Element-web: Open source the integrations server

[mvgorcum]

The integration server provides some of the core ideals of matrix: integrating other services and allowing easy bridging to other networks that are available.

As such a proprietary component used to provide this feels very much out of place in the otherwise open source matrix ecosystem. I get that new vector is looking at ways to monetize some aspects of matrix, which is perfectly fine, and writing some proprietary services or bridges makes sense in that context, but having riot for everyone on mobile and most on web to use a closed source solution seems quite contrary to the matrix philosophy.

This should also solve this issue that keeps riot labeled with nonfreenet on f-droid.

[edit]If this is in the wrong repository, please feel free to move, or ask me to move it.

[t3chguy]

@mvgorcum I'd recommend riot-meta: https://github.com/vector-im/riot-meta

[Half-Shot]

@t3chguy we had a chat about this oob, and I recommended web instead of meta because meta gets very little attention. To be honest, this question technically doesn't even fall under riot :|

[t3chguy]

I was mainly focussing on

but having riot for everyone on mobile and most on web to use a closed source solution

[mvgorcum]

I don't disagree that it may fit a bit better in riot-meta, but that's not a perfect place either. (It should obviously be in the scalar/modular integration server repo, but that one isn't open) Since riot-meta gets almost no attention I ended up deciding to open the issue here, lacking a 'perfect place' for this issue.

Still, though, if the consensus is that the issue doesn't belong here I'm fine with moving it elsewhere. Otherwise we may want to discuss the actual issue rather than its placement :wink:

[lampholder]

I understand that the sentiment at the core of this is "having the shipped riot client coupled to a closed-source component seems antithetical to the general matrix philosophy".

There are few ways to tackle this: open sourcing the existing proprietary integrations manager would I feel only address the underlying concern in a superficial way - if we threw away scalar and switched to (open source) Dimension today, we'd still be shipping Riot clients (android, ios, riot.im/app) with a single hard-coded integrations manager instance.

The recent release of support for .well-known provides a neat way to address this coupling - by selecting the integration manager based on the homeserver's .well-known config, the homeserver administrator is able to direct which integration manager their users connect to.

Coming back to the question of open sourcing scalar - if shipped riot instances can be less closely coupled with an individual hard-coded integrations manager instance then whether or not to open source scalar becomes a purely commercial rather than a philosophical question - the integrations and integrations manager provided by New Vector are the only closed source components of the matrix ecosystem that we work on, and for the foreseeable future these will remain closed source.

[lampholder]

I recommended web instead of meta because meta gets very little attention

Yeah, tbh I use meta mostly for project planning and don't tend to triage issues that arrive there by any other means :\

[mvgorcum]

Thank you for the response.

There are few ways to tackle this: open sourcing the existing proprietary integrations manager would I feel only address the underlying concern in a superficial way - if we threw away scalar and switched to (open source) Dimension today, we'd still be shipping Riot clients (android, ios, riot.im/app) with a single hard-coded integrations manager instance.

I fully agree that the hard coding of the integrations manager is in itself already antithetical to matrix: it not being federated/decentralized in any way. Allowing the configuration of the manager through .well-known would be a good solution to this issue.

However: As it stands there is no other integrations manager available (be it open source or no) that works with the mobile clients. This means that .well-known in practice won't solve the issue for a large group of users. I suspect this is partially due to the fact that The Spec doesn't specify how the integrations API works, which should be fixed as well (though that should probably be discussed in matrix-doc).

Coming back to the question of open sourcing scalar - if shipped riot instances can be less closely coupled with an individual hard-coded integrations manager instance then whether or not to open source scalar becomes a purely commercial rather than a philosophical question - the integrations and integrations manager provided by New Vector are the only closed source components of the matrix ecosystem that we work on, and for the foreseeable future these will remain closed source.

For the foreseeable future it is highly likely that the default server for the vast majority of riot/matrix users is going to be matrix.org. As long as matrix.org defaults to the closed source solution, even .well-known won't solve this problem.

An alternative would probably be to give the user an explicit choice which integrations manager they want to use. As it stands that is chosen by whoever hosts riot, or through the distributed config.json on riot-desktop. Maybe a specific riot.im/app version allowing for a drop-down list to choose between scalar/modular and dimension?

I won't pretend I am not a little disappointed that scalar/modular won't be open sourced, but I can respect that choice. However, I would really like to see a bit more active work on getting at least an integrations manager that is open source working reliably for all riot clients. Specifically one that would also be accessible to the majority of general users (ie consumers that happen to use riot on matrix.org). Maybe the matrix.org foundation could sponsor travis to get the spec written and/or dimension compatible with riot mobile?

[turt2live]

Related: https://github.com/vector-im/riot-web/issues/6430, https://github.com/vector-im/riot-web/issues/4913, https://github.com/matrix-org/matrix-doc/issues/1286

[lampholder]

Tagging this wontfix just so it stops popping up in my list of untriaged issues :)

[PopeRigby]

Is there a discussion tag? That would probably better categorize this issue.

[onitake]

It seems there is a free alternative to the integration server available: https://github.com/turt2live/matrix-dimension
Would this help?

[frankwalter1301]

I understand that the sentiment at the core of this is "having the shipped riot client coupled to a closed-source component seems antithetical to the general matrix philosophy".

There are few ways to tackle this: open sourcing the existing proprietary integrations manager would I feel only address the underlying concern in a superficial way - if we threw away scalar and switched to (open source) Dimension today, we'd still be shipping Riot clients (android, ios, riot.im/app) with a single hard-coded integrations manager instance.

The recent release of support for .well-known provides a neat way to address this coupling - by selecting the integration manager based on the homeserver's .well-known config, the homeserver administrator is able to direct which integration manager their users connect to.

Coming back to the question of open sourcing scalar - if shipped riot instances can be less closely coupled with an individual hard-coded integrations manager instance then whether or not to open source scalar becomes a purely commercial rather than a philosophical question - the integrations and integrations manager provided by New Vector are the only closed source components of the matrix ecosystem that we work on, and for the foreseeable future these will remain closed source.

The use of integration servers provided by the homeserver is now supported. I guess it makes sense now to make open source scalar.