104573 2012-12-10 10:40:42 -0800 [EFL][WK2] Layers are shifted vertically by 1-2 pixels in compositing/geometry tests. 2012-12-30 22:26:36 -0800 1 1 1 Unclassified WebKit Layout and Rendering 528+ (Nightly build) Unspecified Unspecified RESOLVED DUPLICATE 102374 P2 Normal --- 1 ostap73 webkit-unassigned cdumez d-r laszlo.gombos pnormand zimmermann oldest_to_newest 787515 0 ostap73 2012-12-10 10:40:42 -0800 This bug is to keep track and summarize my findings about compositing/geometry tests on EFL port. Currently it doesn't apply to EFL/wk1 because accelerated compositing is disabled. Probably GTK port has the same problem, but they do not support AC now. The easiest way problem reproduced on compositing/geometry/composited-in-columns.html tests. Other geometry tests also may fail or flaky. Problem appears when RenderBoxes are positioned during layout. Layout is based on line height + line gap. For SimpleFontDataFreeType line hight is calculated like ascender + descender (FontMetrics.h) . Line gap is calculated like font_extents.height - font_extents.ascent - font_extents.descent (SimpleFontData::platformInit()). When I run this test font_extents.height is less than font_extents.ascent + font_extents.descent and line gap becomes negative (-1) and this affects page layout. font_extents are received from cairo_scaled_font_extents call and internally cairo gets it from freetype font library. Freetype loads LiberationSerif-Regular.ttf font and it seems if this font is loaded 1st it gets wrong metric in face->size->metrics . I tried loading this test into Minibrowser. In this case font is loaded with correct metrics and page layouts correctly. The problem looks very similar to this bug in freetype library: http://savannah.nongnu.org/bugs/?34156 EFL currently uses 2.4.2. I tried upgrading to 2.4.10 and it didn't help. 787721 1 ostap73 2012-12-10 14:32:56 -0800 r101343 removed this code with comment: - // There seems to be some rounding error in cairo (or in how we - // use cairo) with some fonts, like DejaVu Sans Mono, which makes - // cairo report a height smaller than ascent + descent, which is - // wrong and confuses WebCore's layout system. Workaround this - // while we figure out what's going on. - float lineSpacing = font_extents.height; - if (lineSpacing < font_extents.ascent + font_extents.descent) - lineSpacing = font_extents.ascent + font_extents.descent; It seems this exactly what happens here. Shouldn't we check that lineGap (in new code) is non-negative? 798294 2 ostap73 2012-12-30 22:26:36 -0800 *** This bug has been marked as a duplicate of bug 102374 ***