[cairo] _cairo_surface_wrapper_get_target_extents() too small

[Guillaume Ayoub]

Le vendredi 29 mars 2013 à 15:43 +0100, Simon Sapin a écrit :
> Le 27/03/2013 13:12, Simon Sapin a écrit :
> > Hi,
> >
> > When rendering to PDF with WeasyPrint a document containing a SVG image,
> > SVG elements in the right and bottom of the image are not shown in some
> > cases.
> >
> > git bisect indicates that the bug appeared in this cairo commit:
> >
> > http://cgit.freedesktop.org/cairo/commit/?id=09b42c748e9dbcc923560c7d8bf5298fbffe95ef
> >
> > Reverting it on top of today’s master fixes the bug.
> >
> >
> > WeasyPrint has CairoSVG render the image to an intermediate cairo SVG
> > surface, which is then used in surface pattern as a source to paint in
> > the final PDF surface. The bug appears when the target context has a
> > scale to make the image smaller than its "intrinsic" size. Making the
> > SVG surface smaller (with its content scaled accordingly) and
> > compensating with a scale on the target surface works around the problem.
> >
> > Apparently, the rectangle returned by
> > _cairo_surface_wrapper_get_target_extents() is smaller than it should
> > be. Something is transformed wrong, but I’m not sure what coordinate
> > system each rectangle is supposed to be in.
> 
> 
> Based on further research, it appears that the wrapper->transform value 
> in _cairo_surface_wrapper_get_target_extents() should account for the 
> target context’s transformation matrix, but doesn’t. wrapper->transform 
> is always the identity matrix in our tests, even though the context has 
> a non-identity transform when painting the surface pattern.
> 
> Removing any calls to cairo_save() and cairo_restore() also works around 
> the issue, but for reasons we do not understand.

I finally think that I have found what's happening.

In cairo-analysis-surface, the transformation matrix is stored in
surface->ctm, and this matrix is handled by _add_operation for all the
operations added in the analysis surface. In
_analyze_recording_surface_pattern, the ctm surface is set, everything
is OK for the future operations.

But _analyze_recording_surface_pattern calls
_cairo_recording_surface_replay_and_create_regions, calling
_cairo_recording_surface_replay_internal, calling
_cairo_surface_wrapper_get_target_extents. As replay_and_create_regions
calls replay_internal without the surface_transform parameter, the
original transformation matrix is lost. get_target_extents relies on
wrapper.transform that is not set, and get_target_extents then uses the
original size of the target to reduce the extents rectangle, instead of
using the size multiplied by the transformation matrix.

A simple solution to fix this is to add a surface_transform parameter to
_cairo_recording_surface_replay_and_create_regions (as it is done in
_cairo_recording_surface_replay_with_clip), and to use the invert matrix
of &tmp->ctm as surface_transform parameter of
_cairo_recording_surface_replay_and_create_regions.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 490 bytes
Desc: This is a digitally signed message part
URL: <http://lists.cairographics.org/archives/cairo/attachments/20130329/76a5894a/attachment.pgp>