Re: t0301-credential-cache test failure on cygwin

From: Jeff King <peff@peff.net>
Date: Jul 7, 2022, 18:29 UTC
Message-ID: <Yscl/Jx4g74RwkCK@coredump.intra.peff.net>
In-Reply-To: <4529b11a-e514-6676-f427-ffaec484e8f1@ramsayjones.plus.com>

On Thu, Jul 07, 2022 at 04:17:08PM +0100, Ramsay Jones wrote:

> Also, I would like to understand why the code is written as it is
> currently. I'm sure there must be a good reason - I just don't know
> what it is! I suspect (ie I'm guessing), it has something to do with
> operating in a high contention context [TOCTOU on socket?] ... dunno. ;-)

By the way, I was slightly surprised you did not find the explanation in the commit history. A blame[1] of credential-cache--daemon.c shows that the comment was added by 7d5e9c9849 (credential-cache--daemon: clarify "exit" action semantics, 2016-03-18) which mentions the race in the tests. And then searching for that commit message in the list yields the thread I linked earlier with more context[2].

I mention this not as a criticism, because your digging for backstory was otherwise quite thorough. It's only _because_ it was so thorough that I was surprised you didn't find that commit. ;) So I offer it only as a suggestion for future digging.

-Peff

[1] Likewise this works:

       git log --follow -S'important that we clean up' builtin/credential-cache--daemon.c

    but the --follow is necessary because of the rename when it became a
    builtin. It's cool that "blame" handles this seamlessly. :)

[2] Finding the original patch on the list is my go-to trick when a
    commit message hasn't sufficiently explained things. I know Junio
    was for a while (is still?) kept a git-notes mapping of commits to
    emails. In practice, I usually just do a manual search for a few
    unique-looking phrases from the commit message. That should be
    pretty easy with lore these days, though I do it with a local
    archive indexed by notmuch.

-Peff