[PATCH 1/1] test/basic: replaced find -perm +111 with portable alternative

[Austin Clements]

Quoth Tomi Ollila on Jun 13 at 10:47 pm:
> On Mon, Jun 10 2013, Austin Clements <amdragon at MIT.EDU> wrote:
> 
> > LGTM.  Though, I wonder, why not *just* -perm -100?  That isn't quite
> > a correct test of whether the user can execute it: e.g., if the file
> > is owned by some other user and a group the current user isn't in,
> > then -perm -1 is the correct test, though unless the file has some
> > unusual permissions, -perm -100 is likely to pass anyway.  But the
> > test you have (and the test that was there before) isn't quite correct
> > either: if the file is owned by the current user and has some crazy
> > permission like 0611, the user won't be able to execute it, even
> > though someone else could.
> 
> While giving considerable amount of thought for such an insignificant
> issue I came to realize this:
> 
> The purpose of the '-perm ...' part in that expression is not to check
> whether the file is executable by the user but just to reduce the set
> of files the whole expression returns without need to "blacklist" more
> files that are already blacklisted with '! -name ...' subexpressions
> ("Makefile", ".gitignore" and so on).
> 
> With +111, /ppp and their portable alternative
> ( -perm -100 -or -perm -10 -or -perm 1 ) the implicit reduction this
> part does is smaller than with -100.
> 
> The returned list is then compared with ${TESTS} and if there is no
> exact match then this particular test fails.
> 
> Whatever this test result is, the execution of any file in ${TESTS}
> will fail with "permission denied" if it is not executable by
> the user running the tests.
> 
> I think that as we're doing this "shortcut" instead of full file
> blacklisting, this should reduce the output less rather than
> more and therefore use the version provided in this patch
> instead of changing +111 to -100.
> 
> (In the future I'd like to see that we had some convention to name
> the test scripts and either do comparison to that list or that
> convention also dictates order and this test could be removed. There
> are a few alternatives that we could think of...).

Okay.

(I completely agree that the right solution here is switching to a
naming convention and eliminating the hand-made list of tests.)