diff options
| author | wl <wl> | 2009-09-01 05:26:30 +0000 |
|---|---|---|
| committer | wl <wl> | 2009-09-01 05:26:30 +0000 |
| commit | 88563e645a8ab2d41b0b14fb4141d341f70eeb9a (patch) | |
| tree | a24a9197074cbd1560951a2d2bc989b2584e0863 /src | |
| parent | 41dbc33cb500b495ae5372c5a7797c5bc74a931a (diff) | |
| download | groff-88563e645a8ab2d41b0b14fb4141d341f70eeb9a.tar.gz | |
[pic] Fix a border case for arc computation.
* src/preproc/pic/object.cpp (object_spec::make_arc): It can happen
that test `radius < d' in the loop is satisfied, but the difference
in the two values is on the order of 1e-10. If `radius' is small,
doubling the value can lead to a fairly gross error.
The original code appears to have been intended to deal with the
situation when radius is orders of magnitude less than `d'. The
replacement code simply assigns `radius' the smallest value that
avoids problems with the floating point code further on in the
routine.
Diffstat (limited to 'src')
| -rw-r--r-- | src/preproc/pic/object.cpp | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/src/preproc/pic/object.cpp b/src/preproc/pic/object.cpp index 62415793..a3146040 100644 --- a/src/preproc/pic/object.cpp +++ b/src/preproc/pic/object.cpp @@ -1830,8 +1830,8 @@ linear_object *object_spec::make_arc(position *curpos, direction *dirp) if (radius <= 0) radius = .25; // make the radius big enough - while (radius < d) - radius *= 2.0; + if (radius < d) + radius = d; double alpha = acos(d/radius); double theta = atan2(h.y, h.x); if (cw) |