[PATCH v4 0/2] MR10603: win32u/freetype: Fix tmExternalLeading for CFF fonts.
This fixes font scaling issues with CFF fonts this was discovered from games utilizing DxLib. https://github.com/yumetodo/DxLib/tree/master https://dxlib.xsrv.jp/index.html I've ran tests on Windows and to confirm that this change is consistent with a number of fonts. Before: {width="801" height="600"} After (this matches the windows visual from testing in a VM): {width="767" height="572"} Related Proton Issue: https://github.com/ValveSoftware/Proton/issues/7299 Full Disclosure: An LLM was used to assist in finding this bug, the original fix written by the LLM wrote a loader binary that changed the pointers for `GetTextMetricsA` and `GetTextMetricsW` to point to a custom implementation that always set `tmExternalLeading`. The LLM determined that `GDI returns 0 when (winAscent + winDescent) > (Ascender - Descender)`, after testing on WIndows I found this to be incorrect, the actual difference is that CFF fonts always set `tmExternalLeading` to 0. The test case for this fix was LLM generated but reviewed and tested on both Windows and Linux by me. Test case: ``` WINEPREFIX=/tmp/wine-font winetricks allfonts WINEPREFIX=/tmp/wine-font ./loader/wine ./dlls/gdi32/tests/i386-windows/gdi32_test.exe ``` Patched: ``` font.c:4266: test_CFF_external_leading: tested 51 CFF fonts (0 failures) ``` Unpatched Wine: ``` font.c:4266: test_CFF_external_leading: tested 51 CFF fonts (51 failures) ``` Windows 11: ``` font.c:4266: test_CFF_external_leading: tested 48 CFF fonts (0 failures) ``` -- v4: gdi32: Add test for tmExternalLeading for CFF fonts. win32u/freetype: Fix tmExternalLeading for CFF fonts. https://gitlab.winehq.org/wine/wine/-/merge_requests/10603
From: Arie Miller <renari@arimil.com> --- dlls/win32u/freetype.c | 16 ++++++++++------ 1 file changed, 10 insertions(+), 6 deletions(-) diff --git a/dlls/win32u/freetype.c b/dlls/win32u/freetype.c index a660b166a0f..42f62dc191a 100644 --- a/dlls/win32u/freetype.c +++ b/dlls/win32u/freetype.c @@ -3392,12 +3392,16 @@ static BOOL freetype_set_outline_text_metrics( struct gdi_font *font ) TM.tmHeight = TM.tmAscent + TM.tmDescent; - /* MSDN says: - el = MAX(0, LineGap - ((WinAscent + WinDescent) - (Ascender - Descender))) - */ - TM.tmExternalLeading = max(0, SCALE_Y(pHori->Line_Gap - - ((ascent + descent) - - (pHori->Ascender - pHori->Descender)))); + /* MSDN documents the formula: + * el = MAX(0, LineGap - ((WinAscent + WinDescent) - (Ascender - Descender))) + * but Windows GDI does not follow this for OpenType-CFF fonts: for any font + * containing a 'CFF ' table it reports tmExternalLeading = 0. + */ + if (font->ntmFlags & NTM_PS_OPENTYPE) + TM.tmExternalLeading = 0; + else + TM.tmExternalLeading = max(0, + SCALE_Y(pHori->Line_Gap - ((ascent + descent) - (pHori->Ascender - pHori->Descender)))); TM.tmAveCharWidth = SCALE_X(pOS2->xAvgCharWidth); if (TM.tmAveCharWidth == 0) { -- GitLab https://gitlab.winehq.org/wine/wine/-/merge_requests/10603
From: Arie Miller <renari@arimil.com> --- dlls/gdi32/tests/font.c | 136 ++++++++++++++++++++++++++++++++++++++++ 1 file changed, 136 insertions(+) diff --git a/dlls/gdi32/tests/font.c b/dlls/gdi32/tests/font.c index c56180b1ce6..c6dc066c2d6 100644 --- a/dlls/gdi32/tests/font.c +++ b/dlls/gdi32/tests/font.c @@ -4132,6 +4132,141 @@ static void test_GetTextMetrics(void) ReleaseDC(0, hdc); } +struct cff_external_leading_ctx +{ + int found; + int tested; + int failed; +}; + +static INT CALLBACK test_cff_external_leading_proc(const LOGFONTA *lf, const TEXTMETRICA *ntm, + DWORD type, LPARAM lparam) +{ + struct cff_external_leading_ctx *ctx = (struct cff_external_leading_ctx *)lparam; + struct + { + USHORT majorVersion; + USHORT minorVersion; + SHORT ascender; + SHORT descender; + SHORT lineGap; + } hhea; + USHORT win_ascent, win_descent; + USHORT units_per_em; + LOGFONTA test_lf; + HFONT hfont, old_hfont; + TEXTMETRICW tm; + HDC hdc; + DWORD ret; + LONG hhea_sum, win_sum, raw, msdn_px; + int ppem; + + /* Windows reports OpenType-CFF fonts with DEVICE_FONTTYPE, not + * TRUETYPE_FONTTYPE, so we cannot filter on type alone. Just skip raster + * fonts and let GetFontData('CFF ') select what we need below. */ + if (type & RASTER_FONTTYPE) + return 1; + + /* Build the font so we can probe its tables. */ + hdc = CreateCompatibleDC(NULL); + ppem = MulDiv(24, GetDeviceCaps(hdc, LOGPIXELSY), 72); + + test_lf = *lf; + test_lf.lfHeight = -ppem; + test_lf.lfWidth = 0; + test_lf.lfWeight = FW_NORMAL; + test_lf.lfCharSet = DEFAULT_CHARSET; + hfont = CreateFontIndirectA(&test_lf); + if (!hfont) + { + DeleteDC(hdc); + return 1; + } + old_hfont = SelectObject(hdc, hfont); + + /* Skip non-CFF fonts: this test only covers OpenType-CFF (.otf). */ + ret = GetFontData(hdc, MS_MAKE_TAG('C','F','F',' '), 0, NULL, 0); + if (ret == GDI_ERROR || ret == 0) + goto next; + ctx->found++; + + /* Read the metrics tables we need to evaluate the MSDN-formula prediction. */ + if (GetFontData(hdc, MS_MAKE_TAG('h','h','e','a'), 0, &hhea, sizeof(hhea)) != sizeof(hhea)) + goto next; + if (GetFontData(hdc, MS_MAKE_TAG('h','e','a','d'), 18, &units_per_em, sizeof(units_per_em)) + != sizeof(units_per_em)) + goto next; + if (GetFontData(hdc, MS_MAKE_TAG('O','S','/','2'), 74, &win_ascent, sizeof(win_ascent)) + != sizeof(win_ascent)) + goto next; + if (GetFontData(hdc, MS_MAKE_TAG('O','S','/','2'), 76, &win_descent, sizeof(win_descent)) + != sizeof(win_descent)) + goto next; + + units_per_em = GET_BE_WORD(units_per_em); + if (units_per_em == 0) goto next; + + /* MSDN formula: el = max(0, lineGap - ((winA + winD) - (Asc - Desc))). + * The hhea fields and OS/2 winAscent/Descent are big-endian on disk; the + * subexpression matches what Wine's freetype_set_outline_text_metrics() + * computes when usWinAscent + usWinDescent != 0. */ + hhea_sum = (SHORT)GET_BE_WORD(hhea.ascender) - (SHORT)GET_BE_WORD(hhea.descender); + win_sum = GET_BE_WORD(win_ascent) + GET_BE_WORD(win_descent); + raw = (SHORT)GET_BE_WORD(hhea.lineGap) - (win_sum - hhea_sum); + if (raw <= 0) + { + /* MSDN formula already gives 0 for this font; the bug isn't observable. */ + goto next; + } + msdn_px = (raw * ppem + units_per_em / 2) / units_per_em; + if (msdn_px <= 0) goto next; + + if (!GetTextMetricsW(hdc, &tm)) + goto next; + + ctx->tested++; + /* Windows GDI returns tmExternalLeading = 0 for any font containing a + * 'CFF ' table, regardless of the MSDN formula. Wine prior to the fix in + * freetype_set_outline_text_metrics() returned msdn_px instead. */ + if (tm.tmExternalLeading != 0) + ctx->failed++; + ok(tm.tmExternalLeading == 0, + "%s: CFF font tmExternalLeading = %ld, expected 0 " + "(MSDN-formula prediction was %ld at ppem %d, " + "lineGap=%d, winSum=%ld, hheaSum=%ld, upem=%u)\n", + lf->lfFaceName, tm.tmExternalLeading, msdn_px, ppem, + (SHORT)GET_BE_WORD(hhea.lineGap), win_sum, hhea_sum, units_per_em); + +next: + SelectObject(hdc, old_hfont); + DeleteObject(hfont); + DeleteDC(hdc); + return 1; +} + +static void test_CFF_external_leading(void) +{ + struct cff_external_leading_ctx ctx = {0}; + LOGFONTA lf; + HDC hdc; + + hdc = GetDC(0); + memset(&lf, 0, sizeof(lf)); + lf.lfCharSet = DEFAULT_CHARSET; + EnumFontFamiliesExA(hdc, &lf, test_cff_external_leading_proc, (LPARAM)&ctx, 0); + ReleaseDC(0, hdc); + + if (!ctx.found) + skip("no installed font has a 'CFF ' table\n"); + else if (!ctx.tested) + skip("found %d CFF fonts but none had a non-zero MSDN-formula prediction; " + "install a CJK CFF font (e.g. Source Han Sans, Adobe Source Sans, " + "FOT-NewRodin) to exercise the regression\n", ctx.found); + else + trace("test_CFF_external_leading: tested %d CFF fonts (%d failures)\n", + ctx.tested, ctx.failed); +} + static void test_nonexistent_font(void) { static const struct @@ -8067,6 +8202,7 @@ START_TEST(font) skip("Arial Black or Symbol/Wingdings is not installed\n"); test_EnumFontFamiliesEx_default_charset(); test_GetTextMetrics(); + test_CFF_external_leading(); test_RealizationInfo(); test_GetTextFace(); test_GetGlyphOutline(); -- GitLab https://gitlab.winehq.org/wine/wine/-/merge_requests/10603
I can confirm the issue with the OpenType (PostScript) version of the f.k Kikai Chokoku font[1]. From my perspective, your test code works well as a demonstration, but it feels slightly redundant for actual testing. It might be a good idea to include a dedicated font file for testing and use that in the test code, rather than selecting a suitable typeface each time. [1] https://font.kim/distribution/fk-kikaichokoku-v0320.zip -- https://gitlab.winehq.org/wine/wine/-/merge_requests/10603#note_136517
I forgot to mention that our recent guidelines[2] state that LLM-generated code is not acceptable. [2] https://gitlab.winehq.org/wine/wine/-/wikis/Developer-FAQ#can-i-contribute-c... -- https://gitlab.winehq.org/wine/wine/-/merge_requests/10603#note_136520
Thanks, that makes sense, the test was written this way because I wanted to confirm that this behavior was true for a large number of fonts I also didn't want to deal with potential font licensing issues. But now that the behavior is confirmed your approach makes sense. I will write a new test using the provided font. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/10603#note_136521
On Thu Apr 16 14:52:08 2026 +0000, Akihiro Sagawa wrote:
I forgot to mention that our recent guidelines[2] state that LLM-generated code is not acceptable. [2] https://gitlab.winehq.org/wine/wine/-/wikis/Developer-FAQ#can-i-contribute-c... OK, that is fine I will write a new test case myself.
-- https://gitlab.winehq.org/wine/wine/-/merge_requests/10603#note_136523
participants (3)
-
Akihiro Sagawa (＠sgwaki) -
Arie Miller
-
Arie Miller (＠Arimil)