Re: [PATCH] Fix %K for LTO

[Richard Biener <rguenther at suse dot de>]

On Tue, 25 Mar 2014, Jakub Jelinek wrote:

> On Tue, Mar 25, 2014 at 03:35:29PM +0100, Richard Biener wrote:
> > The following implements a special-case for LTO when doing %K
> > formatting.  LTO currently only keeps FUNCTION_DECL abstract
> > origins (for BLOCKs satisfying inlined_function_outer_scope_p)
> > which means the existing %K formatting usually bails out immediately.
> > This severely complicates the search for warnings like
> > 
> > /usr/include/bits/stdio2.h: In function âmainâ:
> > /usr/include/bits/stdio2.h:293:2: warning: call to â__fread_chk_warnâ 
> > declared with attribute warning: fread called with bigger size * nmemb 
> > than length of destination buffer
> >   return __fread_chk_warn (__ptr, __bos0 (__ptr), __size, __n, __stream);
> >   ^
> > 
> > because you only have an outermost function context to work on
> > and the inline stack can include functions from any translation
> > unit with LTO.  The following patch improves this to
> > 
> > In function 'fread',
> >     inlined from 'main' at t.c:7:10:
> > /usr/include/bits/stdio2.h:293:2: warning: call to '__fread_chk_warn' 
> > declared with attribute warning: fread called with bigger size * nmemb 
> > than length of destination buffer
> >   return __fread_chk_warn (__ptr, __bos0 (__ptr), __size, __n, __stream);
> >   ^
> > 
> > which matches non-LTO behavior.
> > 
> > Bootstrap/regtest pending on x86_64-unknown-linux-gnu, ok for trunk?
> 
> If it works, ok.  I'd of course prefer if it preserved the abstract origins
> properly, otherwise it can't emit proper debug info.  But for now it is fine
> I guess.

Yeah, but we'd ICE here because the abstract BLOCKs cannot be
partitioned, etc. properly.  The debug info "works" at least
for the purpose of handling inlines.

Richard.