Avoid O(n^2) time broadcast' by replacing Writer

[robbert-vdh]

As mentioned by @fatho in #102, the Writer monad can be replaced with the strict state monad to avoid the use for mappend/list concatenation for every element added to the list.. This brings the time complexity down from O(n^2) to O(n). I added the reverse so the order stays the same before and after this change. If that's not needed the reverse can also be removed. Then order of the lists in the tests would then need to be changed.

I thought about removing the monad entirely since it's only used for keeping track of the list and the return value is never used, but an alternative version that still runs on O(n) time would require passing the list around directly which gets very messy.

Resolves #102.

[diegodiv]

The code looks ok to me but how did you test it?

[robbert-vdh]

I thought the SubscriptionTree tests already tested the broadcast' function directly. I was going to add a regression test to make sure the order doesn't change but if it did this test would fail:

icepeak/server/tests/Icepeak/Server/SubscriptionTreeSpec.hs

Lines 109 to 114 in c204311

	it "notifies parents and children about updates" $ do
	broadcast'' ["foo", "bar"]
	`shouldBe` [ (conn1, value)
	, (conn2, value_foo)
	, (conn3, value_foo_bar)
	]

But, there's a sneaky sort right here:

icepeak/server/tests/Icepeak/Server/SubscriptionTreeSpec.hs

Line 92 in aa3320b

broadcast'' path = sortOn fst $ broadcast' path value root

I added another commit before this one that removes the sort so the order is also tested (if you change the orders in broadcast' it will break the test). The CI won't run for that commit but the tests are all green for both commits!

[diegodiv]

LGTM! 🚀

[diegodiv]

Nice catch!

[robbert-vdh]

@OpsBotPrime merge

[OpsBotPrime]

Pull request approved for merge by @robbert-vdh, rebasing now.

[OpsBotPrime]

Rebased as a24b517, waiting for CI …

[OpsBotPrime]

CI job 🟡 started.