Re: [PATCH] bloom: ignore renames when computing changed paths

From: Derrick Stolee <stolee@gmail.com>
Date: Apr 9, 2020, 14:00 UTC
Message-ID: <e3ace6a6-eb2e-0544-c9fa-eaed6e9d6c6f@gmail.com>
In-Reply-To: <20200409134724.GA3494212@coredump.intra.peff.net>

On 4/9/2020 9:47 AM, Jeff King wrote:

Show 16 quoted lines

> On Thu, Apr 09, 2020 at 07:56:43AM -0400, Derrick Stolee wrote:
> 
>>> So we really aren't detecting renames in the first place! And indeed,
>>> checking diffopt.detect_rename shows that it is unset. So I'm curious if
>>> there is a case where that would not be true. I _think_ it would only be
>>> true in a program which ran init_diff_ui_defaults(), but never in
>>> git-commit-graph.
>>
>> So our issue was really that the partial clone prefetch logic was just
>> overly aggressive.
> 
> Right, but I'm not sure how this patch could ever have helped, since
> it's just setting a variable to the value it _should_ have already had.
> 
> Or do you just mean that the issue would have gone away with Jonathan's
> patch to make the prefetching less aggressive?

Yes, with Jonathan's patch we stop downloading blobs during Bloom filter computations. The patch this is "replacing" disabled that download in a different way, depending only on if detect_renames is false and the diff output doesn't need file contents. (Jonathan's is better.)

-Stolee