Re: [PATCH v7] show-branch: convert per-branch flags to commit-slab

From: Junio C Hamano <gitster@pobox.com>
Date: Jul 17, 2026, 17:53 UTC
Message-ID: <xmqqcxwl33kr.fsf@gitster.g>
In-Reply-To: <xmqqse5h63js.fsf@gitster.g>

Junio C Hamano <gitster@pobox.com> writes:

Show 25 quoted lines

>> We've seen a strong uptick in threads that are obviously AI generated,
>> only, and at times it just feels like one is merely talking to a prompt.
>> This just doesn't scale well, as it leads to constant iterations and
>> back and forth without much thinking being involved. So we require the
>> other side to stop every once in a while and invest the necessary time,
>> too. Otherwise the community will simply stop working, and that doesn't
>> serve anyone well.
>>
>> Sorry if I came across as harsh.
>
> Thanks for saying this.
>
> When viewed in that light, what the v7 patch does is extremely
> incoherent.  It gives the impression of having been generated by an
> automated assistant and sent without human oversight, or perhaps
> drafted in a state of severe late-night exhaustion.  For instance,
> the commit message claims to have lifted 'max_revs' completely, yet
> the proposed documentation updates still reference a hard limit of
> 64.  It also removes the local definition of 'UNINTERESTING', even
> though the comment immediately above it still advises our future
> selves to migrate to the shared definition eventually.
>
> It appears the automation was not used merely for structuring the
> reply; the changes in the patch itself show signs of having been
> generated and sent out without any human oversight X-<.

By the way, the two items that follow 'For instance' above are mere examples. Addressing them alone will not suffice.

Like you, I have no desire to interact with a human proxy sitting between an automated generator and the mailing list, relaying what the machine produced under the guise of having reviewed it, without understanding or thinking through the implications. We want to see changes designed from the ground up by human minds, not synthetic slop.

It is fine to use automated assistants to proofread what you have produced yourself. To judge whether to adopt or discard their suggestions, you must possess a deeper understanding of the problem domain than the tools do. At the very least, to employ a 'human produces, machine assists' workflow, you must first compose the draft yourself before asking the tool to polish it, which inherently requires sufficient domain knowledge and forces you to think about the problem.

In contrast, blindly relaying what the machine generated, while claiming to have reviewed it carefully, is something anyone can do without even realizing how superficial that review actually was. The 'machine generates, human reviews' model is fundamentally broken when compared to its converse.

Pardon if this sounds even harsher.