Re: [PATCH] Fix 'git clean' failure on NFS.

From: Török Edwin <edwintorok@gmail.com>
Date: Jun 18, 2010, 13:09 UTC
Message-ID: <4C1B7026.2010501@gmail.com>
In-Reply-To: <4C1B63CB.7080209@viscovery.net>

On 06/18/2010 03:17 PM, Johannes Sixt wrote:

Show 11 quoted lines

> Am 6/18/2010 13:07, schrieb Török Edwin:
>>  		else if (S_ISDIR(st.st_mode)) {
>> -			if (!remove_dir_recursively(path, only_empty))
>> +			if (!remove_dir_recursively(path, only_empty)) {
>> +				did_rm = 1;
>>  				continue; /* happy */
> 
> Is it possible that remove_dir_recursively exits with 0, but it actually
> did not remove the directory? That would result in an endless loop: it
> would have to re-enter remove_dir_recursively on the next iteration,
> return 0 again (without removing the directory), next iteration, etc.

Hmm, I think this could cause problems:
	if ((flag & REMOVE_DIR_KEEP_NESTED_GIT) &&
	    !resolve_gitlink_ref(path->buf, "HEAD", submodule_head))
		/* Do not descend and nuke a nested git work tree. */
		return 0;

What if we limit the amount of times you can loop? The worst case is that you loop == number of files in the directory, if you loop more times than we could break out early.

Or we can compare the number of files listed in one pass, and in another. If they stay equal, we can break out (since we'd infloop), if it is higher (new files added) or lower (successfully removed some) we can keep looping.

How does that sound?

--Edwin