[PATCH] Fix initialization of _M_grouping

[Benjamin Kosnik]

> I see. This currently fails for me.

Ditto. I'm not quite sure how it's supposed to work, and if this is an 
accurate representation of what Nathan is talking about. I'm hoping he 
can yay or nay on it: Nathan? 

It is a pretty interesting testcase to play around with, even though this 
is really a side issue to the whole grouping/sep issue.

I suspect the core that happens is when __sep == '0', not '\0'. Both 
cases should be fixed, but again, I'm not in a position to work on this 
right now.

> This is really, I believe, a related but
> different issue: making sure that everything works ok even for '\0' as the
> thousands separator.

Yes. From my limited time last night, it looks like std::string is all 
right but that other things go wrong where the following '0' after the 
'\0' gets eaten, so that l == 12345 and l != 1023045. This is a bug I think.

> Please keep me updated: if you want me to help on auditing the library for this
> or whatever.

It would be great if you would figure out what is going on in this testcase.

thanks,
benjamin