Use the return value to return the result from _cairo_hash_table_lookup()
(as opposed to filling an output parameter on the stack) as this (a)
results in cleaner code (no strict-alias breaking pointer casts), (b)
produces a smaller binary and (c) is measurably faster.
As we obey the rule in Cairo that we only insert if we know that there is
no existing entry in the hash table, we can therefore perform a much quicker
search knowing that the key is unique.
A limitation of the current API was that the destroy notifier was called
on the mime-data block. This prevents the user from passing in a pointer
to a managed block, for example a mime-data block belonging to a
ref-counted object. We can overcome this by allowing the user to specify
the closure to be used with the destroy notifier.
Reading through the previous commit spotted that the arguments to
edge_compare_for_y_against_x were transposed, but the test-suite had
failed to catch detect it. This is due that in order to actually
solve the equation we need to have a diagonal edge passing near an
off-centre point of interest, which was not among the test cases. So add
some off-centre tests to fully exercise the code.
Benjamin Otte reports that in one particular benchmark cairo_in_fill() is
a hotspot in the profile. Currently we tessellate the entire path and then
search for a containing trapezoid. This is very expensive compared to the
simple method of counting the number of edge crossing between the point of
interest and x=-∞. For example, this speeds tessellate-256 up by almost 3
orders of magnitude.
Write out the original PNG mime-data if attached to the surface during
cairo_surface_write_to_png(). Similar to how the compressed alternate
representations are handled by the other backends.
Note: by automatically attaching and using the mime-data in preference to
the image data, we break the read_from_png(); draw(); write_to_png()
cycle.
I moved the pixel centre to xc,yc but forgot to remove it during
compensation - as caught by the test suite.
Refresh a couple of reference images that depend upon exact pixel-centre
rounding conditions.
pixman limits the src] co-ordinates (and thus [xy]_offset] to 16bits,
so we need to be careful how much of the translation vector to push into
[xy]_offset. Since the range is the same for both, split the integer
component between the matrix and the offset.
test/scale-offset* now at least shows the source image, even if it is
misplaced.
set_font_face was not consuming it's operand but blithely placing an
undefined font_face onto the operand stack, whereas set_source was
performing invalid exchanges on the stack.
<adrianj> ickle_: If we are going to use a different image for jpeg in
mime-data maybe we could create a jpg that contains the text "jpeg". That
way when support for the other image formats is added the mime-data test
could have one image for each type with each image and it is easy to see
that each image is the correct one.
Instead of doing a full-copy of the mime data (which can be 10K-100K,
or even larger) just copy a reference to the original mime to the
snapshot surface (as suggested by Behdad).
