On 2020-02-26 at 17:16:01, Han-Wen Nienhuys wrote:
> On Fri, Feb 21, 2020 at 7:40 AM Jonathan Nieder <jrnieder@gmail.com> wrote:
> > > This adds the reftable library, and hooks it up as a ref backend.
> >
> > As promised, here's a patch to include the reftable spec in git.git.
> > Please include this in the next iteration of this patch series (or if
> > you prefer for it to land separately, that's also fine with me).
> 
> I did.
> 
> > [...]
> > >  * support SHA256 as version 2 of the format.
> >
> > I'd prefer if we error out for now when someone tries to use reftable
> > when the_hash_algo != sha1.  That would buy a bit more time to pin
> > down the details of version 2 (e.g., should it only add an object
> > format field, or object format and oid size?).

If we do this, the testsuite will start to fail with SHA-256 unless you
label those tests with the SHA1 prerequisite.

> You have convinced me that we should go with the 4-byte identifier.
> 
> How about setting version=2 and extending the header by 4 bytes (which
> hold the 4-byte identifier). The footer would also be increased in
> size equivalently. There are no further changes.

Just so you're aware, the values we're going to be using in pack index
version 3 are in sha1-file.c as the format_version values of struct
hash_algo.  They're 0x73686131 for SHA-1  ("sha1", big-endian) and
0x73323536 for SHA-256 ("s256", big-endian).

I know Jonathan is already familiar with those, but I thought I'd point
them out since others may not be.
-- 
brian m. carlson: Houston, Texas, US
OpenPGP: https://keybase.io/bk2204