Smarter linear location (for ST_Line_Interpolate_Point, ST_Line_Substring, ST_Line_Locate_Point)

[strk]

ST_LocatePoint returns a value which you can't easily convert to a vertex number. This isn't helpful if what you're looking for is a substring formed by pre-existing vertices rather than formed by arbitrarily interpolated points.

In turn, re-using existing vertices is required to avoid introducing coordinates drifts when willing to analize subportions of lines.

GEOS/JTS, for comparison, uses a pair of (SEG_NUM/SEG_FRACTION) representing in turn the segment index and the fraction of next segment.

[strk]

#1314 was another version of this.

As of r8212 there's some support for specifying a snap-to-vertex tolerance in the low-level ptarray_substring function. That tolerance was used in turn to solve the problem of using a pre-existent vertex as a location.

Still, it'd be nice to make the whole set of functions more robust numerically

[strk]

r9063 referenced this ticket by accident, #982 was the correct one for it.