Re: [PATCH v2] Re: perf annotate segfaults when source code has goto label that looks like hex number

[Arnaldo Carvalho de Melo <acme@infradead.org>]

Em Mon, Aug 02, 2010 at 11:02:14AM +0200, Peter Zijlstra escreveu:
> On Thu, 2010-07-22 at 16:11 -0300, Arnaldo Carvalho de Melo wrote:
> > > > diff --git a/tools/perf/util/hist.c b/tools/perf/util/hist.c
> > > > @@ -976,13 +976,15 @@ static int hist_entry__parse_objdump_line(struct hist_entry *self, FILE *file,
> > > >     if (line_ip != -1) {
> > > >             u64 start = map__rip_2objdump(self->ms.map, sym->start);
> > > >             offset = line_ip - start;
> > > > +           if (offset < 0 || (u64)line_ip > sym->end)
> > > > +                   offset = -1;
> > 
> > > This part is good idea anyway. Even if label will be interpreted as ip
> > > perf at least will not crash. It may miss-report something if check will
> > > accidentally succeed though.
> > 
> > Yeah, we can possibly find a label which is a valid hex number and that
> > falls inside the address range, but with what we have in objdump this
> > seems to be the best we can have, I'll commit this.
> 
> Wouldn't it be better to re-write perf-annotate to not have wild monkey
> sex with objdump and instead 'borrow' some of the objdump code to
> generate the output ourselves? That way we don't rely on the output
> syntax at all..

Right, the problem is that doing that seems like avoiding regular,
mostly satisfying sex with a known partner to having a month long orgy
blindfold.  You can try, I'll pass for now.

- Arnaldo