[Patch/RFC] Fix libstdc++/9875

[Paolo Carlini]

Nathan Myers wrote:

>In general seekpos should not depend on seekoff, just because seekpos 
>is supposed to work even where seekoff doesn't.
>
This is interesting and, honestly, was not at all completely clear to 
me. Thanks.
Notice, however, that we _always_ did so (before my patch, I mean)

>  Nevertheless this is 
>certainly better than before.  Let's use this, and revisit it when
>we get variable-length encodings working (next week, right? ;-)
>
Right ;)

I'll wait a little more for further comments, anyway...

Paolo.