Re: Testing the return value of fclose() in the backend

On Fri, 30 May 2003, Tom Lane wrote:

> Bruce Momjian <pgman(at)candle(dot)pha(dot)pa(dot)us> writes:
> > Gavin Sherry wrote:
> >> There are various places in the backend, such as FreeFile(), where the
> >> return value of fclose() is not tested.
>
> > We are not checking fclose, probably because fclose failures are quite
> > rare. Should we be concerned?
>
> Probably. Closing a valid file descriptor in itself can't provoke any
> error that I can imagine, but fclose() also implies fflush() --- so if
> you have written data that hasn't yet been forced out of the stdio
> buffers then out-of-disk-space is certainly a foreseeable failure.

Yes. I think I brought that up in my original email. Heap access/WAL
routines 'should not' suffer an fclose() problem because of
fsync() calls. But this isn't necessarily the case for COPY.

>
> fclose failure on an open-for-read-only file seems like Assert()
> material; it "can't happen".

Right. If this generates an error, there are probably more serious issues.

Gavin