Re: [PATCH v3] tag: add tag.gpgSign config option to force all tags be GPG-signed

From: Mkrtchyan, Tigran <tigran.mkrtchyan@desy.de>
Date: Jun 5, 2019, 20:12 UTC
Message-ID: <1933659713.10490718.1559765529240.JavaMail.zimbra@desy.de>
In-Reply-To: <xmqqv9xkroc5.fsf@gitster-ct.c.googlers.com>

----- Original Message -----

Show 6 quoted lines

> From: "Junio C Hamano" <gitster@pobox.com>
> To: "Tigran Mkrtchyan" <tigran.mkrtchyan@desy.de>
> Cc: "git" <git@vger.kernel.org>, "Jonathan Nieder" <jrnieder@gmail.com>, "Johannes Schindelin"
> <Johannes.Schindelin@gmx.de>
> Sent: Wednesday, June 5, 2019 6:25:46 PM
> Subject: Re: [PATCH v3] tag: add tag.gpgSign config option to force all tags be GPG-signed

Show 5 quoted lines

> Tigran Mkrtchyan <tigran.mkrtchyan@desy.de> writes:
> 
>> As many CI/CD tools don't allow to control command line options when
>> executing `git tag` command, a default value in the configuration file
>> will allow to enforce tag signing if required.

Must of them blindly execute git commands with some hard-coded combination of options. It's clear to me, that they are the source of the problem, but git can be the solution.

Show 15 quoted lines

> 
> Hmm.  Would these "many" tools still allow arbigrary configuration
> set to affect their operation?  It sounds like a bigger issue but it
> is a separate one.
> 
>> The new config-file option tag.gpgSign enforces signed tags. Additional
>> ...
>> will skip the signing step.
> 
> This paragraph is well written.
> 
>> The combination of -u <key-id> and --no-sign not allowed.
> 
> This sentence lacks a verb.  Perhaps s/not allowed/is &/.
>

Jup. Sorry.

Show 17 quoted lines

> But more importantly, I think we should justify why this "not
> allowed" makes sense as the design of the feature. A plausible
> alternative design would simply follow the "last one wins" paradigm,
> where
> 
>    git tag -u key # "-u key" implies "-s"
> 
>    git tag -u key --no-sign # "--no-sign' trumps the implied "-s"
> 
>    git tag --no-sign -u key # "-u key"'s implication of "-s" trumps the
>			     # earlier "--no-sign"
> 
> and having "[tag] gpgsign" simply adds to the implication early in
> the chain to be overridden by later command line options.
> 
> Let's explain why "you cannot give -u <key> and --no-sign at the
> same time" is better than "the last one wins".

This is matter of convention. I never pay attention to the order of options on the git command line, but I don't put conflicting options either, I hope. Does git already have position depended options? If yes, then fine with me. Otherwise, I don't want to introduce ambiguity. Yes, less ambiguity is my answer to why it's better than "the last one wins".

Show 23 quoted lines

> 
>> diff --git a/Documentation/git-tag.txt b/Documentation/git-tag.txt
>> index a74e7b926d..2e5599a67f 100644
>> --- a/Documentation/git-tag.txt
>> +++ b/Documentation/git-tag.txt
>> @@ -64,6 +64,13 @@ OPTIONS
>>  -s::
>>  --sign::
>>  	Make a GPG-signed tag, using the default e-mail address's key.
>> +	The default behavior of tag GPG-signing is controlled by `tag.gpgSign`
>> +	configuration variable if it exists, or disabled oder otherwise.
>> +	See linkgit:git-config[1].
>> +
>> +--no-sign::
>> +	Override `tag.gpgSign` configuration variable that is
>> +	set to force each and every tag to be signed.
>>  
>>  -u <keyid>::
>>  --local-user=<keyid>::
> 
> If we justify "-u and --no-sign do not mix", that design needs to be
> explained to the end users in the documentation, not just in the
> proposed log messsage.

Make sense.

Thanks,
   Tigran.