bernardd
commented
7 years ago In general I'm in favour of using the built-in stuff wherever we can, however in this case I'm not sure it will provide what we need (at least in its current form). Reading through the docs it looks like, funactionality wise, it does essentially exactly what TROS/S3 does right now - that is, provide a mechanism to get an S3 upload URL.
However it's missing a few key elements that we currently implement:
- Support for file metadata (we use this for access permissions, which are also absent)
- Feedback to the application as to where the file has been uploaded - there's no hooks or anything else that we can directly use; at best we'd have to incercept the result packet and parse it out. I could also add a hook, of course, but once we start hacking patches into it it lessens the value of using a built-in module.
- Download support (I presume it's meant to be left to the client to share the cryptographically random URL for the file).
While we could work around most of these if we really wanted to, I'm not sure there's any benefit beyond XEP conformity (a change which would also mean work on the client side). There's another, arguably more important issue though:
Because this, at its core, does exactly what we currently do, it means it doesn't do exactly what we want to do in #452. There's no scope for inspecting the images (or whatever other file type) prior to making them available to other users.
So all in all, while it's a nice idea on the surface, I'm going to vote it down for the moment.
What might be worth considering, however, is modifying the TROS protocol to be XEP-363 compliant (with download extensions). At a glance it looks like they're similar enough that it wouldn't be a huge body of work. I'm open to opinions on that one.
toland
bengtan
MongooseIM 2.0.1 supports XEP-363 HTTP File Upload. Using a module already in MIM and backed by a XEP would probably be preferable to supporting our own implementation.
mod_http_uploaddocs