Re: Bad behavior in xhistogram.c in the face of hash collisions?

On Mon, Aug 16, 2021, at 2:40 PM, Greg Hurrell wrote:

I think I may have found a bug in the histogram diff algorithm that
manifests when there is a hash collision.  This behavior seems to exist
in the JGit implimentation (https://git.io/J0Ud8) too, and was brought
across with the port-to-C in 8c912eea94a2.

Thought I had better bump the thread as I did a fire-and-forget on it
a year ago and never followed up because I wasn't super confident
about my findings.

I'd be interested in corroboration of my analysis, to see whether there
really is a bug there. Not quoting my entire email here so as to keep
things brief, but the original can be seen at:

https://public-inbox.org/git/4e0eff48-4a3e-4f0e-9ed2-d01ec38442a5@www.fastmail.com/

(Phillip, CC'ing you for an opinion because I see you have made a few
changes to xdiff/xhistogram.c relatively recently.)

Best wishes,
Greg