Add evalState(T)

[andreasabel]

Fix #404: add evalState and evalStateT.

Use case: stateful interpreter (eval) for C-like expressions e:

let Ur value = HashMap.empty capacity (\ env -> move (evalState (eval e) env))

This pattern breaks the "jail" set by the type signature of HashMap-allocation:

empty :: Int -> (HashMap k v %1 -> Ur b) %1 -> Ur b

HashMap is not Movable but Consumable, so we can get rid of it before moveing the result to Ur.

PS: Your ormolu CI workflow is quite a gatekeeper, running for 12min... I tried to run it locally after stack install ormolu, but this didn't produce the same formatting.

$ ormolu --version
ormolu 0.3.1.0 UNKNOWN UNKNOWN
using ghc-lib-parser 9.0.2.20211226

[aspiwack]

The CI doesn't run for 12 minutes, fortunately (that would be more than a little bug). It was just waiting for someone with push rights to approve the CI. This is a Github thing, to prevent people from using pull requests as a free computing resource (typically to mine bitcoin or the like).

[andreasabel]

The CI doesn't run for 12 minutes,

Maybe I was unlucky? Maybe it does so until it succeeds for the first time (caching logic):

[aspiwack]

This is a little concerning. I assume you've needed that?

[andreasabel]

Yes, otherwise cabal build would fail. I am using GHC 9.2.2.
I think this restriction can be lifted if the underlying problem is fixed:

Building test suite 'test' for linear-base-0.2.0..
[1 of 9] Compiling Test.Data.Destination ...

test/Test/Data/Destination.hs:22:7: error: [-Wdeprecations, -Werror=deprecations]
    In the use of ‘testProperty’ (imported from Test.Tasty.Hedgehog):
    Deprecated: "testProperty will cause Hedgehog to provide incorrect instructions for re-checking properties"
   |
22 |     [ testProperty "alloc . mirror = id" roundTrip,
   |       ^^^^^^^^^^^^

This deprecation was introduced in tasty-hedgehog-1.2.

[andreasabel]

Added a comment to the .cabal file to explain why this bound is needed.

[aspiwack]

Ah now I see the problem with #405 . I didn't remember we had this version of fst available as well.

[aspiwack]

The CI doesn't run for 12 minutes,

Maybe I was unlucky? Maybe it does so until it succeeds for the first time (caching logic):

Ah yes… we've been needing to provide Ormolu via stack, so it needs to be built once which probably takes a little bit of time. I wonder how the cache logic works on forks… maybe that's the rub…

Please let me know if it happens again next time you make a PR.

[treeowl]

How does this help the HashMap situation? We have no way to ensure that a HashMap doesn't escape through the "context" of the monad.

[andreasabel]

@treeowl wrote:

How does this help the HashMap situation? We have no way to ensure that a HashMap doesn't escape through the "context" of the monad.

Er, I cannot follow. Maybe you had some prior discussion of this issue? Do you have a pointer to it?

Atm, the continuation of HashMap.empty has to live in Ur b, do you suggest bad things happen if b is instantiated with, say, IO a?

[aspiwack]

(I'll get to merging this soon, I haven't had time so far, but it will be merged, sorry about the delay)

[aspiwack]

Thanks @andreasabel . Let me merge.