From: René Scharfe. <l.s.r@web.de>
Date: Tue, 02 Mar 2021 16:00:34 GMT
Subject: Re: [PATCH 2/2] pretty: add merge and exclude options to %(describe)
Message-ID: <3be35be6-66f7-9191-1c9f-1b279ca760f9@web.de>
In-Reply-To: <xmqqy2f6rc8f.fsf@gitster.c.googlers.com>

Am 01.03.21 um 18:54 schrieb Junio C Hamano:
> René Scharfe. <l.s.r@web.de> writes:
>
>> Am 17.02.21 um 19:31 schrieb Jeff King:
>>> On Sun, Feb 14, 2021 at 11:10:57AM +0100, René Scharfe. wrote:
>>>
>>>> Allow restricting the tags used by the placeholder %(describe) with the
>>>> options match and exclude.  E.g. the following command describes the
>>>> current commit using official version tags, without those for release
>>>> candidates:
>>>>
>>>>    $ git log -1 --format='%(describe:match=v[0-9]*,exclude=*rc*)'
>>>
>>> An interesting side effect of this series is that it allows remote users
>>> asking for archives to fill in this data, too (by using export-subst
>>> placeholders). That includes servers allowing "git archive --remote",
>>> but also services like GitHub that will run git-archive on behalf of
>>> clients.
>>>
>>> I wonder what avenues for mischief this provides. Certainly using extra
>>> CPU to run git-describe.
>>
>> A repository can contain millions of files, each file can contain
>> millions of $Format:...$ sequences and each of them can contain millions
>> of %(describe) placeholders.  Each of them could have different match or
>> exclude args to prevent caching.  Allowing a single request to cause
>> trillions of calls of git describe sounds excessive.  Let's limit this.
>
> An invocation of "git archive" would have to deal with a single
> commit, no?  I wonder if it is a more fruitful direction to go to
> teach format_subst() to "cache" the mapping from <placeholders> to
> <resulting-string> and reuse.

Yes, git archive only works on a single commit.  Caching cannot help
against a DoS attack using describe placeholders with different match
or exclude arguments.

René