[Patch] Fix libstdc++/13582

[Paolo Carlini]

Whee!

Basically, I have implemented Pétur's notes present in the audit
trail of 12868: we can exploit the external buffer in order to imbue
'on the fly' a new locale without trying to seek back (not a good
idea for pipes and ttys ;)

I think that the final result is sufficiently clean: just an additional
|| in underflow to teach it that since we have just imbued a new locale
we should first try to simply convert the remainder and delay low level
reads (this case is characterized by !_M_reading and __remainder > 0
whereas, usually, !_M_reading implies __remainder == 0: at the beginning
of a series of gets we cannot have a remainder)

Tested x86-linux. Will wait the usual 24 hours or so.

Paolo.

P.S. Well, I would like to remark that apparently our strong competitor
on x86 does bad on most of our recent QoI testcases for filebuf::imbue :)

///////////////
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: CL_13582
URL: <http://gcc.gnu.org/pipermail/libstdc++/attachments/20040109/bf40ee4b/attachment.ksh>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: patch_13582
URL: <http://gcc.gnu.org/pipermail/libstdc++/attachments/20040109/bf40ee4b/attachment-0001.ksh>