[postgis-devel] Postgis 2.4.6 regession tests fails on Solaris

[Greg Troxel]

"Darafei \"Komяpa\" Praliaskouski" <me at komzpa.net> writes:

> Given the rarity of architecture I suggest you to rebuild with PostGIS 2.5
> - we've fixed many things for supporting ARM/MIPS and may happen that your
> sparcv9 will just catch up. We didn't have any SPARC for testing and didn't
> know of any install on SPARC until today. Please send the patches if you'll
> figure them out.

I used to test on sparc.  I will again someday; I have both a sparc64
and a 32-bit sparc.

I think it's important to keep an eye on odd platforms.  Often bugs that
show up there are real bugs that happen not to express on amd64.   But I
certainly understand about limited resources.

I have seen some failures on NetBSD/arm (earmv7hf-el specifically) with
gcc 5, I think in geos, but so far I am suspicious of the compiler.

I would suggest that the OP run 'paranoia' on their system to check the
floating point support, and also build with -O0 as there could be
compiler bugs.

Another thing is to try clang instead of gcc, or a sun compiler.

You could also try to run various sanitizers under clang.  It's possible
postgis has code that invokes undefined behavior which happens to be ok
on intel (or it would have been fixed :-).

sparc uses IEEE754 so that should not be an issue (vs vax which has
floating point that is different).  But, note the part in the mips bug
about different valid representations of nan and hex equivalence checks
not being sound.


The other thing to do is to begin to trace the troublesome functions
with gdb.