Re: portability of rewind

[Bruno Haible]

Eric Blake wrote:
> m4 had been using rewind to swap from appending to a
> temporary file representing a large diversion, to reading from offset 0 to
> copy that diversion to a different location.  But if the implicit fflush
> performed by the rewind fails, there is no way for the application to know
> about it, since rewind must also clear the error bit on the stream.  Which
> means that rewind can cause undetected data loss.

Apparently, yes.

> And which is why I'm
> checking in a patch to m4 today to avoid all use of rewind.
> ...
> Should we update stdio_.h to give a link warning that recommends avoiding
> all use of rewind?

This sounds exaggerated. You can recommend (for output streams and update
streams) to call  fflush (fp)  before  rewind (fp) , no?

OTOH, the code that avoids rewind is also not particularly complicated:

  if (fseeko (fp, 0, SEEK_SET) < 0) ...error-handling...
  clearerr (fp);

Bruno