This is the mail archive of the
libstdc++@gcc.gnu.org
mailing list for the libstdc++ project.
Re: [v3] move locale cache to 3.3
- From: Benjamin Kosnik <bkoz at redhat dot com>
- To: Jerry Quinn <jlquinn at optonline dot net>
- Cc: libstdc++ at gcc dot gnu dot org
- Date: Mon, 3 Mar 2003 20:25:49 -0600
- Subject: Re: [v3] move locale cache to 3.3
- References: <E18pN4i-0003wY-00@tiamat>
>There were a couple of fixups to it, though. I had to export _S_atoms_in and
>_S_atoms_out and reinsert _S_atoms to get check-abi to work.
Yes I noticed this too. Sadly, there were other things as well. Note,
_S_atoms doesn't have to re-appear as in your patch. It's no longer
used, and a symbol is provided for compatibility.
>Also, there were two latent bugs in ios_base::_M_grow_words. First, we don't
>actually want to delete the _Words array if a new allocation fails.
>Otherwise, you could be calling callbacks that look for data that isn't there
>any longer. The spec only says that badbit should be set, and if we delete
>the array, later callbacks will still be expecting that data to be there.
>Second, if the number is too large, we need to throw a failure exception if
>the exception mask has badbit set.
Assuming this is still a problem on mainline....
Can you post a patch for mainline with just this part? Actually, can you
just take this part out right now and then it can be applied to mainline
and gcc-3_3-branch seperately.
>What do you think? The pword stuff is not quite as pretty, but it's still
>fast.
I think it will work very nicely! Thanks for figuring it out.
best,
benjamin