Re: [PATCH 0/1] git-config --add allows values from stdin

From: Phillip Wood <phillip.wood123@gmail.com>
Date: Sep 23, 2019, 09:46 UTC
Message-ID: <8c079514-bc18-cf03-1f82-4c2d2e878453@gmail.com>
In-Reply-To: <20190922031128.GA76333@syl.lan>

Hi Taylor and ZJ

On 22/09/2019 04:11, Taylor Blau wrote:

Show 8 quoted lines

> Hi ZJ,
> 
> On Tue, Sep 17, 2019 at 03:31:34PM +0200, Zeger-Jan van de Weg wrote:
>> When adding or updating configuration values using git-config, the
>> values could all be observed by different processes as these are passed
>> as arguments. In some environments all commands executed are also all
>> logged. When the value contains secrets, this is a side effect that
>> would be great to avoid.

How much extra security does this actually add? - do the processes that can observe the command line arguments also have read access to the git config file?

  At GitLab we use Rugged/libgit2 to circumvent

Show 23 quoted lines

>> this property[1].
>>
>> The following patch allows a value to be set through stdin when the user
>> passes a `--stdin` flag.
> 
> Interesting. I had thought some time ago about making an interactive
> line-oriented 'mode' for using 'git-config(1)', which would allow
> callers to add/delete/fetch multiple variables using only a single
> process.
> 
> This would satisfy a more general use-case than yours: particularly my
> idea was grown out of wanting to specify or read many configuration
> entries at once when using a tool built around Git, such as Git LFS.
> 
> I had not considered tying '--stdin' to the '--add' (implicit or not)
> mode of 'git-config(1)'. It is an interesting idea to be sure.
> 
> On the one hand, it lends itself to other modes, such as '--get'
> combined with '--stdin', or '--unset' in the same fashion. One could
> imagine that each of these would take either a key/value-pair (in the
> case of '--add') or a set of key(s) (in the remaining cases). The most
> desirable aspect is that this would allow for a clear path to this
> series being picked up.

It would be great to be able to --get multiple values and I can see people wanting to be able to --unset them as well.

Show 5 quoted lines

> On the other hand, tying '--stdin' to a particular mode of using 'git
> conifg' seems overly restrictive to me. If I am building a tool that
> wants to fetch some values in the configuration, and then add/unset
> others based on the results using only a single process, I don't think
> that a mode-based '--stdin' flag gets the job done.

That's true but I don't know how common it is compared to a script wanting to read a bunch of config variables at startup (i.e. does it warrant the extra complexity)

Best Wishes

Phillip

Show 34 quoted lines

> One happy medium that comes to mind is a new '--interactive' mode, which
> implies '--stdin' and would allow the above use-case, e.g.:
> 
>    $ git config --interactive <<\EOF
>    get core.myval
>    set core.foo bar
>    unset core.baz
>    EOF
> 
> (An off-topic note is that it would be interesting to allow more
> fanciful options than 'get', e.g., 'get' with a '--type' specifier, or
> some such).
> 
> I'm not sure if anyone actually wants to use 'git-config(1)' in this
> way, but I figured that I would at least share some things that I was
> thinking about when initially considering this proposal.
> 
>> [1]: https://gitlab.com/gitlab-org/gitaly/blob/8ab5bd595984678838f3f09a96798b149e68a939/ruby/lib/gitlab/git/http_auth.rb#L14-15
>>
>> Zeger-Jan van de Weg (1):
>>    Git config allows value setting from stdin
>>
>>   Documentation/git-config.txt |  5 ++++-
>>   builtin/config.c             | 23 +++++++++++++++++++++--
>>   t/t1300-config.sh            | 11 +++++++++++
>>   3 files changed, 36 insertions(+), 3 deletions(-)
>>
>> --
>> 2.23.0
>>
> 
> Thanks,
> Taylor
>