Re: Weird revision walk behaviour

From: Kevin Bracey <kevin@bracey.fi>
Date: May 31, 2018, 14:54 UTC
Message-ID: <28359a94-e584-a963-428d-2cf11f2cb895@bracey.fi>
In-Reply-To: <20180531054355.GA17344@sigill.intra.peff.net>

On 31/05/2018 08:43, Jeff King wrote:

>
> If there are zero parents (neither relevant nor irrelevant), is it still
> TREESAME? I would say in theory yes.

Not sure - I think roots are such a special case that TREESAME effectively doesn't matter. We always test for roots first.

>   So what I was proposing would be to
> rewrite the parents to the empty set.

That feels a bit radical - I believe we need to retain (some) parent information for modes that show it (eg the dangling unfilled circles in gitk). And making it a root I think could cause other problems with making it look like we have a disjoint history. I believe the next simplification step may be trying to follow down to the common root.

> What next here? It looks like we have a proposed solution. Do you want
> to try to work up a set of tests based on what you wrote earlier?

I was hoping Gábor would carry on, as he's made a start... I was just planning to back-seat drive.

> I'd also love to hear from Junio as the expert in this area, but I think
> he's been a bit busy with maintainer stuff recently. So maybe I should
> just be patient. :)
>

Likewise - I have been quite deep into this, but it was a quite short window of investigation a long time ago, and I've not looked at it since. Would like input from someone with more active knowledge.

Kevin