diff options
| author | Andrea Canciani <ranma42@gmail.com> | 2011-07-15 22:02:01 +0200 |
|---|---|---|
| committer | Andrea Canciani <ranma42@gmail.com> | 2011-07-15 22:05:11 +0200 |
| commit | c06af104546868ed748c8f771817f5e9ae9a6a2d (patch) | |
| tree | c4e50cbebc8c969feec4f675238d384687887fa9 /pixman | |
| parent | e814b50877bf313619fbf777dcab98d39874f8a4 (diff) | |
radial: Improve documentation and naming
Add a comment to explain why the tests guarantee that the code always
computes the greatest valid root.
Rename "det" as "discr" to make it match the mathematical name
"discriminant".
Based on a patch by Jeff Muizelaar <jmuizelaar@mozilla.com>.
Diffstat (limited to 'pixman')
| -rw-r--r-- | pixman/pixman-radial-gradient.c | 31 |
1 files changed, 21 insertions, 10 deletions
diff --git a/pixman/pixman-radial-gradient.c b/pixman/pixman-radial-gradient.c index 5e9fd73..ecbca6f 100644 --- a/pixman/pixman-radial-gradient.c +++ b/pixman/pixman-radial-gradient.c @@ -78,11 +78,11 @@ radial_compute_color (double a, { /* * In this function error propagation can lead to bad results: - * - det can have an unbound error (if b*b-a*c is very small), + * - discr can have an unbound error (if b*b-a*c is very small), * potentially making it the opposite sign of what it should have been * (thus clearing a pixel that would have been colored or vice-versa) - * or propagating the error to sqrtdet; - * if det has the wrong sign or b is very small, this can lead to bad + * or propagating the error to sqrtdiscr; + * if discr has the wrong sign or b is very small, this can lead to bad * results * * - the algorithm used to compute the solutions of the quadratic @@ -92,7 +92,7 @@ radial_compute_color (double a, * * - the above problems are worse if a is small (as inva becomes bigger) */ - double det; + double discr; if (a == 0) { @@ -116,15 +116,26 @@ radial_compute_color (double a, return 0; } - det = fdot (b, a, 0, b, -c, 0); - if (det >= 0) + discr = fdot (b, a, 0, b, -c, 0); + if (discr >= 0) { - double sqrtdet, t0, t1; + double sqrtdiscr, t0, t1; - sqrtdet = sqrt (det); - t0 = (b + sqrtdet) * inva; - t1 = (b - sqrtdet) * inva; + sqrtdiscr = sqrt (discr); + t0 = (b + sqrtdiscr) * inva; + t1 = (b - sqrtdiscr) * inva; + /* + * The root that must be used if the biggest one that belongs + * to the valid range ([0,1] for PIXMAN_REPEAT_NONE, any + * solution that results in a positive radius otherwise). + * + * If a > 0, t0 is the biggest solution, so if it is valid, it + * is the correct result. + * + * If a < 0, only one of the solutions can be valid, so the + * order in which they are tested is not important. + */ if (repeat == PIXMAN_REPEAT_NONE) { if (0 <= t0 && t0 <= pixman_fixed_1) |