Re: [PATCH] fts: avoid false-positive cycle-detection

[Jim Meyering]

James Youngman wrote:
> On Sun, Jul 19, 2009 at 9:35 PM, Jim Meyering<address@hidden> wrote:
>> I'm considering a rewrite of rm that uses fts.c,
>> and see that the resulting rm fails coreutils' rm/cycle test
>> because it incorrectly reports a cycle for that unusual case.
>>
>> The following patch fixes fts so that it properly reinitializes one of
>> its cycle-detection data structure for each new command line argument.
>>
>> I might save a little by merely clearing the hash table
>> and reusing it, rather than freeing it altogether, but don't
>> think that would be worth the added code.
>>
>> I have not yet been able to cause trouble with chmod, chown, chcon,
>> or du, but haven't tried very hard, either.
>
>>From your description, I would have assumed that find would be
> affected, but this appears not to be the case:
>
> $ ./find/find  ~/tmp/a  ~/tmp/a
> /home/james/tmp/a
> /home/james/tmp/a/file
> /home/james/tmp/a
> /home/james/tmp/a/file

The difference is that the rewritten rm calls fts_open just once with all
of its non-option command line arguments.  By contrast, ftsfind.c calls
fts_open for each command line argument individually.  Since fts_open is
what initializes the cycle-detection machinery, find is not susceptible.