Received-SPF: pass (google.com: domain of linux-nfs+bounces-2436-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
From: Junio C Hamano <gitster@pobox.com>
To: Patrick Steinhardt <ps@pks.im>
Cc: git@vger.kernel.org,  Chuck Lever <chuck.lever@oracle.com>,
  rsbecker@nexbridge.com,  Jeff King <peff@peff.net>,
  linux-nfs@vger.kernel.org
Subject: Re: [PATCH] reftable: fix tests being broken by NFS'
 delete-after-close semantics
In-Reply-To: <8ac5e94a3930cdd2aee9ea86acda3155674b635c.1711035529.git.ps@pks.im>
	(Patrick Steinhardt's message of "Thu, 21 Mar 2024 16:39:52 +0100")
References: <ZfBwZTL9zqDsac5m@manet.1015granger.net>
	<8ac5e94a3930cdd2aee9ea86acda3155674b635c.1711035529.git.ps@pks.im>
Date: Thu, 21 Mar 2024 10:35:19 -0700
Message-ID: <xmqqsf0j8p48.fsf@gitster.g>
User-Agent: Gnus/5.13 (Gnus v5.13)
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain

Patrick Steinhardt <ps@pks.im> writes:

>  	while ((d = readdir(dir))) {
> -		if (!strcmp(d->d_name, "..") || !strcmp(d->d_name, "."))
> +		/*
> +		 * Besides skipping over "." and "..", we also need to
> +		 * skip over other files that have a leading ".". This
> +		 * is due to behaviour of NFS, which will rename files
> +		 * to ".nfs*" to emulate delete-on-last-close.
> +		 *
> +		 * In any case this should be fine as the reftable
> +		 * library will never write files with leading dots
> +		 * anyway.
> +		 */
> +		if (starts_with(d->d_name, "."))
>  			continue;

Sounds good.  We should count what we positively consider what we
wrote, not random cruft somebody else might have dropped in the
directory, and this is a good first thing to do.

Will queue.  Thanks.