1 - Things to keep in mind if breaking ABI:
3 - There should be a guard #ifndef I_AM_EITHER_CAIRO_OR_THE_X_SERVER
5 - X server will require 16.16 essentially forever. Can we get
6 the required precision by simply adding offset_x/y to the
7 relevant rendering API?
9 - pixman_image_set_indexed() should copy its argument, and X
10 should be ported over to use a pixman_image as the
11 representation of a Picture, rather than creating one on each
14 - We should get rid of pixman_set_static_pointers()
16 - We should get rid of the various trapezoid helper functions().
17 (They only exist because they are theoretically available to
20 - Regions should become 32 bit
22 - There should only be one trap rasterization API.
24 - Make pixman_region_point_in() survive a NULL box, then fix up
29 - Add a general way of dealing with architecture specific
30 fast-paths. The current idea is to have each operation that can
31 be optimized is called through a function pointer that is
32 initially set to an initialization function that is responsible for
33 setting the function pointer to the appropriate fast-path.
35 - Go through things marked FIXME
37 - Add calls to prepare and finish access where necessary. grep for
38 ACCESS_MEM, and make sure they are correctly wrapped in prepare
41 - restore READ/WRITE in the fbcompose combiners since they sometimes
42 store directly to destination drawables.
44 - It probably makes sense to move the more strange X region API
45 into pixman as well, but guarded with PIXMAN_XORG_COMPATIBILITY
47 - Reinstate the FbBits typedef? At the moment we don't
48 even have the FbBits type; we just use uint32_t everywhere.
50 Keith says in bug 2335:
52 The 64-bit code in fb (pixman) is probably broken; it hasn't been
53 used in quite some time as PCI (and AGP) is 32-bits wide, so
54 doing things 64-bits at a time is a net loss. To quickly fix
55 this, I suggest just using 32-bit datatypes by setting
56 IC_SHIFT to 5 for all machines.
58 - Consider whether calling regions region16 is really such a great
59 idea. Vlad wants 32 bit regions for Cairo. This will break X server
60 ABI, but should otherwise be mostly harmless, though a
61 pixman_region_get_boxes16() may be useful.
63 - Consider optimizing the 8/16 bit solid fills in pixman-util.c by
64 storing more than one value at a time.
66 - Add an image cache to prevent excessive malloc/free. Note that pixman
67 needs to be thread safe when used from cairo.
69 - Moving to 24.8 coordinates. This is tricky because X is still
70 defined as 16.16 and will be basically forever. It's possible we
71 could do this by adding extra offset_x/y parameters to the
72 trapezoid calls. The X server could then just call the API with
73 (0, 0). Cairo would have to make sure that the delta *within* a
74 batch of trapezoids does not exceed 16 bit.
76 - Review the pixman_format_code_t enum to make sure it will support
77 future formats. Some formats we will probably need:
79 ARGB/ABGR with 16/32/64 bit integer/floating channels
83 Also we may need the ability to distinguish between PICT_c8 and
84 PICT_x4c4. (This could be done by interpreting the A channel as
85 the depth for TYPE_COLOR and TYPE_GRAY formats).
87 A possibility may be to reserve the two top bits and make them
88 encode "number of places to shift the channel widths given" Since
89 these bits are 00 at the moment everything will continue to work,
90 but these additional widths will be allowed:
92 All even widths between 18-32
93 All multiples of four widths between 33 and 64
94 All multiples of eight between 64 and 128
96 This means things like r21g22b21 won't work - is that worth
97 worrying about? I don't think so. And of course the bpp field
98 can't handle a depth of over 256, so > 64 bit channels arent'
99 really all that useful.
101 We could reserve one extra bit to indicate floating point, but
104 PIXMAN_TYPE_ARGB_FLOAT
105 PIXMAN_TYPE_BGRA_FLOAT
108 image types. With five bits we can support up to 32 different
109 format types, which should be enough for everybody, even if we
110 decide to support all the various video formats here:
112 http://www.fourcc.org/yuv.php
114 It may make sense to have a PIXMAN_TYPE_YUV, and then use the
115 channel bits to specify the exact subtype.
117 Another possibility is to add
122 where the channel widths would get 16 and 32 added to them,
125 What about color spaces such a linear vs. srGB etc.?
129 - Get rid of the switch-of-doom; replace it with a big table
130 describing the various fast paths.
132 - Make source clipping optional.
133 - done: source clipping happens through an indirection.
134 still needs to make the indirection settable. (And call it
137 - Run cairo test suite; fix bugs
138 - one bug in source-scale-clip
140 - Remove the warning suppression in the ACCESS_MEM macro and fix the
141 warnings that are real
144 - make the wrapper functions global instead of image specific
145 - this won't work since pixman is linked to both fb and wfb
147 - Add non-mmx solid fill
149 - Make sure the endian-ness macros are defined correctly.
151 - The rectangles in a region probably shouldn't be returned const as
152 the X server will be changing them.
154 - Right now we _always_ have a clip region, which is empty by default.
155 Why does this work at all? It probably doesn't. The server
156 distinguishes two cases, one where nothing is clipped (CT_NONE), and
157 one where there is a clip region (CT_REGION).
159 - Default clip region should be the full image
161 - Test if pseudo color still works. It does, but it also shows that
162 copying a pixman_indexed_t on every composite operation is not
163 going to fly. So, for now set_indexed() does not copy the
166 Also just the malloc() to allocate a pixman image shows up pretty
171 - Make all the setters not copy their arguments
173 - Possibly combined with going back to the stack allocated
174 approach that we already use for regions.
176 - Keep a cached pixman_image_t around for every picture. It would
177 have to be kept uptodate every time something changes about the
180 - Break the X server ABI and simply have the relevant parameter
181 stored in the pixman image. This would have the additional benefits
184 - We can get rid of the annoying repeat field which is duplicated
187 - We can use pixman_color_t and pixman_gradient_stop_t
188 etc. instead of the types that are defined in