Messages in this thread | | | Date | Tue, 10 Dec 2013 13:07:07 +0100 | From | Ingo Molnar <> | Subject | Re: [GIT PULL 00/21] perf/core improvements and fixes |
| |
* Arnaldo Carvalho de Melo <acme@ghostprotocols.net> wrote:
> Em Tue, Dec 10, 2013 at 12:47:57PM +0100, Ingo Molnar escreveu: > > * Arnaldo Carvalho de Melo <acme@ghostprotocols.net> wrote: > > > Em Tue, Dec 10, 2013 at 12:12:29PM +0100, Ingo Molnar escreveu: > > > > Hm, I've unpulled it because 'perf top' crashes on exit, in > > > > dso__delete(): > > > > 495 if (dso->sname_alloc) > > > 496 free((char *)dso->short_name) > > > Btw., instead of trusting flags I'd argue that using the pointer as a > > flag and clearing the pointer too is a much more robust freeing > > pattern in general: > > > if (dso->short_name) { > > free(dso->short_name); > > dso->short_name = NULL; > > } > > > > or so ... > > This is not an unusual idiom, if you look at > tools/perf/util/ev{list,sel}.c, for instance, you'll see it in many > destructors. > > In this case there is a micro optimization where sometimes the > shortname is just a pointer to the tail part of the long name, hence > the flag.
Sounds fair.
[btw., a tiny nit remains: the cast is probably unnecessary, free() will take any pointer.]
Thanks,
Ingo
| |