[RFA:] ar test failure fix

[Alan Modra]

On Tue, Jun 07, 2005 at 08:15:34PM -0700, Mark Mitchell wrote:
> Perhaps, even, the caller is buggy?

Not really.  The sequence of events was

- bfd_error cleared

- lots of intervening code

- bfd_bread returning zero bytes, then a test on bfd_error to see
  whether the short read was due to a file error of some sort.

You happened to change some of that intervening code in a way that set
bfd_error when there was no error.  ar.c could be made more robust by
reducing the amount of intervening code, but I think it's OK as is.
I also think that it's no business of bfd_bread to clear bfd_error on
success.

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre