Re: [PATCH] Fixes compile warning with -Wimplicit-fallthrough CFLAGS

From: Jeff King <peff@peff.net>
Date: Jan 23, 2018, 18:37 UTC
Message-ID: <20180123183731.GA18333@sigill.intra.peff.net>
In-Reply-To: <xmqqd120ifvu.fsf@gitster.mtv.corp.google.com>

On Tue, Jan 23, 2018 at 10:33:57AM -0800, Junio C Hamano wrote:

Show 19 quoted lines

> Jeff King <peff@peff.net> writes:
> 
> >> diff --git a/apply.c b/apply.c
> >> index 321a9fa68..a22fb2881 100644
> >> --- a/apply.c
> >> +++ b/apply.c
> >> @@ -1450,7 +1450,7 @@ static void recount_diff(const char *line, int size, struct fragment *fragment)
> >>  		switch (*line) {
> >>  		case ' ': case '\n':
> >>  			newlines++;
> >> -			/* fall through */
> >> +			GIT_FALLTHROUGH;
> >
> > Ugh, the semi-colon there makes it look like it's actual code. If we go
> > this route, I wonder if it's worth hiding it inside the macro.
> 
> What?  You mean to shout in all caps without even terminating the
> line with any punctuation?  Please don't---I am sure it will break
> auto indentation people rely on from their editors.

True, that may be even worse. I just wonder if we can do something to make it look more obviously like a non-code attribute. The actual syntax is something like:

  [[fallthrough]];

which is pretty horrid, but at least a bit easier to see. gcc also provides "__attribute__((fallthrough))", but I don't think it works with clang.

I vastly prefer the comment approach if we can use it. Apparently clang doesn't support it, but I have also not managed to get clang (either version 4, 6, or the upcoming 7) to actually report anything via -Wimplicit-fallthrough, either. Maybe I'm holding it wrong.

-Peff