Add base Env monad

[Jason94]

This PR adds a base Env :env monad to environment.lisp. Previously Env :env was just an alias for EnvT :env Identity. Haskell defines Reader like this, and they can get away with it. GHC does a lot of work behind the scenes to fuse lambda calls, so wrapping the Identity monad is zero-cost at runtime for them. Coalton doesn't have that, so it's not.

Here's a little microbenchmark that takes an Integer as the environment and a list of ints as an argument, and maps adding the environment Integer to all of the ints in the list inside the monad.( It does this very inefficiently, so it binds for every element in the list.)

(cl:in-package :cl-user)
(defpackage :bench-env
  (:use
   #:coalton
   #:coalton-prelude
   #:coalton-library/monad/environment
   #:coalton-library/monad/identity
   )
  )

(in-package :bench-env)

(named-readtables:in-readtable coalton:coalton)

(coalton-toplevel
  (define context 100)

  (declare add-env (MonadEnvironment Integer :m => List Integer -> :m (List Integer)))
  (define (add-env ints)
    (rec % ((rem ints)
            (accum Nil))
      (match rem
        ((Nil) (pure (reverse accum)))
        ((Cons x rem)
         (do
          (n <- ask)
          (% rem (Cons (+ n x) accum)))))))


  (declare test-env (List Integer -> Unit))
  (define (test-env ints)
    (run-env (add-env ints) context)
    Unit)

  (declare test-envT (List Integer -> Unit))
  (define (test-envT ints)
    (run-identity (run-envT (add-env ints) context))
    Unit)

  (declare test-both (Unit -> Unit))
  (define (test-both)
    (let ints = (range 0 10000000))
    (lisp :a (ints)
      (cl:time (coalton (test-env (lisp :a () ints)))))
    (lisp :a (ints)
      (cl:time (coalton (test-envT (lisp :a () ints)))))
    Unit)
  )

Here are typical results on my machine, compiled in release mode:

Evaluation took:
  1.186 seconds of real time
  1.182656 seconds of total run time (1.007186 user, 0.175470 system)
  [ Real times consist of 0.547 seconds GC time, and 0.639 seconds non-GC time. ]
  [ Run times consist of 0.543 seconds GC time, and 0.640 seconds non-GC time. ]
  99.75% CPU
  4,382,138,882 processor cycles
  1,919,869,824 bytes consed
  
Evaluation took:
  1.226 seconds of real time
  1.220518 seconds of total run time (1.082556 user, 0.137962 system)
  [ Real times consist of 0.356 seconds GC time, and 0.870 seconds non-GC time. ]
  [ Run times consist of 0.357 seconds GC time, and 0.864 seconds non-GC time. ]
  99.59% CPU
  4,527,260,615 processor cycles
  2,719,164,848 bytes consed

The base Env runs ~3.26% faster than the EnvT' which actually surprised me that it was so close. I'm not good enough at optimizing CL/Coalton to really know why that might be, other than speculating that the EnvT & Identity monads really don't do that much.
The base Env did cons ~29% fewer bytes than the EnvT version did, which is in line with what I would have expected. Also curious, the base version actually spent more time in GC, but I suspect that's just noise or a bad test setup on my part.

[Jason94]

I re-ran the benchmark with 12M iterations, in the SBCL console. Previously I was testing in Sly inside Emacs, which I noticed was leading to some funky freezes. The results were much closer to what I expected. I also tested with the heuristic inline on/off, and it made a very small but possibly statistically significant speed-up turning it on. (I'm not surprised the automatic inliner didn't have much impact, because the monad code is aggressively (inline) annotated.)

Test – 12,000,000 iterations

Expand benchmark table

test	evaluation_index	real_time_seconds	gc_time_seconds	non_gc_time_seconds	processor_cycles	bytes_consed	avg_real_time (s)
test-env	1	1.449	0.7	0.749	5350510319	2303851120
test-env	2	1.022	0.347	0.675	3773787805	2303875024	1.00344444444444
test-env	3	1.073	0.418	0.655	3962464643	2303851152
test-env	4	0.997	0.357	0.64	3679401138	2303842304
test-env	5	0.963	0.311	0.652	3557912970	2303883904
test-env	6	1.05	0.403	0.647	3876228632	2303851248
test-env	7	0.966	0.326	0.64	3566461968	2303875072
test-env	8	0.977	0.336	0.641	3609443166	2303851216
test-env	9	0.982	0.34	0.642	3626700521	2303907792
test-env	10	1.001	0.345	0.656	3693185784	2303818496
test-envT	1	1.39	0.477	0.913	5133118632	3263214272
test-envT	2	1.306	0.377	0.929	4824995212	3263189584	1.30822222222222
test-envT	3	1.323	0.399	0.924	4883256152	3263199104
test-envT	4	1.278	0.363	0.915	4720169883	3262978960
test-envT	5	1.371	0.46	0.911	5063699824	3262997184
test-envT	6	1.307	0.383	0.924	4825740540	3262997280
test-envT	7	1.318	0.388	0.93	4867550873	3263224528
test-envT	8	1.271	0.356	0.915	4692395315	3262997216
test-envT	9	1.305	0.397	0.908	4817381833	3262978928
test-envT	10	1.295	0.377	0.918	4784951222	3263030016
test-env (inliner disabled)	1	1.473	0.734	0.739	5442874825	2303851280
test-env (inliner disabled)	2	1.04	0.375	0.665	3841716845	2303851216	1.02255555555556
test-env (inliner disabled)	3	0.959	0.324	0.635	3542639888	2303875008
test-env (inliner disabled)	4	1.078	0.44	0.638	3979465403	2303851216
test-env (inliner disabled)	5	0.991	0.345	0.646	3660430017	2303875040
test-env (inliner disabled)	6	0.999	0.357	0.642	3689732241	2303883920
test-env (inliner disabled)	7	0.998	0.36	0.638	3687453041	2303875056
test-env (inliner disabled)	8	1.085	0.369	0.716	4005687118	2303851152
test-env (inliner disabled)	9	1.083	0.427	0.656	3998785915	2303842272
test-env (inliner disabled)	10	0.97	0.326	0.644	3584094207	2303879520
test-envT (inliner disabled)	1	1.348	0.411	0.937	4980579509	3263199056
test-envT (inliner disabled)	2	1.344	0.416	0.928	4961919891	3262978960	1.35755555555556
test-envT (inliner disabled)	3	1.283	0.359	0.924	4736679357	3262997184
test-envT (inliner disabled)	4	1.34	0.41	0.93	4949700789	3262997280
test-envT (inliner disabled)	5	1.341	0.408	0.933	4950213831	3263224528
test-envT (inliner disabled)	6	1.285	0.367	0.918	4746497825	3262997216
test-envT (inliner disabled)	7	1.321	0.398	0.923	4881190331	3262978928
test-envT (inliner disabled)	8	1.47	0.42	1.05	5428045965	3263030016
test-envT (inliner disabled)	9	1.531	0.538	0.993	5653814711	3263151888
test-envT (inliner disabled)	10	1.303	0.391	0.912	4809611574	3262978960

The first row looked like it was an outlier (possibly because of some processor or SBCL warm-up time, but I honestly don't know how that works at that level.) Removing the first row of each test, the `Env` code was ~23.3% faster than the `EnvT :e Identity` code, which is in line with what I expected.