There are still two more of these that don't show up in openapi.json
because they don't show up in HTTP routes (only in the websocket):
* GameUpdateEvents
* FrozenCombat
I'm not sure if there's a way to forcibly include those in the
openapi.json, if I should add a no-op API to force it to happen, or if I
should just ignore it. For now I'm going with option 3.
operation_ids give us better function names when generating the
typescript API from the openapi.json. BaseModel.Config.title does the
same for type names. Response models (or 204 status codes) need to be
explicit or the API will be declared as returning any.
The React UI running in a browser can't connect to the backend without
punching a hole for CORS, which isn't done by default. We don't need the
API key to protect from browsers, and anything else running on the
user's machine that can access the backend (that's hosted on only
localhost) already has enough control to do damage without using
Liberation as an attack vector.
https://github.com/dcs-liberation/dcs_liberation
https://stackoverflow.com/a/44888113/632035
Pickle can't deal with loading sets (and probably dicts) of objects with
custom __hash__ functions that depend on their state because __hash__
can be called before __setstate__. Make the hash function stupider (but
still correct) by just relying on the object ID.
This may not be the way to do this long term, but it is how the old map
works so it's at least not a regression. It might be better to generate
events for the between-turn changes in state instead.
https://github.com/dcs-liberation/dcs_liberation/issues/2039
This automatically logs state changes (in a readable form) to the js
console. We'll probably want to turn this off in production to cut down
on log noise.
The UI needs to be able to identify these to the server and vice versa,
so they'll need IDs that don't change. Rather than constructing an ID
based on the control points names, make them an owned part of the
control point. The constructed ID would be fine, but a UUID will make
them more suitable for the database, and this was always fairly gross
anyway.
Some follow up work if anyone is interested: a bunch of the data that's
computed in the various properties can now probably be computed *once*
and persisted to the FrontLine type.
