Tom, When you say it happens all the time, do you mean every call has messed up (non numeric) results or is this just on particular uses? (e.g. getting a particular message rather than total failure)
Depending on which situation, I can give you some test harness code to try and track this down. It's not reproducible for me (I'm on Ubuntu rather than netbsd) but if it is for you, this may be a way forward. Jim On 12/09/2016 21:42, Arnt Gulbrandsen wrote: > Tom Ivar Helbekkmo writes: >> Actually, I guess that's pretty good -- it indicates that this isn't an >> optimization problem after all, right? > > I also have a debian box where it doesn't occur and one where it > occurs depending on compiler switches. Now I know it happens on netbsd > always. So it might depend on the gcc vagaries and object alignment... > something. > > The code pattern where it happens is one of the most common in all of > the aox source. Variants of > > Estring d = "a" + fn(42) + "c"; > > are used hundreds of times. > > Arnt >
