Re: [PATCH v3 08/13] graph: tidy up display of left-skewed merges

From: Derrick Stolee <stolee@gmail.com>
Date: Oct 17, 2019, 12:34 UTC
Message-ID: <863fb1b9-8829-64db-d355-56a97e979427@gmail.com>
In-Reply-To: <xmqqsgnt8hlh.fsf@gitster-ct.c.googlers.com>

On 10/16/2019 12:00 AM, Junio C Hamano wrote:

Show 41 quoted lines

> "James Coglan via GitGitGadget" <gitgitgadget@gmail.com> writes:
> 
>> This effect is applied to both "normal" two-parent merges, and to
>> octopus merges. It also reduces the vertical space needed for pre-commit
>> lines, as the merge occupies one less column than usual.
>>
>>         Before:         After:
>>
>>         | *             | *
>>         | |\            | |\
>>         | | \           | * \
>>         | |  \          |/|\ \
>>         | *-. \
>>         | |\ \ \
> 
> Looking at these drawings reminded me of a tangent that is brought
> up from time to time.  We do not do great job when showing multiple
> roots.
> 
> If you have a history like this:
> 
>       A---D---E
>          /
>     B---C
> 
> drawing the graph _with_ the merge gives a reasonable representation
> of the entire topology.
> 
>     * 46f67dd E
>     *   6f89516 D
>     |\  
>     | * e6277a9 C
>     | * 13ae9b2 B
>     * afee005 A
> 
> But if you start drawing from parents of D (excluding D), you'd get
> this:
> 
>     * e6277a9 C
>     * 13ae9b2 B
>     * afee005 A

I hit this very situation recently when I was experimenting with 'git fast-import' and accidentally created many parallel, independent histories. Running "git log --graph --all --simplify-by-decoration" made it look like all the refs were in a line, but they were not. (The one way I knew something was up: the base commits also appeared without a decoration. That was the only clue that the histories did not continue in a line.)

Show 11 quoted lines

> 
> and the fact that B and A do not share parent-child relationships is
> lost.  An easy way to show that would be to draw the bottom three
> lines of the full history output we saw earlier:
> 
>     | * e6277a9 C
>     | * 13ae9b2 B
>     * afee005 A
> 
> either with or without the vertical bar to imply that A may have a
> child.

The natural extension of this would be multiple columns:

 | | | | | *
 | | | | *
 | | | *
 | | *
 | *
 *

This does not appear too cumbersome, and such a situation should be rare.

> This is not something that has to be done as part of this series,
> but I am hoping that the internal simplification and code
> restructuring that is done by this series would make it easier to
> enhance the system to allow such an output.

I agree. An excellent idea for a follow-up.

Thanks, -Stolee