Re: [Patch] Fix libstdc++/11378 (take2) + xsputn optmizations

[Paolo Carlini <pcarlini at unitus dot it>]

Andreas Jaeger wrote:

> > However, Andreas, please correct me as soon as possible if I'm wrong,
> > mixing fseek and read and write (vs fread, fwrite) is not possible. We
> >    
> You can only do this if you take care to sync streams and file
> descriptors with e.g. fclean on GNU systems.
Interesting... I didn't know about this... But the issue seems a little 
tricky. Currently we are calling fflush instead of fclean, which seems 
more general, in seekoff. Last time I tried (in the old non unified 
scheme, however) definitely mixing read, write and fseek didn't work...

> > want non blocking read and write as our low level I/O primitives. Do
> > you expect troubles with lseek64?
> >    
> You can use lseek64 and off64_t, there should be no problem (if lseek
> is part of the C library).
Good! Indeed, the type which goes with lseek64 is off64_t, according to 
the official docs too.

Paolo.