Re: [PATCH 2/2] use zstd zlib wrapper

From: Jeff King <peff@peff.net>
Date: Sep 15, 2016, 06:28 UTC
Message-ID: <20160915062845.sfackmnezb7oy3j5@sigill.intra.peff.net>
In-Reply-To: <CAGZ79kYcB-x40_w1fcWL3NSp8JU9SPrTAEiru-6Jpb7fDM1Y0w@mail.gmail.com>

On Wed, Sep 14, 2016 at 06:22:17PM -0700, Stefan Beller wrote:

Show 5 quoted lines

> > Disappointingly, the answer seems to be "no".
> 
> After having looked at the data, I disagree with the conclusion.
> And for that I think we need to reason about the frequency
> of the operations happening.

I definitely agree that reads outnumber writes, and it's OK to have an asymmetric tradeoff between the two. zstd5 isn't _too_ bad in that respect. I guess I was just disappointed that the pack size was still bigger, as I was really hoping to see some speed tradeoff without getting a worse pack.

The other thing to weigh against is "if we were designing it today" versus "is it worth the compatibility headaches now". A 6% improvement in "rev-list --objects" is not that amazing for a data format change. Bitmaps were an _easier_ data format change and are more like a 99% speedup.

They do not apply to every operations, but we may be able to do similar space/time tradeoffs that are easier to handle in terms of backwards compatibility, and which yield bigger results.

-Peff