Propagate status errors from pattern->status to cr->status.
Originally 2005-05-08 Owen Taylor <otaylor@redhat.com>:
src/cairo-pattern.c src/cairoint.h: If allocation of pattern objects fails, return special static nil pattern objects.
If adding a color stop fails to allocate memory, set pattern->status. (And fix a memory leak.) Make public functions return when pattern->status is set, (and no longer return a cairo_status_t).
src/cairo-pattern.c src/cairo.h doc/public/cairo-sections.txt: Add cairo_pattern_status()
Check the status of gstate->source and of mask patterns passed in.
If cairo_create() fails, return a special static object, cairo_nil.
Don't return early if cr->status is set. cr->status should not affect reference counting.
Ignore any magic object with a reference count of -1.
Add a status field to cairo_path_t.
Add documentation for the new approach for handling errors in these functions---always returning a valid pointer with at least a status. (cairo_append_path): Propagate path status errors to the context. Add note to documentation on initializing path->status.
Add missing cairo_private qualifier to a couple functions.
Track new status field in cairo_path_t. (cairo_path_destroy): Don't destroy cairo_path_nil. Add documentation. (_cairo_path_data_create): (_cairo_path_data_create_flat): (_cairo_path_data_append_to_context): Add documentation. (_cairo_path_data_create_in_error): New function to create a placeholder cairo_path_t just to propagate a cairo_status_t error.
Also in this patch is a change to the xlib surface that creates two separate Pictures, one for source and one for destination operands which separates the source clipping from destination clipping. Cairo now specifies that sources are never clipped by any clipping applied to them as destinations.
Move cairo_clip_t (renamed from cairo_clip_rec_t) from cairoint.h to cairo-gstate-private.h. Eliminate stack of clip state from surfaces. Add new surface clipping API.
Manage clip objects entirely within the gstate, loading the whole thing into the surface just before drawing.
Source surfaces need not have clipping modified as the surface interface now specifies that source surfaces are always unclipped.
Eliminate nested clipping contexts, leaving clip management entirely to the gstate. Create new clip API for the gstate which uses per-surface serial numbers to match gstate clipping against current surface clipping values.
Surfaces no longer track clipping regions at all, so the old _cairo_surface_get_clip_extents has been replaced with _cairo_surface_get_extents. For PDF/PS surfaces, this function is expected to return a rectangle covering the entire fixed point coordinate space to leave rendering unclipped by the surface.
Region clipping capability is now signalled by a non-NULL function pointer in set_clip_region.
Each surface now contains two Pictures, one for source and one for destination operands so that source operands are never clipped by destination clipping.
CAIRO_STATUS_BAD_NESTING removed
self-copy now passes (Xlib only, until libpixman changes land)
reviewed by: krh, otaylor, cworth
Add --only-section-tmpl option so that changes to inline documentation does not lead to churn in the .sgml template files.
Commit new templates now that gtk-doc has ripped all the inline portions out.
New template files added for new sections.
Update to match current API.
Make parameter names match those in the .c file and its documentation.
Add cairo_xlib_surface_create_with_render_format.
Reduce Xlib constructors down to two simple forms. Add width, height to constructors and eliminate any synchronous size queries from the implementation.
Update to match new cairo_xlib_surface_t create functions.
Remove destroy_closure argument from cairo_pdf_surface_create_for_stream. Rename width,height to width_in_points, height_in_points for better clarity.
Brush a bunch of dust off of the PS backend and bring it up to date with the latest API conventions from the PDF backend. These include: accepting a filename rather than a FILE in the primary constructor, providing a stream-based interface for more flexibility, and accepting a surface size in device-space units (points) rather than inches.
Make it a little more clear that the width and height being passed around are in units of points.
Update to the latest cairo-ps.h changes as described above. Notice how much more sane things become now that the surface size is described in device-space units.
This scale value is used only when computing gradient values before the defined range, in which case stop 0 is used for both ends of the interpolation, making the value of 'scale' not actually matter, except that valgrind notices we're using an undefined value.
