On Thu, Aug 22, 2019 at 04:30:34PM +0300, Dan Carpenter wrote:
> On Thu, Aug 15, 2019 at 12:12:37PM +0100, John Levon wrote:
> >
> > I took another pass, and this works for me:
> >
> > diff --git a/parse.c b/parse.c
> > index ca4726b8..44c59707 100644
> > --- a/parse.c
> > +++ b/parse.c
> > @@ -2859,6 +2859,7 @@ struct token *external_declaration(struct token
> > *token, struct symbol_list **lis
> > /* Parse declaration-specifiers, if any */
> > token = declaration_specifiers(token, &ctx);
> > mod = storage_modifiers(&ctx);
> > + mod |= ctx.ctype.modifiers & MOD_NORETURN;
> > decl = alloc_symbol(token->pos, SYM_NODE);
> > /* Just a type declaration? */
> > if (match_op(token, ';')) {
> >
> > (I suppose a proper fix would collate all function-level attributes
> > but...)
>
> This patch seems like a hack, but I will apply it...
I agree, it would be nice to see this cleaned up properly, but as the
above is sufficient for us, it'll do for us! Thanks!
> > However, it seems like smatch is still not quite passing its knowledge
> > along: if I have:
> >
> > extern void die() __attribute((__noreturn__));
> >
> > void mydie()
> > {
> > die();
> > }
> >
> > then the die() call is nullified, but smatch doesn't realise that means
> > all paths of mydie() are __noreturn__ too.
>
> The code to handle that is really ancient. You need to do:
> ./smatch --info test.c | tee warns.txt
> grep no_return_funcs warns.txt || echo FAIL
> ./smatch_scripts/gen_no_return_funcs.sh warns.txt -p=levon
> mv levon.no_return_funcs smatch_data/
>
> Then pass -p=levon to smatch on the next run.
OK, thanks. That doesn't match too well with the way we use smatch, but
this is much less of a big deal (I only needed to annotate this in a
couple of places).
regards
john
