Re: [PATCH 19/20] abbrev: support relative abbrev values

["Ævar Arnfjörð Bjarmason" <avarab@gmail.com>]

On Thu, Jun 14 2018, Junio C Hamano wrote:

> Ævar Arnfjörð Bjarmason <avarab@gmail.com> writes:
>
>> On Wed, Jun 13 2018, Junio C Hamano wrote:
>>
>>> Ævar Arnfjörð Bjarmason <avarab@gmail.com> writes:
>>>
>>>> E.g. here's a breakdown of my dotfiles repo:
>>>>
>>>>     $ git -c core.abbrev=4 log  --pretty=format:%h|perl -nE 'chomp;say length'|sort|uniq -c|sort -nr
>>>>         784 4
>>>>          59 5
>>>>           7 6
>>>>
>>>> I don't have a single commit that needs 7 characters, yet that's our
>>>> default. This is a sane trade-off for the kernel, but for something
>>>> that's just a toy or something you're playing around with something
>>>> shorter can make sense.
>>>>
>>>> SHA-1s aren't just written down, but also e.g. remembered in wetware
>>>> short-term memory.
>>>
>>> That's a fine example of what I called "supporting absurdly small
>>> absolute values like 4"; I still do not see why you want "negative
>>> relative values" from that example.
>>
>> Because hardcoding -2 is very different than setting it to 5, because
>> the -2 will scale to the size of the repository, but 5 is just 7-2 where
>> 7 is our default value.
>>
>> So, in general if you want less future proof hashes by some
>> probabilistic metric (whether you use core.validateAbbrev or not) you'd
>> use -2 or -3, just like you might use +2 or +3 if you'd like to have
>> more future-proof hashes (especially with core.validateAbbrev=true).
>
> That still does not make much sense to me at all.
>
> I do agree that something shorter than the default 7 may be more
> appropriate for our wetware short-term memory, and it would make
> sense to grow the "riskier to collide than the default heuristics
> but more memorable" variant as the project grows, _ONLY_ _IF_ our
> wetware short-term memory scales with the project we happen to be
> working on.  But our wetware does not scale with the project we work
> on; at least mine does not.

Yes, it's a trade-off, but just because something is a trade-off doesn't
make it useless.

Aside from the feature I'm proposing, the same thing applies to the
short SHA-1 currently. My ~1k commit dotfiles has 7 characters, but
linux.git has 12. Does that make printing the short SHA-1 at all useless
and we should just fall back to 40 characters if it's say >= 12? No.

12 is still better than 40, but if we could get away with it 10 would be
better than 12. Right now the "get away with it" calculation is a
hardcoded constant, this makes it configurable.

The reason to make it configurable is because you may want more future
proof *or* less future-proof SHA-1s depending on the use-case. Printing
a longer SHA-1 has a cost, including but not limited to:

 * Remembering it for a short time
 * Seeing it on one screen and typing it into another computer manually
 * `git log --oneline` output in a terminal where horizontal space is at
   a premium