This reverts commit 07122e64fa.
The extra testing did find a pdf bug, and that should be fixed,
but the extra maintenance burden of running another iteration
of all tests does not seem justfied at all---particularly since
it looks like dozens of new reference images would be needed
for the svg backend.
Also, the new "failures" of the image backend with this new
testing look like a misunderstanding of exactly what the new
testing is actually drawing.
Without these checks, a user could hit an assertion failure
by passing a finished surface to cairo_surface_write_to_png.
Now we return a nice CAIRO_STATUS_SURFACE_FINISHED error in
that case instead.
As identified by Vladimir Vukicevic,
_cairo_xlib_surface_create_similar_with_format() was erroneously passing
down the source Visual when creating a surface with a different
XRenderPictFormat.
If the xserver doesn't support the RENDER extension or simply doesn't
have the matching PictFormat then xrender_format might be NULL. Check
and fallback in this case.
Test surfaces using similar surfaces with both CONTENT_COLOR and
CONTENT_COLOR_ALPHA, if applicable. This seems justified by the apparent
bugs in the pdf backend when going from an ARGB32 similar surface to
a destination RGB24 surface as well as isolated bugs in the image
backend.
The original goal was to try and trick the test suite into producing
a xlib surface with mismatching Visual/XRenderPictFormat. This succeeds,
although with a little bit of brute force in the xlib backend, but the
search to reproduce a BadMatch error fruitless.
I don't know if we made minor changes to the pdf backend, or
if this was due to a poppler change, (we recently bumped the
poppler required for testing up to 0.8.0), but any way around
it these test results look just as good as the reference images
they are replacing.
This test case does show a bug, and we have lots of ideas on how
to fix the bug, but we want to wait until after the 1.6 release
before we fix it. See:
http://bugs.freedesktop.org/show_bug.cgi?id=15349
This reverts commit 731e121c80.
This commit introduced various problems, (some likely noticeable
in the test suite, and others perhaps not). For some details, see
the latest comments in the original bug report leading to the
fix now being reverted:
http://bugs.freedesktop.org/show_bug.cgi?id=15349
In tests such as smask-mask (and others) the cairo_mask operation
is used in the construction of a mask. In this case, the single
document->mask_id value was being incremented at inappropriate
times.
We fix this by adding a new _cairo_svg_document_allocate_mask_id
that returns the current value and increments it. That way,
callers can hold onto this reliable value for the desired lifetime
that the code needs the identifier.
The smask-fill test was failing for PDF output because in some places
where the surface size is changed (eg when emitting patterns or
smasks) the cairo_to_pdf matrix was not updated.
Fix this by adding a function to handle the surface size change and
replace all the duplicated code for changing surface size with a call
to this function.
This reverts commit 5f38e215ed.
It's true that the dependencies can fall out of date, but it's really
unkind to not provide users with this information. I think we can
compromise by committing to update this information before major
releases.
Right now the two filters are implemented identically, so there's
no real change for now. But in the future, it's conceivable that an
X server could implement some crazy, high-quality filter for BEST
without regard to performance, (since that's what BEST means).
Meanwhile, GOOD actually captures what we want by default which is
a good mix of both quality and performance.
It appears the reference image wasn't generated on a gold-standard
machine, (and yes, the gold standard is, *ahem* whatever happens
to match what Carl's machine does---eventually we'll make the test-
based tests less picky about system configuration like this).
make -j check has the unfortunate side-effect of performing the summary
before the TESTS have completed, both hiding in the stream of test
output and making it useless. Forcibly disable the jobserver so that the
summary is always last at the expense of not compiling the tests in parallel.
If the .notdef glyph is the first glyph in the subset to be mapped in
scaled font, we do not know if the subset will scaled or unscaled. We
can put it in the unscaled subset as Type1-fallback will embded an
empty glyph if it can not get the path.
Bug 15348 references the following PDF that was printing incorrectly
when running through poppler and cairo to generate PostScript.
http://launchpadlibrarian.net/12921700/UbuntuDesktop.pdf
The PostScript output had too much space between each word causing
strings of glyphs printed with the TJ operator to overlap.
The original PDF file contains an CFF font with CID Identity-H
encoding. The PDF file is using character code 0 (glyph 0 due to
Identity-H encoding) as a space character. The CFF specification
defines glyph 0 to be the .notdef glyph.
The PS backend subsets CFF fonts as a Type1-fallback
font. Type1-fallback creates it's own empty .notdef glyph with an
arbitrary glyph advance of 500. The problem here is the TJ operator
used to output the glyphs depends on the glyph advance being
correct. pdf-operators.c uses the glyph advance returned by
_scaled_glyph_init(). However the PostScript interpreter sees the
glyph advance of 500 for .notdef. This problem does not occur when
generating PDF as the PDF font dictionary contains an list of glyph
advances that override the font metrics.
Fix this by making Type1-fallback not treat .notdef as special and to
create it the path and metrics obtained from _scaled_glyph_init(). As
a special case, make it not fail if _scaled_glyph_init() is unable to
return a path for .notdef. This was probably the reason Type1-fallback
previously created it's own .notdef glyph as calling
_scaled_glyph_init(_GLYPH_INFO_PATH) for glyph 0 returns
CAIRO_INT_STATUS_UNSUPPORTED for some fonts.
This ensures the Type1-fallback font metrics match the metrics used
by pdf-operators.c to position the glyphs. This also results in the
removal of some duplicated code.
5050c55f93 fixed type1-fallback to use the glyph advance instead of
glyph width in the stored glyph metrics in the font. However it did
not fix the same bug in Type 2 charstrings (used by CFF fallback in
PDF). This problem was not noticed since the glyph widths in the PDF
font dictionary overrides these values.
Fix this in case any software reading cairo PDFs uses these values.
If _cairo_scaled_glyph_lookup() returns CAIRO_INT_STATUS_UNSUPPORTED
it will be caught by the ASSERT_NOT_REACHED in
_emit_unscaled_font_subset in PS/PDF. It is more useful to catch this
closer to the source.
This fixes the filter-bilinear-extents test case and the
related bug entry:
bad clipping with EXTEND_NONE
http://bugs.freedesktop.org/show_bug.cgi?id=15349
Though there are still differences in the PDF and PostScript
backends, (primarily because we can't capture cairo's filter
modes in those file formats).
This test exercises code that computes the extents of a surface
pattern with CAIRO_FILTER_BILINEAR, (where the filtering
effectively increases the extents of the pattern).
The original bug was reported by Owen Taylor here:
bad clipping with EXTEND_NONE
http://bugs.freedesktop.org/show_bug.cgi?id=15349
This covers the known-to-broken 1.3 servers such as appeared
in Fedora 8. It also leaves the workaround off, (since it's
a severe slowdown), for the known-to-be-working 1.4.99.901
server as appears in Fedora 9 Betas.
XRenderAddGlyph() does not split its image data across multiple requests
and so the glyph surface must be smaller than XMaxRequestSize or else
the server will disconnect the client, causing "Fatal IO error 104".
As this will require an extension to the XRender spec, we can work
around the issue by using our fallbacks if we detect that the glyph will
be too large for a single request.
See bugs https://bugs.freedesktop.org/show_bug.cgi?id=4339 and
http://bugs.freedesktop.org/show_bug.cgi?id=13266 for examples.
