Implement Read Marker API

[lukebarnard1]

This implements the design at https://docs.google.com/document/d/1UWqdS-e1sdwkLDUY0wA4gZyIkRp-ekjsLZ8k6g_Zvso/edit#heading=h.8jqdhkapw60, summarised below.

Update: see #2128 for further alterations. The summary below has been updated.

m.read_marker event

{
    "content": {
        "event_id”: “$123456789098765hhjGslf:matrix.org"
    },
    "room_id": "!someroomid:matrix.org",
    "type": "m.read_marker"
}

To update it, the client sends a PUT.

RESTful HTTP endpoint

PUT /_matrix/client/r0/rooms/%21someroomid%3Amatrix.org/read_markers HTTP/1.1
Content-Type: application/json

{
	"m.read": "$347926718ewsc:example.com",
	"m.read_marker": "$4536432343kljf:example.com"
}

The REST API hits:

• the RR handler if “m.read” has been set. This is optional and means that we don’t end up doing one call for RM and one separate call for RR, which is the common case when the user has caught up on their scrollback;
• the RM handler if “m.read_marker” has been set. This handler asserts that the RM event ID is ahead in the stream in comparison to the current one and then uses the room account data API to store the m.read_marker event.

[erikjohnston]

Should be: # Copyright 2017 Vector Creations Ltd

[erikjohnston]

Just use a tuple: (room_id, user_id)

[erikjohnston]

This should be a function inside the storage function. Conventionally functions that start with _ are private.

This also should have a desc="..." field so that that metrics work correctly. Usually its just the name of the calling function

[erikjohnston]

I'd move this entire block of code into the storage function and have a dedicated is_event_after function or some such

[erikjohnston]

Also, you upset pep8:

$ flake8 synapse tests 
synapse/handlers/read_marker.py:20:1: F401 'synapse.util.logcontext.PreserveLoggingContext' imported but unused
synapse/handlers/read_marker.py:22:1: F401 'synapse.types.get_domain_from_id' imported but unused
synapse/handlers/read_marker.py:28:1: E302 expected 2 blank lines, found 1
synapse/rest/client/v2_alpha/account_data.py:87:17: E128 continuation line under-indented for visual indent
synapse/rest/client/v2_alpha/account_data.py:88:17: E128 continuation line under-indented for visual indent
synapse/rest/client/v2_alpha/account_data.py:89:13: E124 closing bracket does not match visual indentation
synapse/rest/client/v2_alpha/read_marker.py:18:1: F401 'synapse.api.errors.SynapseError' imported but unused
synapse/rest/client/v2_alpha/read_marker.py:48:43: E703 statement ends with a semicolon

[lukebarnard1]

Actually, it turns out that there's a bug if you haven't set a RM for a room previously. In the handler on line 46 it assumes that there is a m.read_marker key.

[lukebarnard1]

Fixed!

[erikjohnston]

Outdated comment?

[erikjohnston]

We probably should be using a storage function that pulls out based on type too. I think there already is one?

[lukebarnard1]

There's one for global_account_data but not for_room

[erikjohnston]

This could also be written as:

existing_read_marker = account_data.get("m.read_marker", None)

[erikjohnston]

We don't need to notify the entire room about this change, just the user.

[erikjohnston]

For future reference we tend to put spaces at the start:

"Cannot set m.read_marker through this API."
" Use /rooms/!roomId:server.name/read_marker"

This makes it more obvious that there is at least one space, which is quite important for e.g. SQL

[erikjohnston]

Try to avoid taking a reference to hs if possible, and instead just pull out everything you need. This is for cleanliness and makes it clearer what the dependencies of this class is.

[erikjohnston]

I usually do:

read_event_id = body.get("m.read", None)
if read_event_id:
    ...

but its a matter of taste

[erikjohnston]

I think you can actually write this as:

is_after = (to_1, so_1) > (to_2, so_2)

[lukebarnard1]

Sadly,

>>> ( 1, 2 ) < ( 1, 2 )
False
>>> ( 1, 2 ) < ( 1, 3 )
True

So we'll have to go for the form with two comparisons. is_event_after is now also an accurate name with this change...

[erikjohnston]

That's exactly what you want though, isn't it?

[erikjohnston]

Just a few nits, nothing particularly important. Feel free to ignore the various style comments depending on your taste

[erikjohnston]

This is not equivalent to (to_1, so_1) > (to_2, so_2)?

[lukebarnard1]

in IRL we disproved my lovely boolean algebra; it had an error in it. sads.

[erikjohnston]

lgtm