This is the mail archive of the
libstdc++@gcc.gnu.org
mailing list for the libstdc++ project.
Re: ext/pb_ds/regression/trie_no_data_map_rand
- From: Paolo Carlini <paolo dot carlini at oracle dot com>
- To: Daniel Berlin <dberlin at dberlin dot org>, libstdc++ <libstdc++ at gcc dot gnu dot org>, bkoz at redhat dot com
- Date: Tue, 15 Jul 2008 11:48:59 -0500 (CDT)
- Subject: Re: ext/pb_ds/regression/trie_no_data_map_rand
- Accept-language: en-US
- References: <4aca3dc20807150939w55f9716avb00770e309332aab@mail.gmail.com>
Hi,
I suspect you meant to CC another Paolo... ;)
Anyway:
> I was trying to track down what PRE is doing to these testcases
> that causes them to crash, and I noticed that while I can't get
> them to crash without turning on PRE (IE -fno-tree-pre "fixes" them),
> they seem to have a *massive* number of valgrind failures consisting
> of "invalid reads" even with PRE turned off.
I don't think those failures are normal, but Benjamin knows that code *much* better than me. While we are at it, I think those testcases take far too much to compile and should be split to simpler units (probably also easier to debug, then ... ;)
If, while waiting for Benjamin feedback, you could debug a bit more, it would be great, because, all-in-all, the chances that only the new PRE triggers undefined behavior seem pretty small to me, just a guess of course.
Paolo.