Re: [RFC] Generation Number v2

From: Stefan Beller <sbeller@google.com>
Date: Oct 29, 2018, 19:22 UTC
Message-ID: <CAGZ79ka-FTqaXdrMixjUp2THJ3L0YvEnkKxs3XFgB3WEEy2-Tg@mail.gmail.com>
In-Reply-To: <6367e30a-1b3a-4fe9-611b-d931f51effef@gmail.com>

Show 14 quoted lines

> Based on the performance results alone, we should remove minimum
> generation numbers, (epoch, date) pairs, and FELINE index from
> consideration. There are enough examples of these indexes performing
> poorly.
>
> In contrast, maximum generation numbers and corrected commit
> dates both performed quite well. They are frequently the top
> two performing indexes, and rarely significantly different.
>
> The trade-off here now seems to be: which _property_ is more important,
> locally-computable or backwards-compatible?
>
> * Maximum generation number is backwards-compatible but not
>    locally-computable or immutable.

These maximum generation numbers sound like the reverse of the generation numbers as they are currently implemented, i.e. we count all commits between the commit A and all heads. How would this impact creation of a commit?

The current generation numbers can be lazily updated or not updated at all. In my understanding of the maximum generation numbers, a new commit would make these maximum generation numbers invalid (i.e. they have to be recomputed).

Are there ways out by deferring the computation of maximum generation numbers while still being able to work with some commits that are un-accounted for?

When recomputing these numbers, the current generation number (V0) has the property that already existing numbers can be re-used as-is. We only need to compute the numbers for new commits, and then insert this to the appropriate data structures (which is a separate problem, one could imagine a split generation numbers file like the split index)

For the V2 maximum generation numbers, would we need to rewrite the numbers for all commits once we recompute them? Assuming that is true, it sounds like the benchmark doesn't cover the whole costs associated with V2, which is why the exceptional performance can be explained.

(Note: The table as read in https://public-inbox.org/git/6367e30a-1b3a-4fe9-611b-d931f51effef@gmail.com/ seems line broken, using gmails web interface is not good for ASCII art and patches, git-send-email would fare better)

>
> * Corrected commit-date is locally-computable and immutable,
>    but not backwards-compatible.

How are these dates not backwards incompatible?
We don't have to expose these dates to the user, but
- just like generation numbers - could store them and use them
but not tell the user about it.

We would need to be really careful to not expose them at all as they look like the real dates, so that could make for an awkward bug report.

The approach of "corrected commit date" sounds like we could have a very lazy approach, i.e. no extra data structures needed for many commits (as the corrected date equals the real date) and only need to store the corrections for some commits. Such an approach however would not make it easy to know if we operate on corrected dates, or if we even checked them for correctness before.

So if we'd have an additional row in the generation numbers file telling the corrected date, then we should be able to be backwards compatible?