[Atp] Re: Possible changes for IETF version

[Bryan Newbold <[e-mail]>]

On Tue, Jul 7, 2026 at 6:36 PM Eric Rescorla <[e-mail]> wrote:

> Has there been some discussion about the version transition story? It
> appears that v4 repositories will be unreadable by existing
> implementations, right? If so, how does a repository know when it's safe to
> move to v4? Conversely, how long do other entities need to support v3?
>

We had some operational experience with this from the v2 to v3 migration
back in 2023. There were fewer implementations and operators at the time,
but there were multiple software implementations even then.

Changes to just the commit object itself (eg, field names) are disruptive,
but I think manageable. Implementations need to read and verify both
versions during a transition period, and most services in the network are
built on a handful of open SDK packages. Folks can deploy test PDS
instances in the live network using the new commit version, and track which
services fail to consume and index them. Then larger PDS operators can
start rolling out the new commit version, including emitting new "no-op"
commits for active-but-quiet accounts. Folks can observe the full-network
firehose to measure progress on the migration, and track down hold-out PDS
instances and urge them to upgrade. After a grace period,
consuming/indexing services could drop legacy support. A small number of
tools (eg, account migration services and repository explorers) might want
to retain legacy support longer, to allow recovery from CAR file exports.

The transparency (mostly open source implementations, and ability to
observe the firehose) and spirit of cooperation in the network have made
this kind of change possible in the past. That window may be closing as
things become larger and more heterogeneous, but i'm confident that the
ecosystem could pull this off. If we decide it is worth the effort.

--bryan