From 0750dd58958fbaef1854df1ab4e9d020f0b3e4c4 Mon Sep 17 00:00:00 2001 From: Cosimo Lupo Date: Wed, 22 Jun 2022 18:31:51 +0100 Subject: [PATCH] Add reproducer for issue #621 data copied from test font at changa.zip in #621 Thanks Marc Foley! --- .../decomposeTransformedComponents_test.py | 91 +++++++++++++++++++ 1 file changed, 91 insertions(+) diff --git a/tests/filters/decomposeTransformedComponents_test.py b/tests/filters/decomposeTransformedComponents_test.py index 0cbf2604c..7c528016b 100644 --- a/tests/filters/decomposeTransformedComponents_test.py +++ b/tests/filters/decomposeTransformedComponents_test.py @@ -93,3 +93,94 @@ def test_decompose_compatibly(self, FontClass): ).process() assert len(glyphsets[0]["base"]) == 1 assert len(glyphsets[1]["base"]) == 1 + + def test_decompose_compatibly_nested_transformed_components(self, FontClass): + # This replicates three glyphs from the 'changa.zip' test fonts at + # https://github.com/googlefonts/ufo2ft/issues/621 + # In both fonts, the "exclam" glyph is made of one simple contour and one + # component ("period"); "exclamdown" in turn is made of one "exclam" component + # that is flipped vertically and horizontally; "period" is a single contour. + # But only in the Bold.ufo, the "exclam" contains a scaled down "period"; in + # the Regular.ufo, the "period" component only has an offset. + # This would previously trigger a situation whereby after "exclamdown" was + # decomposed, its points were no longer interpolation compatible across masters + # because the order in which the contours were decomposed was different. + # This is because filters used to modify glyphs in-place in alphabetical order, + # so 'exclam' comes before 'exclamdown', and in the Bold.ufo, 'exclam' has + # a 2x3 transform so is decomposed (with the period appended at the end), but + # then 'exclamdown' needs decomposing as well (for it's flipped) and the + # already decomposed 'exclam' contours are drawn onto it in the same order; + # whereas in Regular.ufo, the 'exclam' does not contain transformed components + # so it's kept as composite (for the time being, it will be decomposed later on + # because it's mixed), but when it's the turn of 'exclamdown', the period's + # contour gets appended to it before the rest of the rest of the 'exclam' + # (deepCopyContours follows a post-order depth-first traversal so the children + # get decomposed before the parent) -- leading to cu2qu crashing... Pfew! + regular_ufo = FontClass() + period = regular_ufo.newGlyph("period") + period.width = 230 + pen = period.getPen() + pen.moveTo((50, 62)) + pen.curveTo((50, 13), (61, -6), (115, -6)) + pen.curveTo((168, -6), (180, 13), (180, 62)) + pen.curveTo((180, 110), (168, 131), (115, 131)) + pen.curveTo((61, 131), (50, 110), (50, 62)) + pen.closePath() + + exclam = regular_ufo.newGlyph("exclam") + exclam.width = 250 + pen = exclam.getPen() + pen.moveTo((93, 196)) + pen.lineTo((156, 196)) + pen.lineTo((186, 627)) + pen.curveTo((186, 637), (181, 645), (161, 645)) + pen.lineTo((87, 645)) + pen.curveTo((67, 645), (63, 637), (63, 627)) + pen.closePath() + pen.addComponent("period", (1, 0, 0, 1, 10, 0)) + + exclamdown = regular_ufo.newGlyph("exclamdown") + exclamdown.width = 250 + pen = exclamdown.getPen() + pen.addComponent("exclam", (-1, 0, 0, -1, 250, 509)) + + bold_ufo = FontClass() + period = bold_ufo.newGlyph("period") + period.width = 277 + pen = period.getPen() + pen.moveTo((30, 99)) + pen.curveTo((30, 23), (50, -6), (139, -6)) + pen.curveTo((227, -6), (247, 23), (247, 99)) + pen.curveTo((247, 175), (227, 206), (139, 206)) + pen.curveTo((50, 206), (30, 175), (30, 99)) + pen.closePath() + + exclam = bold_ufo.newGlyph("exclam") + exclam.width = 297 + pen = exclam.getPen() + pen.moveTo((84, 230)) + pen.lineTo((214, 230)) + pen.lineTo((254, 618)) + pen.curveTo((254, 633), (247, 645), (217, 645)) + pen.lineTo((81, 645)) + pen.curveTo((51, 645), (44, 633), (44, 618)) + pen.closePath() + pen.addComponent("period", (0.87, 0, 0, 0.87, 28, -1)) + + exclamdown = bold_ufo.newGlyph("exclamdown") + exclamdown.width = 297 + pen = exclamdown.getPen() + pen.addComponent("exclam", (-1, 0, 0, -1, 298, 509)) + + # We test that, even with DecomposeTransformedComponentsFilter(pre=True) and + # the above nested/transformed/mixed component setup, we don't crash cu2qu + # with errors about masters with inconsistent contour order after decomposition + # of "exclamdown". + glyphsets = TTFInterpolatablePreProcessor( + [regular_ufo, bold_ufo], + filters=[DecomposeTransformedComponentsFilter(pre=True)], + ).process() + assert len(glyphsets[0]["exclam"]) == 2 + assert len(glyphsets[0]["exclamdown"]) == 2 + assert len(glyphsets[1]["exclam"]) == 2 + assert len(glyphsets[1]["exclamdown"]) == 2