Vincent Lefevre <address@hidden> writes:
> Geoff Clare said in the austin-group-l list:
>
> | This one is debatable. Step 2c says that when "file" is a
> | If attempting to remove one of the names returned by readdir()
> | results in an ENOENT error, then it could be argued that the name
> | was not an "entry contained in file" at that point, and therefore
> | an implementation which ignores the error would still conform to
> | the requirements of step 2c.
>
> which was more or less my point of view.
That argument applies to ENOENT, but I don't see why it applies to
ESTALE. ESTALE could be caused by a lot of things other than file
removal, e.g., the remote file system was unexported. See
<http://nfs.sourceforge.net/> and search for ESTALE to see more
examples.