New candidate JEP: 357: Migrate from Mercurial to Git

[Aleksey Shipilev]

On 7/16/19 5:58 PM, David Lloyd wrote:
> On Tue, Jul 16, 2019 at 10:48 AM Aleksey Shipilev <shade at redhat.com> wrote:
>> Yes, but we are comparing "fresh clone from master server" use case here. I carry around tarballs of
>> Mercurial workspaces from https://builds.shipilev.net/workspaces/, and it is also blazingly fast.
>> But that misses the point I want to make.
> 
> Generally with Git there is no value in a fresh clone from master
> unless you well and truly have no other checkouts anywhere (or the
> project is small, which is not the case here).  It would be very
> unusual for a developer to be *required* to clone from upstream more
> than one time unless they were developing from a new, empty system and
> didn't have reasonable access to any other system that had a clone. 

Neither does Mercurial for local development. But here we are, comparing clone performance. I find
myself doing fresh JDK clones very often across multiple systems -- this is why I maintain my stash
of Mercurial workspace tarballs.

See what JEP says here: "The reduction in metadata preserves local disk space and ***reduces clone
times, since fewer bits have to go over the wire.***". It seems to conflate on-the-wire size and
on-the-disk size. My experiment shows that claim is problematic: the amount of bits going over the
wire is lower for fresh clones with clonebundles. I suspect that additive pulls into the local repo
produce much less on-the-wire traffic to really care.

Don't you agree with that point?

-- 
Thanks,
-Aleksey