Re: [v3] testsuite/ext cleanups

[Zack Weinberg <zack at codesourcery dot com>]

"Aaron W. LaFramboise" <aaronavay62@aaronwl.com> writes:

> However, as reiserfs has a similar problem, and no doubt future
> filesystems and implemenations may have similar problems, it seems like
> a good idea to NOT rely on the behavior that un-openable files (whether
> they're directories, or something else) will be discarded and the search
> will continue.  I think its a good thing for GCC to have as few
> assumptions about the nature of the filesystem as possible.  That means
> the build process should eliminate cases where this sort of confusion
> might occur.

As a dyed-in-the-wool Unix weenie I am uninterested in catering to
broken file systems (whether they are on non-Unix operating systems,
as NTFS, or on Unix, as reiserfs).

Also, I've never particularly liked the feature that an un-openable
file/directory/whatever with exactly the name requested in #include
would be silently skipped if the failure matched some set of
"acceptable" error codes.  I think we should always error out in that
case.  (This would make the issue of broken file systems moot, but
would also break people's builds - but, as shown by this problem, not
in ways that they weren't already broken, in the presence of a broken
file system.)

zw