we should optimise rest endpoint dispatch

[richvdh]

every incoming GET request gets compared against a list of 227 regexps, many of which are redundant since they duplicate each endpoint onto the various places the client API lives. This is surely an unnecessary overhead.

[erikjohnston]

Something like https:/nitely/kua would probably work, though its py3 only.

[richvdh]

possibly, though I'm not sure how it would tie into twisted.

I was mostly just imagining that it would be a good start to say: for the ClientRestResource which is mounted at /_matrix/client/r0, only register regexes which will actually match that prefix, rather than all of the /_matrix/client/api/v1 etc regexes.

[clokep]

every incoming GET request gets compared against a list of 227 regexps

I was a bit curious how this compares to current day:

Method	Endpoints
GET	232
PUT	129
POST	169
DELETE	45

I was pretty easily able to reduce this quite a bit by messing with the client_patterns to create a single regular expression instead of multiple patterns. I'm not sure if that's a worthwhile optimization or if we need an entire new strategy. Anyway the results were:

Method	Endpoints
GET	80
PUT	39
POST	66
DELETE	15

(These were for the /_matrix/client endpoints using the demo server. so exact endpoints could vary a bit based on configuration.)