On Sat, Apr 16, 2022 at 07:18:47AM -0600, Daniel Dickman wrote: > > > > On Apr 14, 2022, at 11:49 AM, Stuart Henderson <[email protected]> wrote: > > > > My thinking is that, if the code has behaviour which is considered > > undefined by the C standard assumed by the compiler, no level of > > optimization is safe. Maybe now you get lucky and -O works (on whichever > > architecture you've tested) but I don't think it's reasonable to assume > > that this is the case everywhere, or will be the case following compiler > > updates. > > I haven't looked very deeply at epic but if the note is referring to > strict aliasing then I would follow the advice about sticking to -O. > > John Regehr wrote up a nice piece on this a few years ago: > > https://blog.regehr.org/archives/1307
We have a decade or even more when we didn't hear any random crashes reports with '-O', and FreeBSD has it as a default flag. Epic developer takes crash reports seriously, when I had "openbsd only crash" he helped me with it. Also, he wasn't against adding unveil and pledge. Anyway, if there is a proposition not to take any risk - probably it worth it, performance isn't critical thing for an IRC client.
