Re: [PATCH 14/14] completion: add default merge strategies

From: Junio C Hamano <gitster@pobox.com>
Date: Jun 25, 2019, 01:43 UTC
Message-ID: <xmqq1rzih228.fsf@gitster-ct.c.googlers.com>
In-Reply-To: <CAMP44s0y_3Ltv_hnoVByTRwZrDU0+PfRo3R=Lxm6wDk+=sypJg@mail.gmail.com>

Felipe Contreras <felipe.contreras@gmail.com> writes:

Show 10 quoted lines

> On Mon, Jun 24, 2019 at 12:24 PM Junio C Hamano <gitster@pobox.com> wrote:
>>
>> Felipe Contreras <felipe.contreras@gmail.com> writes:
>>
>> > In case the command fails.
>>
>> It is unclear what you wanted to say with this.  What command?
>> After "git merge" fails?
>
> Yes. The command that __git_list_merge_strategies() uses.

Next round, write that in the proposed log message, please. An issue in the proposed commit log message that puzzles reviewers is something we expect future readers of "git log" to also stumble on.

>  % cd /tmp
>  % git merge -s help
> fatal: not a git repository (or any parent up to mount point /)
> Stopping at filesystem boundary (GIT_DISCOVERY_ACROSS_FILESYSTEM not set).

I think the recent <20190612085606.12144-1-pclouds@gmail.com> established a good pattern we should follow; when a command we run to get list of things to use in completion fails, we refrain from caching that broken output, and arrange so that we will try again. It looks to me that "git merge -s help" barfing outside a repository is a good candidate to follow that pattern. Outside a repository, the user will not be able to perform a merge with any strategy, so not completing the command line when the user say "git merge -s <TAB>" outside a repository is not the end of the world, as long as we follow the right codepath to grab the available strategies once the user goes into a repository where "git merge -s help" works, no?

Thanks.