Re: Re* [PATCH v4 0/3] advice: add "all" option to disable all hints

From: Junio C Hamano <gitster@pobox.com>
Date: May 3, 2024, 20:08 UTC
Message-ID: <xmqq4jbe64l9.fsf@gitster.g>
In-Reply-To: <xmqqbk5m65i8.fsf@gitster.g>

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

Show 15 quoted lines

> Eric Sunshine <sunshine@sunshineco.com> writes:
>
>> different. Aren't both commands addressing the same use-case of
>> comparing one version of a series against a subsequent version? In
>> your response[3], you seemed to agree with that observation and
>> suggested instead simply increasing the default creation factor for
>> both commands (which sounds reasonable to me).
>
> I think Dscho's use case was compare only single updated series of
> his with seen that has tons of irrelevant other patches, to have the
> command sift the patches belong to other topics away while making
> comparison with the older incarnation of his topic.  I never use the
> command for such a comparison, but I can imagine that in such a
> context lower criteria may help reduce false matches of his new
> commits with unrelated commits on 'seen'.

It seems that Dscho was in agreement that format-patch's use case should try to be more aggressive at least back then. In the message in the thread you pointed

 https://lore.kernel.org/git/nycvar.QRO.7.76.6.1903211209280.41@tvgsbejvaqbjf.bet/

he does not give us the exact reason why he does not think the "more aggressive" mode is not suitable for other use cases, though.

A similar thread was raised more recently:

 https://lore.kernel.org/git/rq6919s9-qspp-rn6o-n704-r0400q10747r@tzk.qr/

Also, there was an attempt to clarify what the "factor" really meant, but we did not get a real conclusion other than the UNIT to measure the "factor" is called "percent" (without specifying what 100% is relative to, "percent" does not mean much to guide users to guess what the right value would be).

 https://lore.kernel.org/git/85snn12q-po05-osqs-n1o0-n6040392q01q@tzk.qr/

So, yes, --creation-factor=<value> is messy, I think a very high value, much higher than the hardcoded default of 60, is more appropriate for use in format-patch, but we do not know what bad effect it would have if we used much higher default everywhere.