Re: [RFC 2/3] am: semi working --cover-at-tip

[Nicolas Morey-Chaisemartin <NMoreyChaisemartin@suse.de>]

Le 14/11/2017 à 07:00, Junio C Hamano a écrit :
> Nicolas Morey-Chaisemartin <NMoreyChaisemartin@suse.de> writes:
>
>>  	if (!git_config_get_bool("commit.gpgsign", &gpgsign))
>>  		state->sign_commit = gpgsign ? "" : NULL;
>> +
>>  }
> Please give at least a cursory proof-reading before sending things
> out.
>
>> @@ -1106,14 +1131,6 @@ static void am_next(struct am_state *state)
>>  
>>  	oidclr(&state->orig_commit);
>>  	unlink(am_path(state, "original-commit"));
>> -
>> -	if (!get_oid("HEAD", &head))
>> -		write_state_text(state, "abort-safety", oid_to_hex(&head));
>> -	else
>> -		write_state_text(state, "abort-safety", "");
>> -
>> -	state->cur++;
>> -	write_state_count(state, "next", state->cur);
> Moving these lines to a later part of the source file is fine, but
> can you do so as a separate preparatory patch that does not change
> anything else?  That would unclutter the main patch that adds the
> feature, allowing better reviews from reviewers.
>
> The hunk below...

Sure. I usually do all this later in the process.
>> +/**
>> + * Increments the patch pointer, and cleans am_state for the application of the
>> + * next patch.
>> + */
>> +static void am_next(struct am_state *state)
>> +{
>> +	struct object_id head;
>> +
>> +	/* Flush the cover letter if needed */
>> +	if (state->cover_at_tip == 1 &&
>> +	    state->series_len > 0 &&
>> +	    state->series_id == state->series_len &&
>> +	    state->cover_id > 0)
>> +		do_apply_cover(state);
>> +
>> +	am_clean(state);
>> +
>> +	if (!get_oid("HEAD", &head))
>> +		write_state_text(state, "abort-safety", oid_to_hex(&head));
>> +	else
>> +		write_state_text(state, "abort-safety", "");
>> +
>> +	state->cur++;
>> +	write_state_count(state, "next", state->cur);
>> +}
> ... if you followed that "separate preparatory step" approach, would
> show clearly that you added the logic to call do_apply_cover() when
> we transition after applying the Nth patch of a series with N patches,
> as all the existing lines will show only as unchanged context lines.

Agreed. The split of am_clean should probably have its own commit too.

>
> By the way, don't we want to sanity check state->last (which we
> learn by running "git mailsplit" that splits the incoming mbox into
> pieces and counts the number of messages) against state->series_len?
> Sometimes people send [PATCH 0-6/6], a 6-patch series with a cover
> letter, and then follow-up with [PATCH 7/6].  For somebody like me,
> it would be more convenient if the above code (more-or-less) ignored
> series_len and called do_apply_cover() after applying the last patch
> (which would be [PATCH 7/6]) based on what state->last says.

I thought about that.
Is there a use case for cover after the last patch works and removes the need to touch am_next (can be done out of the loop in am_run).

If that multiple series in a mbox is something people do, your concern could be solved by flushing the cover when state->series_id goes back to a lower value.

Nicolas