The _cairo_ft_unscaled_font_fini() was skipped during the destroy for
fonts generated by cairo_ft_font_face_create_for_ft_face(). However
this causes a mutex to be 'leaked' for each font.
Every time we assign or return a hard-coded error status wrap that value
with a call to _cairo_error(). So the idiom becomes:
status = _cairo_error (CAIRO_STATUS_NO_MEMORY);
or
return _cairo_error (CAIRO_STATUS_INVALID_DASH);
This ensures that a breakpoint placed on _cairo_error() will trigger
immediately cairo detects the error.
As pointed out by Jeff Muizelaar, this allows for more concise code, as
_cairo_error(CAIRO_STATUS_NO_MEMORY)
return CAIRO_STATUS_NO_MEMORY
can become
return _cairo_error(CAIRO_STATUS_NO_MEMORY);
Since the objects can be shared and may be in use simultaneously across
multiple threads, setting the status needs to be atomic. We use a
locked compare and exchange in order to avoid overwriting an existing
error - that is we use an atomic operation that only sets the new status
value if the current value is CAIRO_STATUS_SUCCESS.
At some point during the blitz, I accidentally wrote
_cairo_error (CAIRO_STATUS_SUCCESS) and then proceeded to paste it into
the next 30 error sites! s/CAIRO_STATUS_SUCCESS/CAIRO_STATUS_NO_MEMORY/
As we now generate empty paths, we must be able to handle empty paths
in the user facing API. cairo_append_path() has an explicit check, and
raises an error, for a NULL path->data, so we need to check the
path->num_data first for empty paths.
pixman does not (yet?) support arbitrary strides and will fail to
generate an image for the data. We misinterpret that failure as a
CAIRO_STATUS_NO_MEMORY, so instead return CAIRO_STATUS_INVALID_FORMAT
before attempting to create the pixman image.
Avoid returning the "generic" _cairo_surface_nil (which corresponds to
CAIRO_STATUS_NO_MEMORY) when the user asks us to create a surface with
an invalid format or content.
Generate a real empty path structure instead of returning
_cairo_path_nil, if we have been asked to create an empty path.
(Also add a couple of missing _cairo_error()s and an appropriate test
case.)
Spotted by Fred Kiefer.
libtool causes the .gcda files to be generated in the .libs/ directory,
separate from the .c source file. lcov expects them to be in the same
directory - so massage the lcov.info file to remove the reference to the
.libs/.
Also separate the target for generating the lcov output, so that it can
be run independently from triggering the tests. And improve convenience
of using the other lcov targets.
Moments after pushing the new test case did I realise the issue...
We do not attempt to write out the surface to the user stream until
we perform the cairo_surface_destroy() by which point we have lost
the ability to interrogate the error status. We can avoid this by
explicitly calling cairo_surface_finish() and then checking the
error status - and we see that the error is indeed reported
correctly.
No bug. Nothing to see here. Please move along. (Apart from the
request for the status to be return from cairo_surface_destroy!)
From bug https://bugs.freedesktop.org/show_bug.cgi?id=7049, we find
that the error status from the user supplied write function to
cairo_*_surface_create_for_stream is ignored and not propagated back
to the surface/context - leading to silent data loss. Incorporate
the suggested test case, a write function that simply returns
CAIRO_STATUS_WRITE_ERROR, into create-for-stream.c.
Use the png_struct->error_ptr to propagate the error status from the
user/stdio read and write functions through the png_error() to the
cairo_surface_write_to_png*() and cairo_surface_read_from_png*()
functions. From there the error is returned back to the user either
directly or as the most appropriate error surface.
(Fixes https://bugs.freedesktop.org/show_bug.cgi?id=6909)
When limiting the target instruction set for compatibility with older
processors, e.g. -march=i386, the Intel atomic primitives generate a
call to a non-existent function, __sync_fetch_and_add_4(). To detect
this scenario change the configure test from AC_TRY_COMPILE to
AC_TRY_LINK which will then cause Cairo to fall back to using mutexes
for its atomic operations.
Introduce an opaque cairo_reference_count_t and define operations on it
in terms of atomic ops. Update all users of reference counters to use
the new opaque type.
Test for the availability of the Intel __sync_* atomic primitives and
use them to define a few operations useful for reference counting -
providing a generic interface that may be targeted at more architectures
in the future. If no atomic primitives are available, use a mutex based
variant. If the contention on that mutex is too high, we can consider
using an array of mutexes using the address of the atomic variable as
the hash.
The wrapping of GCC attributes (such as cairo_private) needs to be
visible to any header file, including those that avoid cairoint.h such
as cairo-boilerplate. To achieve this we move the pre-processor magic to
its own header file and include it as required.
The content stream compression that was previously implemented was
inadvertently bypassed when the new stream handling for meta surface
patterns was implemented.
The analysis surface will calculated the tight bounding box for each
page. A new paginated-surface backend function set_bounding_box() has
been added for passing the page bounding box to the target surface at
the end of the analysis phase.
The changes to the PS file when EPS is enabled are:
- Add EPS header
- Use tight bounding box instead of page size
- Use save/restore to ensure PS interpreter is left in the same state
Previously, the TrueType subsetting would fail if any of the "cvt",
"fpgm", or "prep" tables were missing from the source font. However
these tables are optional and not required in the subsetted font if
they do not appear in the source font.
The "name" table has been removed from the subsetted font as the
Type42 specification does not require this table.
If atsui and ft were both enabled, the code crashed trying to subset
type-1 fonts; fixed by checking if fonts really are ft before using
them as ft fonts. This is a temporary fix until we support subsetting
across all font backends.
