fraction) and always cramps the denominator.

To match AMSMath behavior, fractionStyle should be forced to kMTLineStyleDisplay and the cramped parameter for the denominator should be false when isContinuedFraction is true.

    MTLineStyle fractionStyle = frac.isContinuedFraction ? kMTLineStyleDisplay : self.fractionStyle;
    MTMathListDisplay* numeratorDisplay = [MTTypesetter createLineForMathList:frac.numerator font:_font style:fractionStyle cramped:false];
    MTMathListDisplay* denominatorDisplay = [MTTypesetter createLineForMathList:frac.denominator font:_font style:fractionStyle cramped:!frac.isContinuedFraction];

[kostub]

Fixed in 37ef1f3 — \cfrac numerator/denominator now render in kMTLineStyleDisplay (not fractionStyle) and the denominator is passed cramped:false.