Date: Fri, 7 Jan 2011 08:48:58 +0100 From: Roman Divacky <rdivacky@freebsd.org> To: Erik Cederstrand <erik@cederstrand.dk> Cc: Tijl Coosemans <tijl@coosemans.org>, freebsd-current@freebsd.org Subject: Re: FYI: clang static analyzer page has moved to http://scan.freebsd.your.org/freebsd-head/ Message-ID: <20110107074858.GA87281@freebsd.org> In-Reply-To: <1EF28C3A-954D-4A4B-8069-F0E767EF3669@cederstrand.dk> References: <20110105131439.GN23329@acme.spoerlein.net> <20110105193653.GA49285@stack.nl> <7FA66A47-CB15-4C22-8614-B58E986CFFA4@cederstrand.dk> <201101062056.55807.tijl@coosemans.org> <1EF28C3A-954D-4A4B-8069-F0E767EF3669@cederstrand.dk>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, Jan 06, 2011 at 11:59:07PM +0100, Erik Cederstrand wrote: > > Den 06/01/2011 kl. 20.56 skrev Tijl Coosemans: > > > On Thursday 06 January 2011 09:01:09 Erik Cederstrand wrote: > >> Den 05/01/2011 kl. 20.36 skrev Jilles Tjoelker: > >>> On Wed, Jan 05, 2011 at 05:55:45PM +0100, Ulrich Sp?rlein wrote: > >>>> - get IPA to work with clang, or at least file a bug > >>>> - mark functions as __dead2 (please don't do that) > >>> > >>> Why not? > >> > >> Because the analyzer is supposed to find bugs. Only the function that > >> really doesn't return should be marked as such. If we begin spewing > >> __dead2's everywhere, it's bound to silence a valid bug somewhere > >> down the line when e.g. a conditional in a print_help() function is > >> changed subtly so it doesn't always reach exit(). > > > > On the other hand you can't really expect the compiler/analyser to know > > what a procedure in another file does, so in that case you need __dead2 > > anyway. [...] > > I have high expectations of LLVM :-) LLVM already has some knowledge of what's going on in other files (see LTO) so why shouldn't it be able to detect the __noreturn__ ? All the necessary information should be readily available. the static analyzer has nothing to do with LLVM. it's a clang component and uses only the info that clang knows. and clang (ie. the C frontend) does not perform any analysis of this kind, thus it does not know that stuff is dead. fwiw - my trivial (but working) patch brought down the number of reports by mere 5% so the bulk is probably somewhere else...
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20110107074858.GA87281>