filter unwated cflags? (-ffast-math)

[Alexander Neundorf]

On Wednesday 17 April 2013, Michael Pyne wrote:
> On Tue, April 16, 2013 20:29:25 Albert Astals Cid wrote:
> > El Dimarts, 16 d'abril de 2013, a les 16:39:32, Bernd Buschinski va
> 
> escriure:
> > > hi,
> > 
> > Hi
> > 
> > > I am not sure how do deal with unwanted cflags(cxxflags), so I ask here
> > > :)
> > > 
> > > For kjs for example -ffast-math is known to cause problems, as we
> > > stricly rely on IEEE compatible floating point, NaN, Inf, signed Zero
> > > and all the stuff.
> > > 
> > > So the question should I filter them? or leave it to the user to
> > > optimize(break) his system?
> > 
> > Well, if you know it'll break, you should shout. If that shouting means
> > just a huge message or stopping the compilation i am not sure, but if
> > it'll give you extra work for no reason in forme of bugs, etc. i'd go
> > for stopping the compilation.
> 
> Having run into this before, I can only +1 this about 1000 times or so. KJS
> with -ffast-math and friends is unsupportable so there's no reason to let
> the compile go on if we can detect it. That code should be considered
> completely miscompiled, if a user wants to use the flag anyways they can
> take it on themselves to figure out CMake in addition.

+1 

Alex