> I have gone back and done more coding and more testing. The regression
> test for geometry no longer causes the back end to core dump. The
> abort was not happening the "close_ps()" function that I had hacked,
> but is in interpt_sl() routine. This routine dumps core if asked to
> find an intersection between a line segment and a line which in fact
> do not intersect. What I did was to fix close_ps() to not call
> interpt_sl() with parameters that do not intersect. I handle such
> special cases separately (and hopefully cleanly) in close_ps().
> Please let me know what you think. If you think of I will clean up
> and send proper patches (in right order this time, hopefully!)
Things look good. Would it be possible to fix interpt_sl() while you are
looking at this? Otherwise it will lurk in the code waiting to bite
someone else later. At the moment it is not directly callable as an SQL
function, but could/should be now that the "line" type is visible to
users.
Anyway, no need really to "send proper patches"; from here on how about
sending me patches based on the last file (or patch) you sent? Use "diff
-c" to generate the patch...
We should settle on an external representation for the "line" type;
although a point/slope representation is nice and intuitive, I'd suggest
trying the Ax+By+C=0 representation (used internally too) since we can
then avoid having representation problems with vertical lines (which
have infinite slope). Another possibility is to use a line segment
representation (two points) but we might have to be careful about
precision and rounding issues.
Once things settle down a bit I'll integrate the whole thing into the
source tree; there are several other files which have been touched in
the catalog and elsewhere and we'll need to move those patches to the
current source tree and test them before committing.
- Tom