Re: [PATCH RFC] git-am: support any number of signatures

From: Christian Couder <christian.couder@gmail.com>
Date: Sep 25, 2014, 10:04 UTC
Message-ID: <CAP8UFD0kfSBhSwu5Mb46XEHqYCE0SEsZd_3c0Sm4WzEc-NNc7w@mail.gmail.com>
In-Reply-To: <xmqqbnq6jm0s.fsf@gitster.dls.corp.google.com>

On Tue, Sep 23, 2014 at 7:15 PM, Junio C Hamano <gitster@pobox.com> wrote:

Show 23 quoted lines

> Christian Couder <christian.couder@gmail.com> writes:
>
>>
>> This is probably not as simple as you would like but it works with
>> something like:
>>
>> $ git interpret-trailers --trailer "Acked-by: Michael S. Tsirkin
>> <mst@redhat.com>" --trailer "Reviewed-by: Michael S. Tsirkin
>> <mst@redhat.com>"  --trailer "Tested-by: Michael S. Tsirkin
>> <mst@redhat.com>" 0001-foo.patch >to_apply/0001-foo.patch
>>
>> and then:
>>
>> $ git am to_apply/*.patch
>
> If I understand it correctly, Michael is envisioning to implement
> his "git am -s art" (I would recommend against reusing -s for this,
> though.  "git am --trailer art" is fine) and doing so by using
> interpret-trailers as an internal implementation detail, so I would
> say the above is a perfectly fine way to do so.  An equivalent of
> that command line is synthesized and run internally in his version
> of "git am" when his "git am" sees "--trailer art" option using
> those am.{"a","r","t"}.trailer configuration variables.

Yeah, that's the idea, except that I think "--trailer art" should mean a trailer like:

art: <default value>

(if there is no trailer.art.key config variable defined).

Having am.{"a","r","t"}.trailer configuration variables to define full trailers seems too specific and quite confusing regarding how git interpret-trailers work without those variables.

Show 16 quoted lines

>> Also by using something like:
>>
>> $ git config trailer.a.key Acked-by
>> $ git config trailer.r.key Reviewed-by
>> $ git config trailer.t.key Tested-by
>>
>> the first command could be simplified to:
>
> So I think this mechanism buys Michael's use case very little, if
> any.  It might be useful in other contexts, though.
>
> What would be more interesting is if the primitives you have,
> e.g. "replace", "append", etc. are sufficient to express his use
> case and similar ones.  For example, when working on multiple
> trailers (e.g. "am --trailer art" would muck with three kinds), how
> should "do this if exists at the end and do that otherwise" work?

The way the "trailer.<foo>.ifexists" and "trailer.<foo>.where" work is
quite orthogonal to the way we decide what the content of the trailer
is.
If we make "--trailer art" mean "--trailer Acked-by: Michael --trailer
Reviewed-by: Michael --trailer Tested-by: Michael", then it should
work as if we had passed the latter to the command line.

Show 6 quoted lines

> To an existing message ends with Michael's Signed-off-by:, if his
> "git am --trailer arts" is called to add these three and then a
> Signed-off-by: from him, should it add an extra S-o-b (because his
> existing S-o-b will no longer be the last one after adding Acked and
> others), or should it refrain from doing so?  Can you express both
> preferences?

The default for "trailer.where" is "end", and for "trailer.ifexists" it is "addIfDifferentNeighbor". That means that by default it will add the four new trailers at the end.

If either "trailer.ifexists" or "trailer.S-o-b.ifexists" is set to "addIfDifferent", then only the first 3 new trailers will be added at the end. So yes you can express both preferences.

Show 5 quoted lines

> Another thing that got me wondered this morning while I was thinking
> about this topic was if "replace" is flexible enough.  We may want
> to have "if an entry exists (not necessarily at the end), remove it
> and then append a new one with this value at the end" to implement
> "Last-tested-by: me@my.domain", for example.

That's what "replace" does already. That's why I changed the previous name for this option from "overwrite" to "replace". You have an "overwrite behavior" with where = after and ifexist = replace, and you have a "remove old one and append new one behavior" with where = end and ifexist = replace.

Thanks, Christian.