xmltester failures under MingW 32 bit

[robe]

strk,

Attached is the log file of my failures on geos 3.3.0 rc1 compilation of tar ball.

[strk]

Failure for #350 is indeed the originally reported faiure. This is annoying. Are you sure you're building with -ffloat-store ? Could you retry that with r3338 or r3335 ? Or generally bisect to see if you got any success for it ?

As per the #398 failure it does look like a minor coordinate drift. We're checking for exact equality (pointwise) which is the reason for the failure. Note that also JTS checks for pointwise equality, which is the reason why I've tried to track as closely as possible JTS as per operations precision, getting back from long doubles to single doubles...

[robe]

nope not sure -- do I pass with --ffloat-store in the configure step?

[strk]

It'll be automatic if you're using automake and your compiler supports it. Check with:

grep AM_CXXFLAGS Makefile

[robe]

Looks like it does

AM_CXXFLAGS = -DGEOS_INLINE -pedantic -Wall -ansi -Wno-long-long -ffloat-store

[robe]

With geos 3.3.0rc2 i still get 4 failures. I'll attach my new log in case it's different.

[strk]

Could you also check:

grep CXXFLAGS Makefile

I want to make sure -O2 is used. Also, which compiler version ? Both failing tests are very sensible to numeric robustness issues...

[robe]

strk output is:

AM_CXXFLAGS =  -DGEOS_INLINE  -pedantic -Wall -ansi -Wno-long-long  -ffloat-store
CXXFLAGS = -g -O2

[robe]

Just for extra information:

I'm running: gcc version 3.4.5 (mingw-vista special r3

I'll try a newer one once I remember how I toggled back and forth between the 2 gccs before.

[strk]

unless someone commits to fix this there's no point in targetting a real milestone.