Re: [PATCH v2 0/5] commit-graph: fsck zero/non-zero generation number fixes

From: Taylor Blau <me@ttaylorr.com>
Date: Aug 21, 2023, 21:25 UTC
Message-ID: <ZOPWY3ydDS1leV5A@nand.local>
In-Reply-To: <20230817195108.GC3032779@coredump.intra.peff.net>

On Thu, Aug 17, 2023 at 03:51:08PM -0400, Jeff King wrote:

Show 13 quoted lines

> > One very minor thing is that how much value are we getting by
> > reporting the object names of one example from each camp, instead of
> > just reporting a single bit "we have commits not counted and also
> > counted their generations, which is an anomaly".
> >
> > Obviously it does not matter.  Even if we stopped doing so, the code
> > would not become much simpler.  We'd just use a word with two bits
> > instead of two pointers to existing in-core objects, which does not
> > have meaningful performance implications either way.
>
> Yeah, I wasn't sure if the commit names were valuable or not. Two bits
> would definitely work (though I have a slight preference for two
> boolean variables, just because I find the syntax easier to read).

I think having a single example of both a commit with zero and non-zero generation is marginally useful. I think keeping track of two commit pointers is more straightforward than the bit-field, and it does not complicate things too much, so I think it is worth keeping.

> I don't think we've heard from Taylor, but I saw his original patches
> were in 'next'. I'm happy to clean up what I posted, but I'm also happy
> if we just merge what's in next and move on.

Sorry that this fell to the bottom of my queue, which I am just digging out of now that 2.42.0 has been tagged.

I think that the clean-up you suggested is worthwhile. Let's replace what we have in 'next' with the reroll that I'm about to submit...

Thanks, Taylor