Registration fallback doesn't seem to be specced

[richvdh]

but https://matrix.org/_matrix/static/client/register is a thing

[richvdh]

dunno if it works though, tbh

[richvdh]

(it doesn't: matrix-org/synapse#6709)

[turt2live]

is this not https://matrix.org/docs/spec/client_server/r0.6.0#fallback ? (because register is UIA)

[richvdh]

well, not really: https://matrix.org/docs/spec/client_server/r0.6.0#fallback specifies fallbacks for individual UIA steps, but still expects the client to do the mechanics of UIA itself.
https://matrix.org/_matrix/static/client/register is supposed to take care of the entire UIA registration flow from start to finish.

However, given it seems to have been broken ever since we removed the non-UIA registration endpoints, and it has never been specced, maybe we should just say that clients have to implement UIA if they want to support registration.

[richvdh]

ok, this has been broken forever and nobody seems to care that much, so rather than fixing it and speccing it, I think we should remove it.

[turt2live]

@richvdh lack of activity on an issue isn't an indicator of it being useless. By existing in synapse it is considered to exist, for better or worse, which is what issues under spec-omission track: spec that's literally just missing but otherwise considered truthful.

[KitsuneRal]

However, given it seems to have been broken ever since we removed the non-UIA registration endpoints, and it has never been specced, maybe we should just say that clients have to implement UIA if they want to support registration.

Which is a pretty high bar, given how versatile UIA is. To have an end-to-end UIA fallback would be really nice - especially if it were aligned with SSO flow, ending with HTTP, rather than JavaScript, callback.

[bmarty]

I can confirm that it was working, but today it does not work anymore. At least Element Android still relies on it if a registration stage is unknown and can't be handled natively

[richvdh]

@bmarty I think you are confusing UIA stage fallback (https://matrix.org/_matrix/client/v3/auth/.../fallback/...) which is specced with registration fallback (https://matrix.org/_matrix/static/client/register), which is not, and is broken.

[richvdh]

By existing in synapse it is considered to exist, for better or worse,

The thing is, it doesn't exist in Synapse. All that exists in Synapse is a thing that serves you error messages.

If we suddenly decided it was a useful thing that we wanted to re-introduce, I'd expect us to go through an MSC process.

I really don't buy the idea that we should spec a thing that hasn't existed for over three years.

[richvdh]

To have an end-to-end UIA fallback would be really nice

I don't disagree with this, but still - I'd expect it to go through MSC, and in any case the OAuth2 changes are looking more promising IMHO.

[turt2live]

yea, 2 years ago the MSC process didn't feel correct for this sort of thing, but now it does. Re-closing.