#615 closed defect (fixed)
ST_SymDifference triggers a crash in WKTWriter (?)
| Reported by: | strk | Owned by: | |
|---|---|---|---|
| Priority: | blocker | Milestone: | 3.3.7 |
| Component: | Default | Version: | 3.3.6 |
| Severity: | Unassigned | Keywords: | |
| Cc: |
Description
For some obscure reason computing the SymDifference between the geometries showed in here: http://trac.osgeo.org/postgis/ticket/2177
Results in a crash deep in WKTWriter (which should be unrelated).
There could be some bogus debug line around...
Change History (13)
comment:1 by , 10 years ago
comment:2 by , 10 years ago
It is curious to note that Union operation succeeds (but SymDifference fails)
comment:3 by , 10 years ago
The offending geometry (portion) is this one:
01030000000100000007000000773A98D22192304099FAA9431D424740765F7CC2209230406643FDA024424740773A98D22192304098FAA9431D424740773A98D22192304097FAA9431D424740773A98D22192304096FAA9431D4247405EA791E509923040D3DDAD7E59424740773A98D22192304099FAA9431D424740
It's a simple polygon. One ring of 7 points. Only, it self-intersects: Self-intersection[16.5708285924579 46.5165180759406] And UnaryUnion does NOT fix that
comment:6 by , 10 years ago
ST_Node can't correctly node it, so no way to polygonize. It's extent is (basically a point):
BOX(16.5704635124779 46.5165180759406,16.5708285924579 46.5183561657865)
Snapping its inputs to a grid of 1e-13 gives us something usable. 1e-14 isn't enough.
comment:7 by , 10 years ago
Looking at the log it's interesting to note that the loop isn't really infinite, but rather catches a self-intersection always at different places ! The X ordinate value of the reported intersection is always the same while the Y value seems to augment over time. Here's a short excerpt:
{{{{ CBR: removed-bits geom 1 is INVALID: Self-intersection at or near point 0.57082859245789663 0.51651807594703314 (0.57082859245789663305 0.51651807594703313953) CBR: removed-bits geom 1 is INVALID: Self-intersection at or near point 0.57082859245789663 0.51651807594704024 (0.57082859245789663305 0.51651807594704024496) ... CBR: removed-bits geom 1 is INVALID: Self-intersection at or near point 0.57082859245789663 0.51651807594735288 (0.57082859245789663305 0.51651807594735288376) ... ... CBR: removed-bits geom 1 is INVALID: Self-intersection at or near point 0.57082859245789663 0.51651807597061605 (0.57082859245789663305 0.51651807597061605293) }}}
Eventually, the stack limit is reached.
comment:8 by , 10 years ago
The previous comment wasn't formatted correctly. Here's another go:
(0.57082859245789663305 0.51651807594059562234) (0.57082859245789663305 0.5165180759406098332) (0.57082859245789663305 0.51651807594061693862) (0.57082859245789663305 0.51651807594062404405) (0.57082859245789663305 0.51651807594063114948) (0.57082859245789663305 0.51651807594063825491) (0.57082859245789663305 0.51651807594064536033) (0.57082859245789663305 0.51651807594065246576) (0.57082859245789663305 0.51651807594065957119) (0.57082859245789663305 0.51651807594066667662) ... (0.57082859245789663305 0.51651807597058052579) (0.57082859245789663305 0.51651807597058763122) (0.57082859245789663305 0.51651807597059473665) (0.57082859245789663305 0.51651807597060184207) (0.57082859245789663305 0.5165180759706089475) (0.57082859245789663305 0.51651807597061605293)
I've 4225 _distinct_ numbers in there, in my log (killed before dying)
comment:9 by , 10 years ago
The segfault happens at iteration 9033 here. The final invalidity I see on removed-bits geom is:
Self-intersection at or near point 0.57082859245789663 0.51651807597262689 (0.57082859245789663305 0.51651807597262688887)
it's still the same X and the highest Y
comment:10 by , 10 years ago
| Resolution: | → fixed |
|---|---|
| Status: | new → closed |
Alright I found a fix for this: running the UnaryUnion _only_ when the given geometry is a multicomponent _and_ the invalidity found is an intersection between components (or a "too few points").
I'm not 100% sure that there's no way to enter the infinite recursion with these kind of invalidities but for now this bug is fixed. The only influence this change had on existing tests was that the test for #488 (which was found to expect a bogus result) is now giving a TopologyException, like done by JTS itself. I think it's an acceptable change.
It's an infinite loop entered while trying to fix invalidities produced by CommonBitsRemoval. The code attempts a UnaryUnion to fix, but the UnaryUnion gets back to an invalidity
I'm surprised that UnaryUnion doesn't drop the "too few points" geometry component: