On 6/1/25 05:11, Mr wrote:
> Sorry I reread the thread more carefully and saw there WAS objection, though I
> did not understand it! it looks bad, but does your answer suggest another
> workaround? something as easy to apply by the profane as such a patch?

Don't use -ffast-math.

If you do use -ffast-math, you must use -fno-finite-math-only.

---

I've not looked at source code fixes because I don't know how to easily 
determine where the exposures in the code are due these g++13 changes!

Only some of which appeared to be backed in later compiler version last 
I 'glanced', rather than completely. The partial retrenchment 'might' 
explain some (or all) the different fail signatures seen.

Fixing the source code for this likely requires a very deep dive into 
what all the g++13 compiler changes were, what's been backed out or not 
etc... Then probably some custom source scanning code to highlight 
exposures - if new compiler warnings haven't been introduced which do 
this work for us.

The work is way down on my todo list because -fno-finite-math-only is 
working for me.

Bill P.

Post a reply to this message