On Tue, Sep 27, 2016 at 05:04:23PM +0200, Jakub Jelinek wrote:
> On Tue, Sep 27, 2016 at 04:54:28PM +0200, Marek Polacek wrote:
> > On Tue, Sep 27, 2016 at 10:48:50AM -0400, Jason Merrill wrote:
> > > On Tue, Sep 27, 2016 at 9:56 AM, Michael Matz <m...@suse.de> wrote:
> > > > On Tue, 27 Sep 2016, Jakub Jelinek wrote:
> > > >
> > > >> Just compare that to the number of real bugs the warning found in gcc
> > > >> codebase. It is really worth it for -Wextra.
> > > >
> > > > All those bugs would also have been found as well when it had simply
> > > > accepted
> > > > /fall.*thr/i
> > > > anywhere in the preceding comment on one line. But all the recent
> > > > spelling changes of comments to cater for the strictness exactly shows
> > > > how
> > > > misguided that is. The above would accept "Don't fall through" as well.
> > > > I say: so what?
> > >
> > > I agree.
> >
> > All right, I'm not opposed to making the comment parsing more benevolent.
> > We still should have enough time to fine-tune it.
>
> Perhaps we want -Wimplicit-fallthrough{,=1,=2,=3,=4}, where
> =1 would match indeed /fall.*thr/i (note, it will be really costly in this
> case, one will have to parse all comments in detail in the preprocessor,
> so I'd be against making it the default),
Perhaps we could use POSIX regcomp/regex functions; do you (or anyone else)
have an idea how expensive they are and if it's feasible to use them in the
preprocessor?
Marek
