From: Junio C Hamano <gitster@pobox.com>
Date: Mon, 29 Feb 2016 21:51:12 GMT
Subject: Re: [PATCHv19 00/11] Expose submodule parallelism to the user
Message-ID: <xmqqsi0b89tb.fsf@gitster.mtv.corp.google.com>
In-Reply-To: <56D4B805.7090702@kdbg.org>

Johannes Sixt <j6t@kdbg.org> writes:

> Am 29.02.2016 um 22:19 schrieb Junio C Hamano:
>> Stefan Beller <sbeller@google.com> writes:
>>
>>> Maybe we want to remove the struct child_process from the
>>> function signature of the callbacks and callbacks need to rely on
>>> the data provided solely thru the pointer as passed around for
>>> callback purposes, which the user is free to use for any kind
>>> of data.
>>
>> I think that is the most sensible.
>
> I also think that is the better approach. Dumping out the argv array
> is not the best end-user experience. It is just a debugging aid, and
> for that we have (or should extend if necessary) GIT_TRACE
> infrastructure. Moreover, a command that failed should have printed
> error messages, and it is not necessary to follow it up with another
> "A child process exited with code N" message.

Exactly.  Stefan, please do not go in the route of butchering the
start_command(), finish_command(), etc. API set with the "uncleaned"
variants.

Thanks.