Add Accum effect and carriers

[turion]

This is still work in progress, but I'm happy for some guidance. There are a few places where I've just tried to follow the conventions, but I'm not sure whether it makes sense.

[turion]

• Tests
• Does an example make sense?
• Should I add benchmarks?

[turion]

Fixes #389

[robrix]

Sorry I’m late on this one, but thanks for the PR!

[robrix]

This is a fantastic start, thank you @turion!

[robrix]

@turion: Heads up that I’m about to release v1.1, which means that the @since annotations will all be 1.1.1.

[turion]

@robrix I noticed that the monoid instance for W in the tests is Integer, which is commutative. That makes the tests less sensitive because it doesn't detect the order in which things are added to monoidal logs. Does it make sense to generalise it somewhat, e.g. to Maybe Integer?

[turion]

@robrix I cannot figure out why this one test fails. It claims that the AccumT carrier isn't lawful, but I can't figure out why.

[turion]

Ah. I did indeed test a property that isn't fulfilled by the transformers AccumT. In runAccumT :: AccumT w m a -> w -> m (a, w), the given w is not appended to the resulting log.

[turion]

I think it doesn't make sense to base Accum carriers on State carriers (like it is done for Writer). Accum is fundamentally different.

[turion]

For some reason, the test suite often hangs. I wonder whether I introduced a bug in mfix or so.

[turion]

I haven't found the mistake in mfix yet, but it's a bit cumbersome to implement mfix for continuation monads. I'll have a look at the StateC implementation tomorrow and try again.

Also, I haven't fixed the expected law yet, but looking at the test it seems it's going to be simple.

[turion]

Hacking on fused effects is a lot of fun, but maintaining compatibility with older GHC versions is not :/ I usually drop older versions as soon as the most idiomatic code for newer versions doesn't compile there anymore

[robrix]

Hacking on fused effects is a lot of fun, but maintaining compatibility with older GHC versions is not :/ I usually drop older versions as soon as the most idiomatic code for newer versions doesn't compile there anymore

Unfortunately, we don’t have a good way to gauge what our users’ needs are here. We’ve discussed this in #185, but I don’t feel like we came to any really solid answer. For the moment I’d prefer not to drop back-compat for changes that would otherwise be a minor version bump unless it’s completely intractable to maintain it (and even then, let’s chat about it some more first 😊).

[turion]

Hacking on fused effects is a lot of fun, but maintaining compatibility with older GHC versions is not :/ I usually drop older versions as soon as the most idiomatic code for newer versions doesn't compile there anymore

Unfortunately, we don’t have a good way to gauge what our users’ needs are here. We’ve discussed this in #185, but I don’t feel like we came to any really solid answer. For the moment I’d prefer not to drop back-compat for changes that would otherwise be a minor version bump unless it’s completely intractable to maintain it (and even then, let’s chat about it some more first blush).

I meant it more like this: My motivation to contribute with some new, interesting features is high, but it is very low to figure out how to fix the 8.2.2 pipeline, because I need to chase an uninteresting bug with a compiler version that is different from the rest of my stack.

[turion]

Side question: Is there any way to run only a specific set of tests in hedgehog?

[turion]

@robrix Your suggestion to make the lambda in mfix was great. Unfortunately I had the wrong mfix implementation in the first place, but I've fixed that now.

[turion]

The only thing left is fixing older GHC versions. Let's see how the pipelines work out.

[turion]

Ping @robrix

[patrickt]

Can you merge master into this? I’d like to add it as part of 1.1.2.0. (Apologies for the long wait!)

[robrix]

Approved CI. Let the good times roll!

[robrix]

@turion: I'm terribly sorry this has taken me so long to get back to, and I'm so grateful for your patience. You've done a fantastic job here, and I'm delighted to add it to the library!

[robrix]

@turion Looks like you'll need to import Data.Semigroup, or else use mappend. I think we do something like this elsewhere, so a quick grep might help.

Edit: the tests for Cut are one such example: we import Data.Semigroup as S and then use S.<> in the body of the file. Using the symbol qualified prevents us from triggering warnings about a redundant import.

[robrix]

@turion: Apologies for hijacking your PR, I just couldn't wait to see CI go green ✅ 😀

[robrix]

Ah, I guess we need to CPP-guard (part of) the tests, too.

[robrix]

@turion: I'm really feeling your earlier point about supporting older versions of ghc 😅

[robrix]

It's green! It's green! Quick, merge before CI notices!!

[turion]

Wow, the last two hours apparently were an exciting ride!

[turion]

@turion: I'm really feeling your earlier point about supporting older versions of ghc sweat_smile

Yeah, I respect your effort to go all the way down to 8.2 a lot. In my hobby projects (which are less used than fused-effects) I typically discard an old GHC version in the moment I would have to use CPP otherwise.

[turion]

Hah, it seems you actually agree in principle 😆 #185 (comment)