[postgis-users] Unable to complete checks on Solaris x64

[Sandro Santilli]

On Mon, Jul 22, 2013 at 10:44:52PM +0200, Dechaux Eric CEN (BCQ STIG) wrote:

> However, the make check fails with two kind of errors during the regress tests :
> 
>  PostgreSQL 9.2.4 on i386-pc-solaris2.10, compiled by gcc (GCC) 3.4.3 (csl-sol210-3_4-branch+sol_rpath), 64-bit
>  Postgis 2.0.3 - r11128 - 2013-07-22 18:47:57
>    GEOS: 3.3.8-CAPI-1.7.8
>    PROJ: Rel. 4.8.0, 6 March 2012
> 

[..]

> Here is the diff for the first failure :
> 
> --- lwgeom_regress_expected     Mon Feb 27 23:08:12 2012
> +++ /tmp/pgis_reg/test_27_out   Mon Jul 22 20:49:06 2013
> @@ -1,5 +1,5 @@
>  BOX(0 0.1,11 12)
> -BOX3D(0 0.1 -55,11 12 12)
> +BOX3D(0 0.1 -0.04,11 12 12)
>  11184
>  15824
>  20464

This one could be big-endian vs. little-endian reading of 3D extent info.
Could you please file a ticket for this alone ?

> All other failures, on the contrary, crashed the postgresql process. Here is what it shown in the postgresql log file for the tickets checks :

[..]

> -bash-3.2$ pstack core
> core 'core' of 26050:   /opt/stig/idicsi-2.0/bin/amd64/postgres -D /tmp/postgis-2.0.313818
>  00000000001b6cad ???????? ()
>  fffffd7ffed11f09 _SUNW_Unwind_RaiseException () + 46
>  fffffd7ffb7dc79e __cxa_throw () + 5e
>  fffffd7ffb923ea1 _ZN4geos4geom10LinearRing20validateConstructionEv () + 161
>  fffffd7ffb9247ea _ZN4geos4geom10LinearRingC1EPNS0_18CoordinateSequenceEPKNS0_15GeometryFactoryE () + 6a
>  fffffd7ffb920115 _ZNK4geos4geom15GeometryFactory16createLinearRingEPNS0_18CoordinateSequenceE () + 35
>  fffffd7ffe6861d2 GEOSGeom_createLinearRing_r () + 22
>  fffffd7ffe7338a3 LWGEOM2GEOS () + 1b3
>  fffffd7ffe734fb8 lwgeom_make_valid () + 48
>  fffffd7ffe6f2c68 ST_MakeValid () + 68

[..]

> If I understand well, the crash occurs inside GEOS, in the LinearRing constructor, inside the validateConstruction method.
> 
> Has somebody seen this king of crash ? Can it comes from my GEOS build ?

I think this kind of issue has been seen many times on windows,
where it would crash everytime there's an exception thrown by the
GEOS library. Something about exceptions crossing DLL boundary.
Did you build GEOS yourself ? Does the testcase there ends successfully ?
Also, doe "make check" under postgis/liblwgeom/cunit work for you ?

--strk;