[RFC] libstdc++/37475

[Paolo Carlini <paolo dot carlini at oracle dot com>]

Hi all, hi Bo,

I'd like to ask the opinion of list - Bo in particular, our std::codecvt 
expert - about this PR, which is about a codecvt::do_in / do _out corner 
case: when the functions are called with from_end > from but to_end == 
to, thus no space at all provided to store the conversion. Currently we 
return "ok" in this case (we do nothing, just skip completely the 
conversion loop), where submitter expects "partial" because of Table 53 
of the Standard which describes "partial - not all source characters 
converted".

Frankly, I don't have a strong opinion, the current behavior rather 
makes sense to me because in this specific case *no conversion is 
attempted at all", but I know there are quite a few tricky issues having 
to do with codecvt and our current implementation is a delicate balance, 
result of our interpretation of a few DRs and the constraints due to the 
underlying use of C library functions...

Opinions?

Thanks,
Paolo.