ST_Line_Locate_Point inaccurate closest point

[strk]

After the robustness fix in r8107 there's a regress test failing in regress_lrs. The test shows that closest_point_on_segment and lw_dist2d_distancepoint disagree on what's the projection of a far-away point onto a single-segment line.

[nicklas]

I think you have a typo in r8107. pa→npoints is 1-based, not 0-based, right?

Then shouldn't it be instead?:

if ( t >= pa->npoints && 
  ( proj.x == end.x) && (proj.y == end.y) )

That would explain it since the answer from st_closestpoint is the same as the second last vertex, not the last.

/Nicklas

[strk]

That doesn't fix. It is the equality comparison that puzzles me:

( proj.x == end.x) && (proj.y == end.y)

End should be the last point and proj would be the return from closest_point_on_segment

[strk]

Oh.. wonder. =# SELECT 'line_locate_point_1', ST_line_locate_point('LINESTRING(709243.393033887 163969.752725768,708943.240904444 163974.593889146,708675.634380651 163981.832927298)', 'POINT(705780 15883)'); NOTICE: t:3, pa→npoints:3, end:(708943 163975)

Indeed "end" is _not_ the last point, but the second-last. It's actually the endpoint of the first segment. I guess I should be using "seg" rather than "t"

[strk]

To better explain: the issue is that 't' is not the segment of the closest point, unless the query point is _on_the_line_.

Fixed with r8112

Thanks for looking at this, Nicklas

[strk]

Better fix about the seg vs. npoints with r8115 (we actually had to subtract 2, not 1, from npoints)