19:06:31 <vtnerd> spirobel: 'doubt btw that any of this is a bottleneck" => there is at least a 11% improvement to lws when switching to binary from json
19:06:52 <vtnerd> I made a mistake when running the perf tests -> I ran expand_transaction_1 when it could be skipped
19:07:06 <vtnerd> mind you this was a test with msgpack+epee tx compact (which is the fastest possible)
19:08:20 <vtnerd> and this was the total scan time, not the unpack time diff
19:09:52 <vtnerd> 'monero adoption is this obscure epee stuff binary endpoints for both parameters' => yes ZMQ has JSON fields now that are exactly what you want. we can probably add this to some of the http endpoints, but its a bit funky because the two systems use different output systems
19:11:28 <vtnerd> the problem with the ZMQ style is that the JSON output requires unpacking the tx from its internal format, and then writing out each component piece-by-piece
19:11:54 <vtnerd> this slows down `monerod` responses by a decent amount, but I forget the overhead cost
19:12:40 <vtnerd> removing expand_transactions_1 from the pathway would help, so the diff maybe somewhat small (but JSON is definitely slower)