On Fri, 2013-01-25 at 14:06:32 -0800, Jonathan Nieder wrote: > Guillem Jover wrote: > > Some packages might be using -Werror, adding -Wall might cause FTBFS, > > please see the recently created FAQ [0] for the procedure to follow to > > add a new flag to the default set. > > When working with upstream projects that don't use -Wall, adding it > can create a lot of noise and make it hard to notice important > warnings. So I'm not convinced it's a good idea. A few well selected > warnings such as -Wimplicit-declaration could be more useful
Yeah, having thought about it, I do agree with that, more so because -Wall includes some warnings that are just for coding-style. I think we should only ever add warnings that point to actual potential bugs, and in that case after a while, possibly after having fixed most of them, we could even try to go further and make the clear cases -Werror=flag. Some of the flags I think would be fine to add (after the procedure from the FAQ) could be -Warray-bounds, -Wimplicit-declaration, -Wvolatile-register-var, -Wreturn-type and -Wsequence-point. From those I could see eventually the possibility to turning all of them into errors except for -Wreturn-type (because parts of it trigger on just old-style, while others are real bugs) and -Wsequence-point (as it's stated it can produce false positives). > By the way, do you think tweaking policy to discourage -Werror (except > in cases of circular build-deps where mistakes are hard to recover > from, like binutils) would be feasible? Well, I'd happily second that. I've always thought blanket -Werror on release builds is in general a bad idea, while it can be a very useful tool during development. Thanks, Guillem -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
