Richard van der Hoff

this is a different problem to vector-im/element-web#5601, and vector-im/element-meta#347 is *completely* unrelated.

this is related to https://github.com/matrix-org/matrix-doc/pull/2654, though I don't think it will do much to change the situation here.

on Linux, you only need the development libraries (hence the recommendations to install `libpq-devel` in some sections). Is there nothing similar in the macOS world?

> We should probably be requiring or allowing psycopg2-binary instead. This seems to be a mess - cf https://www.psycopg.org/articles/2018/02/08/psycopg-274-released/ and https://github.com/psycopg/psycopg2/issues/674. To cut a long story short: use of `psycopg2-binary`...

> Do we need partial-state versions of /state_ids and /state too? I think a partial-state version of `/state_ids` is the empty set? But yes, we can't just rely on the...

(of course, faster backfill is not inherently tied to faster *joins* - it's extending the "partial state" concept beyond joins and into backfill. As if it weren't already complicated enough.)

This may end up being intertwined with other work on backfill which seeks to reduce the amount of state we need to download, such as https://github.com/matrix-org/synapse/issues/13618.

`frontend_proxy` is already totally redundant: there is no difference between it and `generic_worker`. The only thing left is to remove it from the documentation, which I hope #13451 will do....

cf https://github.com/matrix-org/synapse/issues/13476

> What happens if I decide (because of mentioned risk) to remove retention: enabled: true from configuration and many rooms in my HS have retention configured? Is it safe? I...