RE: [ft] Glitchy stroker output

[Stephan T. Lavavej]

> Yes.  I have to step through the whole code... This may take some
> time.

Let's see if I can save you some time...


I tried to reduce the outline in order to understand what's making
FT_Outline_Render() go bonkers.  Here's what I came up with:

http://stl.nuwen.net/pub/freetype/repro4.c

n_contours = 1
n_points = 4
points[0] = (10, 0) on the curve
points[1] = (0, 10) on the curve
points[2] = (-10, 0) on the curve
points[3] = (0, -10) on the curve
contours[0] = 3
flags = 0
coverage 128 | y -10 | x [ -1,   0]
coverage 128 | y  -9 | x [ -2,  -2]
coverage 255 | y  -9 | x [ -1,  -1]
coverage 128 | y  -9 | x [  0,   0]
coverage 128 | y  -8 | x [ -3,  -3]
coverage 255 | y  -8 | x [ -2,  -1]
coverage 128 | y  -8 | x [  0,   0]
coverage 128 | y  -7 | x [ -4,  -4]
coverage 255 | y  -7 | x [ -3,  -1]
coverage 128 | y  -7 | x [  0,   0]
coverage 128 | y  -6 | x [ -5,  -5]
coverage 255 | y  -6 | x [ -4,  -1]
coverage 128 | y  -6 | x [  0,   0]
coverage 128 | y  -5 | x [ -6,  -6]
coverage 255 | y  -5 | x [ -5,  -1]
coverage 128 | y  -5 | x [  0,   0]
coverage 128 | y  -4 | x [ -7,  -7]
coverage 255 | y  -4 | x [ -6,  -1]
coverage 128 | y  -4 | x [  0,   0]
coverage 128 | y  -3 | x [ -8,  -8]
coverage 255 | y  -3 | x [ -7,  -1]
coverage 128 | y  -3 | x [  0,   0]
coverage 128 | y  -2 | x [ -9,  -9]
coverage 255 | y  -2 | x [ -8,  -1]
coverage 128 | y  -2 | x [  0,   0]
coverage 128 | y  -1 | x [-10, -10]
coverage 255 | y  -1 | x [ -9,  -1]
coverage 128 | y  -1 | x [  0,   0]
coverage 128 | y   0 | x [-10, -10]
coverage 255 | y   0 | x [ -9,  -1]
coverage 128 | y   0 | x [  0,   0]
coverage 128 | y   1 | x [ -9,  -9]
coverage 255 | y   1 | x [ -8,  -1]
coverage 128 | y   1 | x [  0,   0]
coverage 128 | y   2 | x [ -8,  -8]
coverage 255 | y   2 | x [ -7,  -1]
coverage 128 | y   2 | x [  0,   0]
coverage 128 | y   3 | x [ -7,  -7]
coverage 255 | y   3 | x [ -6,  -1]
coverage 128 | y   3 | x [  0,   0]
coverage 128 | y   4 | x [ -6,  -6]
coverage 255 | y   4 | x [ -5,  -1]
coverage 128 | y   4 | x [  0,   0]
coverage 128 | y   5 | x [ -5,  -5]
coverage 255 | y   5 | x [ -4,  -1]
coverage 128 | y   5 | x [  0,   0]
coverage 128 | y   6 | x [ -4,  -4]
coverage 255 | y   6 | x [ -3,  -1]
coverage 128 | y   6 | x [  0,   0]
coverage 128 | y   7 | x [ -3,  -3]
coverage 255 | y   7 | x [ -2,  -1]
coverage 128 | y   7 | x [  0,   0]
coverage 128 | y   8 | x [ -2,  -2]
coverage 255 | y   8 | x [ -1,  -1]
coverage 128 | y   8 | x [  0,   0]
coverage 128 | y   9 | x [ -1,   0]

The outline is (I think) a diamond centered at (x = 0, y = 0).  At y = 0,
the outline should span x = [-9, 9] or [-10, 10] or something like that.
(I'm not exactly sure how the outline coordinates relate to the pixel
coordinates, but it doesn't matter.)  Instead, we get:

coverage 128 | y   0 | x [-10, -10]
coverage 255 | y   0 | x [ -9,  -1]
coverage 128 | y   0 | x [  0,   0]

Truncation! I haven't rendered this as a bitmap, but none of the
x-coordinates go beyond 0, so we've clearly got right-side mangling.

Now, let's see what happens when I simply translate the diamond to be
centered at (x = 10, y = 10):

n_contours = 1
n_points = 4
points[0] = (20, 10) on the curve
points[1] = (10, 20) on the curve
points[2] = (0, 10) on the curve
points[3] = (10, 0) on the curve
contours[0] = 3
flags = 0
coverage 128 | y   0 | x [  9,  10]
coverage 128 | y   1 | x [  8,   8]
coverage 255 | y   1 | x [  9,  10]
coverage 128 | y   1 | x [ 11,  11]
coverage 128 | y   2 | x [  7,   7]
coverage 255 | y   2 | x [  8,  11]
coverage 128 | y   2 | x [ 12,  12]
coverage 128 | y   3 | x [  6,   6]
coverage 255 | y   3 | x [  7,  12]
coverage 128 | y   3 | x [ 13,  13]
coverage 128 | y   4 | x [  5,   5]
coverage 255 | y   4 | x [  6,  13]
coverage 128 | y   4 | x [ 14,  14]
coverage 128 | y   5 | x [  4,   4]
coverage 255 | y   5 | x [  5,  14]
coverage 128 | y   5 | x [ 15,  15]
coverage 128 | y   6 | x [  3,   3]
coverage 255 | y   6 | x [  4,  15]
coverage 128 | y   6 | x [ 16,  16]
coverage 128 | y   7 | x [  2,   2]
coverage 255 | y   7 | x [  3,  16]
coverage 128 | y   7 | x [ 17,  17]
coverage 128 | y   8 | x [  1,   1]
coverage 255 | y   8 | x [  2,  17]
coverage 128 | y   8 | x [ 18,  18]
coverage 128 | y   9 | x [  0,   0]
coverage 255 | y   9 | x [  1,  18]
coverage 128 | y   9 | x [ 19,  19]
coverage 128 | y  10 | x [  0,   0]
coverage 255 | y  10 | x [  1,  18]
coverage 128 | y  10 | x [ 19,  19]
coverage 128 | y  11 | x [  1,   1]
coverage 255 | y  11 | x [  2,  17]
coverage 128 | y  11 | x [ 18,  18]
coverage 128 | y  12 | x [  2,   2]
coverage 255 | y  12 | x [  3,  16]
coverage 128 | y  12 | x [ 17,  17]
coverage 128 | y  13 | x [  3,   3]
coverage 255 | y  13 | x [  4,  15]
coverage 128 | y  13 | x [ 16,  16]
coverage 128 | y  14 | x [  4,   4]
coverage 255 | y  14 | x [  5,  14]
coverage 128 | y  14 | x [ 15,  15]
coverage 128 | y  15 | x [  5,   5]
coverage 255 | y  15 | x [  6,  13]
coverage 128 | y  15 | x [ 14,  14]
coverage 128 | y  16 | x [  6,   6]
coverage 255 | y  16 | x [  7,  12]
coverage 128 | y  16 | x [ 13,  13]
coverage 128 | y  17 | x [  7,   7]
coverage 255 | y  17 | x [  8,  11]
coverage 128 | y  17 | x [ 12,  12]
coverage 128 | y  18 | x [  8,   8]
coverage 255 | y  18 | x [  9,  10]
coverage 128 | y  18 | x [ 11,  11]
coverage 128 | y  19 | x [  9,  10]

The y = 10 spans are:

coverage 128 | y  10 | x [  0,   0]
coverage 255 | y  10 | x [  1,  18]
coverage 128 | y  10 | x [ 19,  19]

Perfect!

Experimenting further:

A diamond centered at (x = 10, y = 0), which would be "FT_Vector points[] =
{ { 1280, 0 }, { 640, 640 }, { 0, 0 }, { 640, -640 } };", which has
non-negative x coordinates and some negative y coordinates, renders
perfectly.

A diamond centered at (x = 0, y = 10), which would be "FT_Vector points[] =
{ { 640, 640 }, { 0, 1280 }, { -640, 640 }, { 0, 0 } };", which has some
negative x coordinates and non-negative y coordinates, experiences
truncation.

A diamond centered at (x = -10, y = 0), which would be "FT_Vector points[] =
{ { 0, 0 }, { -640, 640 }, { -1280, 0 }, { -640, -640 } };", isn't rendered
at all!

Therefore, it appears that negative x coordinates make FT_Outline_Render()
produce mangled spans.
 
Thanks,
Stephan