Context Navigation

← Previous Ticket
Next Ticket →

#865 closed defect (fixed)

GEOSClipByRect failure

Reported by:	komzpa	Owned by:	geos-devel@…
Priority:	major	Milestone:	3.6.3
Component:	Default	Version:	3.6.2
Severity:	Unassigned	Keywords:
Cc:	Algunenano

Description

Failing test PR for GEOS: https://github.com/OSGeo/geos/pull/96/files

Extended testcase in PostGIS: https://trac.osgeo.org/postgis/ticket/4038

Change History (8)

comment:1 by dbaston, 6 years ago

Resolution:	→ worksforme
Status:	new → closed
Summary:	ClipByBox2D failure → GEOSClipByRect failure

Cannot reproduce the failure. Maybe the suggested infinite loop was in the PostGIS code somewhere?

comment:2 by Algunenano, 6 years ago

Trying to analyze the endless loop. Running the query from Postgis reaches:

#1  0x00007ff35817a014 in geos::operation::intersection::distance (rect=..., ring=std::vector of length 2, capacity 2 = {...})
    at RectangleIntersectionBuilder.cpp:247

With the following values:

(gdb) p ring
$21 = std::vector of length 2, capacity 2 = {{static _nullCoord = {static _nullCoord = <same as static member of an already seen type>, 
      x = nan(0x8000000000000), y = nan(0x8000000000000), z = nan(0x8000000000000)}, x = 3.0481343214686657e-14, y = -1, 
    z = nan(0x8000000000000)}, {static _nullCoord = {static _nullCoord = <same as static member of an already seen type>, 
      x = nan(0x8000000000000), y = nan(0x8000000000000), z = nan(0x8000000000000)}, x = 3.0481259877257472e-14, y = -1, 
    z = nan(0x8000000000000)}}
(gdb) p ring.size()
$22 = 2
(gdb) p ring[1].x
$24 = 3.0481259877257472e-14
(gdb) p ring[1].y
$25 = -1
(gdb) p ring[0].x
$26 = 3.0481343214686657e-14
(gdb) p ring[0].y
$27 = -1

And the rectangle:

(gdb) p rect
$28 = (const geos::operation::intersection::Rectangle &) @0x7ffe21add4f8: {xMin = 3.0481343214686657e-14, yMin = -20000000, 
  xMax = 20000000, yMax = -1}

This then calls:

geos::operation::intersection::distance (rect=..., x1=3.0481259877257472e-14, y1=-20000000, x2=3.0481343214686657e-14, y2=-1)
    at RectangleIntersectionBuilder.cpp:187

The setup:

(gdb) p rect.position(3.0481259877257472e-14, -20000000)
$30 = geos::operation::intersection::Rectangle::Outside
(gdb) p rect.position(3.0481343214686657e-14, -1)
$31 = geos::operation::intersection::Rectangle::TopLeft

Which means that pos & endpos will always be zero, thus never reaching the break statement and looping until the end of time.

I don't know GEOS codebase to add a test based on this, so I decided to follow the same idea as Kopmza and create a test directly with GEOSClipByRect. The issue is that using geometry::toText() is removing some significant bytes and the test doesn't fail. To avoid this I patched the function to always print 30 digits and took the output: Its here (https://gist.github.com/Algunenano/045dfc55f2b530df0fed12beb2d4db23) and its pretty long.

With that test my installation enters the infinite loop. Setup:

$ uname -a
Linux Mordor 4.17.8-1-ARCH #1 SMP PREEMPT Wed Jul 18 09:56:24 UTC 2018 x86_64 GNU/Linux
$ clang -v
clang version 6.0.1 (tags/RELEASE_601/final)
Target: x86_64-pc-linux-gnu
Thread model: posix
InstalledDir: /usr/bin
Found candidate GCC installation: /usr/bin/../lib/gcc/x86_64-pc-linux-gnu/8.1.1
Found candidate GCC installation: /usr/bin/../lib64/gcc/x86_64-pc-linux-gnu/8.1.1
Found candidate GCC installation: /usr/lib/gcc/x86_64-pc-linux-gnu/8.1.1
Found candidate GCC installation: /usr/lib64/gcc/x86_64-pc-linux-gnu/8.1.1
Selected GCC installation: /usr/bin/../lib64/gcc/x86_64-pc-linux-gnu/8.1.1
Candidate multilib: .;@m64
Candidate multilib: 32;@m32
Selected multilib: .;@m64
$ ./configure --prefix=/usr CC=clang CXX=clang++ CPPFLAGS="-D_FORTIFY_SOURCE=2" CFLAGS="-pipe -fno-plt -Wall -Wextra -g3 -gdwarf-4 -fno-omit-frame-pointer" CXXFLAGS="-pipe -fno-plt -Wall -Wextra -g3 -gdwarf-4 -fno-omit-frame-pointer" LDFLAGS="-Wl,-O1,--sort-common,--as-needed,-z,relro,-z,now"

I've also tested with GCC 8.1.1 (same flags, only changing clang -> gcc, clang++ -> g++) and it also happens. Before I was trying with optimization flags (-march=native -mtune=native -O3 -Wall -Wextra -g3 -gdwarf-4 -fno-omit-frame-pointer) and it was also happening, but the inlining made it pretty hard to debug the issue.

Finally, PR with the test: https://github.com/libgeos/geos/pull/109