Acceptable characters for MXC opaque identifiers should be more clearly described

[joepie91]

From the spec:

Content locations are represented as Matrix Content (MXC) URIs. They look like:

mxc://<server-name>/<media-id>

<server-name> : The name of the homeserver where this content originated, e.g. matrix.org
<media-id> : An opaque ID which identifies the content.

However, it is not specified anywhere (as far as I can tell) what the valid character range is for these opaque media-ids. In particular, it doesn't specify whether the media-id may contain slashes - a detail that's quite semantically important for a lot of HTTP request routing implementations, which often treat a "URL parameter" as a string of any characters other than a slash, eg. as in /media/:id where /media/foo/bar would not match.

[richvdh]

This is mentioned under #174, and an attempt to answer it is in the as-yet-unimplemented MSC1597: https://github.com/matrix-org/matrix-doc/blob/rav/proposals/id_grammar/proposals/1597-id-grammar.md#opaque-ids.

[joepie91]

Whoops, my bad - I didn't realize that that included MXC identifiers as well. I guess this one can be closed, then?

[richvdh]

It's probably useful to keep it open as a specific place to discuss media ids

[uhoreg]

matrix-org/matrix-spec-proposals#1597 (a.k.a. matrix-org/matrix-spec-proposals#1598) was merged as an MSC, so really this is just a spec-pr-missing

[richvdh]

matrix-org/matrix-spec-proposals#1597 (a.k.a. matrix-org/matrix-spec-proposals#1598) was merged as an MSC, so really this is just a spec-pr-missing

as noted over on matrix-org/matrix-spec-proposals#1598, the merge of that MSC dates to a time where MSCs could be merged before implementation, so although matrix-org/matrix-spec-proposals#1597 presents a proposal which seems to have some support, it's never been adopted by the ecosystem.

[richvdh]

some notes on this while I'm in the area:

• synapse's generated media ids consist of 24 letters in the range [a-zA-Z] (https://github.com/matrix-org/synapse/blob/v1.25.0/synapse/rest/media/v1/media_repository.py#L160, https://github.com/matrix-org/synapse/blob/v1.25.0/synapse/util/stringutils.py#L36.
• synapse assumes that media ids do not contain /, # or ?: https://github.com/matrix-org/synapse/blob/v1.25.0/synapse/storage/databases/main/room.py#L663.

[turt2live]

for additional context:

• mmr also makes the same assumptions: https://github.com/turt2live/matrix-media-repo/blob/d38f198b026ac0ac950b13eb1e861f9378e74def/api/webserver/webserver.go#L99-L101
• mmr's media IDs are 64 character alphanumeric strings:
  • https://github.com/turt2live/matrix-media-repo/blob/d38f198b026ac0ac950b13eb1e861f9378e74def/controllers/upload_controller/upload_controller.go#L117
  • https://github.com/turt2live/matrix-media-repo/blob/d38f198b026ac0ac950b13eb1e861f9378e74def/util/random.go#L18-L27

[richvdh]

@neilalexander reports that Dendrite's media ids are

Hex 64 characters I think

Which I assume means that they consist of the characters [0-9a-f].

It'd be interesting to confirm what other media repo impls (eg Conduit) use for their media IDs, but I'd be a bit surprised if they fell outside the range proposed by MSC1597, which, for the record, is:

must be strings consisting entirely of the characters [0-9a-zA-Z.=_-]. Their length must not exceed 255 characters and they must not be empty.

I wonder what we can do to progress this, or what's actually blocking us fixing it.

• Is there any reason we can't simply add an explicit grammar for media IDs to the spec, along the lines of MSC1597's proposal? It doesn't seem like it would be a breaking change in practice. If we did so, should we do a new MSC, or are we happy with MSC1597 to stand in for it?

• If the spec today did give a specific grammar for media IDs, what difference would that make in practice? Would the grammar actually be enforced anywhere, or would it simply be for reference? If it's not enforced, we run the risk of one implementation breaking another by using IDs outside the grammar, but what would enforcement look like?

[KitsuneRal]

AIUI, enforcement might imply that homeservers reject requests to download media with a non-compliant id. That would make the particular media unreachable through normal means but at least clients would be exempted from checking for stray / or an unescaped #.

Ideally, of course, the grammar should be enforced when uploading media; but I have no good ideas here except client applications nagging users to bug homeserver owners if what the client app received for a media id is not compliant.

[richvdh]

per matrix-org/matrix-spec-proposals#1597 (comment), I think allowing = and forbidding ~ is an error.

[turt2live]

oh, and ftr mmr's IPFS media ID is ipfs:someotherid per matrix-org/matrix-spec-proposals#2706