stack safe

[safareli]

fix #9

updated work made in purescript-deprecated/purescript-eff#31

benchmark

bench	type	n	mean	stddev	min	max
>>=R build	Eff	100	43.32 μs	129.40 μs	881.00 ns	411.46 μs
>>=R build	Aff	100	15.08 μs	39.84 μs	1.84 μs	128.39 μs
>>=R run	Eff	100	211.71 μs	198.96 μs	137.53 μs	776.26 μs
>>=R run	Aff	100	354.32 μs	368.47 μs	230.62 μs	1.40 ms
>>=R build	Eff	1000	2.71 μs	2.18 μs	1.51 μs	8.61 μs
>>=R build	Aff	1000	2.54 μs	1.99 μs	1.35 μs	8.16 μs
>>=R run	Eff	1000	1.72 ms	1.14 ms	379.97 μs	3.76 ms
>>=R run	Aff	1000	1.94 ms	1.04 ms	988.37 μs	4.46 ms
>>=R build	Eff	5000	1.65 μs	1.98 μs	893.00 ns	7.24 μs
>>=R build	Aff	5000	1.55 μs	1.53 μs	848.00 ns	5.85 μs
>>=R run	Eff	5000	4.71 ms	2.58 ms	1.33 ms	8.47 ms
>>=R run	Aff	5000	5.84 ms	3.66 ms	2.15 ms	11.92 ms
>>=L build	Eff	100	29.42 μs	87.50 μs	764.00 ns	278.34 μs
>>=L build	Aff	100	2.08 μs	2.01 μs	1.26 μs	7.75 μs
>>=L run	Eff	100	117.08 μs	155.40 μs	63.07 μs	559.29 μs
>>=L run	Aff	100	168.60 μs	20.01 μs	150.76 μs	211.12 μs
>>=L build	Eff	1000	2.16 μs	1.95 μs	1.30 μs	7.59 μs
>>=L build	Aff	1000	1.44 μs	1.70 μs	784.00 ns	6.25 μs
>>=L run	Eff	1000	597.75 μs	241.14 μs	288.23 μs	1.01 ms
>>=L run	Aff	1000	1.48 ms	423.76 μs	989.80 μs	2.39 ms
>>=L build	Eff	5000	1.58 μs	2.03 μs	766.00 ns	7.33 μs
>>=L build	Aff	5000	1.27 μs	1.25 μs	763.00 ns	4.78 μs
>>=L run	Eff	5000	3.89 ms	2.37 ms	1.32 ms	7.66 ms
>>=L run	Aff	5000	4.45 ms	2.94 ms	1.81 ms	10.19 ms
map build	Eff	100	13.74 μs	38.48 μs	730.00 ns	123.11 μs
map build	Aff	100	1.30 μs	1.34 μs	772.00 ns	5.06 μs
map run	Eff	100	143.35 μs	208.15 μs	41.52 μs	722.96 μs
map run	Aff	100	125.66 μs	43.96 μs	91.94 μs	221.56 μs
map build	Eff	1000	1.37 μs	1.58 μs	745.00 ns	5.84 μs
map build	Aff	1000	9.88 μs	24.13 μs	1.11 μs	78.02 μs
map run	Eff	1000	340.58 μs	123.69 μs	214.02 μs	561.88 μs
map run	Aff	1000	712.70 μs	321.60 μs	389.96 μs	1.24 ms
map build	Eff	5000	1.85 μs	1.61 μs	1.20 μs	6.41 μs
map build	Aff	5000	5.35 μs	11.37 μs	1.05 μs	37.25 μs
map run	Eff	5000	1.18 ms	882.39 μs	620.83 μs	3.39 ms
map run	Aff	5000	3.82 ms	2.50 ms	1.63 ms	8.60 ms
apply build	Eff	100	18.52 μs	52.71 μs	754.00 ns	168.43 μs
apply build	Aff	100	9.30 μs	22.66 μs	1.04 μs	73.25 μs
apply run	Eff	100	249.50 μs	188.70 μs	157.87 μs	764.58 μs
apply run	Aff	100	956.74 μs	2.05 ms	204.81 μs	6.79 ms
apply build	Eff	1000	1.95 μs	1.66 μs	1.29 μs	6.65 μs
apply build	Aff	1000	9.34 μs	20.20 μs	1.76 μs	66.05 μs
apply run	Eff	1000	1.00 ms	363.33 μs	569.89 μs	1.62 ms
apply run	Aff	1000	3.81 ms	1.94 ms	1.70 ms	7.48 ms
apply build	Eff	5000	1.32 μs	1.38 μs	741.00 ns	5.22 μs
apply build	Aff	5000	4.38 μs	7.95 μs	1.03 μs	26.50 μs
apply run	Eff	5000	4.23 ms	1.94 ms	2.23 ms	7.67 ms
apply run	Aff	5000	11.20 ms	2.57 ms	9.25 ms	18.19 ms
-	-	-	-	-	-	-
>>=R build	Eff	10000	3.33 μs	2.78 μs	1.36 μs	5.29 μs
>>=R build	Aff	10000	34.65 μs	30.98 μs	12.75 μs	56.56 μs
>>=R run	Eff	10000	7.57 ms	3.30 ms	5.24 ms	9.91 ms
>>=R run	Aff	10000	17.67 ms	41.47 μs	17.64 ms	17.70 ms
>>=R build	Eff	50000	8.73 μs	8.88 μs	2.46 μs	15.01 μs
>>=R build	Aff	50000	24.72 μs	22.66 μs	8.70 μs	40.74 μs
>>=R run	Eff	50000	109.05 ms	36.31 ms	83.38 ms	134.73 ms
>>=R run	Aff	50000	204.59 ms	73.84 ms	152.38 ms	256.80 ms
>>=R build	Eff	100000	19.40 μs	9.01 μs	13.03 μs	25.77 μs
>>=R build	Aff	100000	81.61 μs	10.48 μs	74.20 μs	89.03 μs
>>=R run	Eff	100000	306.49 ms	127.54 ms	216.31 ms	396.67 ms
>>=R run	Aff	100000	419.11 ms	81.90 ms	361.20 ms	477.02 ms
>>=R build	Eff	1000000	6.00 μs	5.19 μs	2.33 μs	9.67 μs
>>=R build	Aff	1000000	34.99 μs	38.85 μs	7.52 μs	62.47 μs
>>=R run	Eff	1000000	4.13 s	2.18 s	2.59 s	5.67 s
>>=R run	Aff	1000000	4.28 s	573.69 ms	3.88 s	4.69 s
>>=L build	Eff	10000	6.52 μs	6.44 μs	1.97 μs	11.08 μs
>>=L build	Aff	10000	33.31 μs	35.44 μs	8.25 μs	58.36 μs
>>=L run	Eff	10000	22.02 ms	4.88 ms	18.56 ms	25.47 ms
>>=L run	Aff	10000	30.54 ms	18.12 ms	17.73 ms	43.36 ms
>>=L build	Eff	50000	4.70 μs	3.77 μs	2.03 μs	7.36 μs
>>=L build	Aff	50000	23.86 μs	19.90 μs	9.78 μs	37.93 μs
>>=L run	Eff	50000	110.11 ms	31.00 ms	88.20 ms	132.03 ms
>>=L run	Aff	50000	242.86 ms	92.22 ms	177.65 ms	308.07 ms
>>=L build	Eff	100000	11.27 μs	7.01 μs	6.31 μs	16.22 μs
>>=L build	Aff	100000	42.01 μs	48.56 μs	7.67 μs	76.34 μs
>>=L run	Eff	100000	324.61 ms	107.09 ms	248.89 ms	400.34 ms
>>=L run	Aff	100000	447.80 ms	86.00 ms	386.99 ms	508.62 ms
>>=L build	Eff	1000000	7.66 μs	8.06 μs	1.96 μs	13.36 μs
>>=L build	Aff	1000000	70.69 μs	77.81 μs	15.67 μs	125.72 μs
>>=L run	Eff	1000000	3.18 s	774.65 ms	2.64 s	3.73 s
>>=L run	Aff	1000000	11.36 s	8.53 s	5.33 s	17.40 s
map build	Eff	10000	7.29 μs	5.64 μs	3.31 μs	11.28 μs
map build	Aff	10000	46.05 μs	46.44 μs	13.21 μs	78.89 μs
map run	Eff	10000	5.24 ms	217.28 μs	5.08 ms	5.39 ms
map run	Aff	10000	15.01 ms	12.58 ms	6.12 ms	23.91 ms
map build	Eff	50000	4.32 μs	3.01 μs	2.19 μs	6.45 μs
map build	Aff	50000	16.23 μs	14.19 μs	6.20 μs	26.26 μs
map run	Eff	50000	14.28 ms	2.40 ms	12.58 ms	15.97 ms
map run	Aff	50000	61.19 ms	23.59 ms	44.51 ms	77.87 ms
map build	Eff	100000	4.27 μs	3.30 μs	1.93 μs	6.60 μs
map build	Aff	100000	40.82 μs	42.18 μs	11.00 μs	70.65 μs
map run	Eff	100000	51.75 ms	1.39 ms	50.77 ms	52.73 ms
map run	Aff	100000	175.20 ms	15.22 ms	164.44 ms	185.96 ms
map build	Eff	1000000	5.06 μs	3.95 μs	2.27 μs	7.86 μs
map build	Aff	1000000	28.48 μs	26.48 μs	9.75 μs	47.21 μs
map run	Eff	1000000	1.30 s	619.66 ms	857.32 ms	1.73 s
map run	Aff	1000000	2.00 s	305.70 ms	1.78 s	2.21 s
map build	Eff	350000	6.09 μs	6.00 μs	1.84 μs	10.33 μs
map build	Aff	350000	75.70 μs	18.64 μs	62.52 μs	88.88 μs
map run	Eff	350000	324.03 ms	61.57 ms	280.49 ms	367.57 ms
map run	Aff	350000	644.66 ms	69.64 ms	595.41 ms	693.90 ms
map build	Eff	700000	4.85 μs	4.71 μs	1.52 μs	8.18 μs
map build	Aff	700000	53.11 μs	50.24 μs	17.59 μs	88.64 μs
map run	Eff	700000	607.53 ms	155.39 ms	497.65 ms	717.41 ms
map run	Aff	700000	1.76 s	620.00 ms	1.32 s	2.20 s
apply build	Eff	10000	5.55 μs	5.87 μs	1.40 μs	9.70 μs
apply build	Aff	10000	114.71 μs	118.45 μs	30.95 μs	198.46 μs
apply run	Eff	10000	9.93 ms	2.98 ms	7.82 ms	12.03 ms
apply run	Aff	10000	59.19 ms	131.00 μs	59.10 ms	59.28 ms
apply build	Eff	50000	5.21 μs	4.50 μs	2.03 μs	8.39 μs
apply build	Aff	50000	43.94 μs	46.97 μs	10.73 μs	77.15 μs
apply run	Eff	50000	143.75 ms	57.67 ms	102.97 ms	184.52 ms
apply run	Aff	50000	384.62 ms	41.34 ms	355.39 ms	413.86 ms
apply build	Eff	100000	6.90 μs	7.03 μs	1.93 μs	11.86 μs
apply build	Aff	100000	59.84 μs	25.76 μs	41.62 μs	78.05 μs
apply run	Eff	100000	335.42 ms	67.59 ms	287.63 ms	383.21 ms
apply run	Aff	100000	846.18 ms	192.05 ms	710.38 ms	981.98 ms
apply build	Eff	1000000	7.54 μs	7.68 μs	2.11 μs	12.97 μs
apply build	Aff	1000000	55.95 μs	62.42 μs	11.81 μs	100.09 μs
apply run	Eff	1000000	7.13 s	4.00 s	4.30 s	9.95 s
apply run	Aff	1000000	25.64 s	1.79 s	24.37 s	26.90 s

[safareli]

This is worth to still have in mind purescript-deprecated/purescript-eff#31 (comment)

[safareli]

💀 this should be updated before merge

[safareli]

In this PR I have added some explanation of how FFI part works.
If something needs more clarification, or if you spot a typo let me know/comment.

[safareli]

There might be concern about portability of code relying on stack safety of Effect. For that to not be an issue other backends should make there Effect implementation stack safe too. The implementation here is not too complex, but it preserves original Eff runtime representation by using JS function as an object. I'm not sure if functions can have properties in c++ or Go or rlang like in JS. If all this languages support similar feature then this implementation can be ported and runtime representation preserved. otherwise if stack safety is desired in other backends, just a regular tree like representation should be used with some unsafePerformEffect like function which interprets the tree.

[safareli]

ping

[natefaubion]

While I do think this is really cool and useful, I don't think this is something that we can merge as part of core Effect and say that it is stack safe. It is only stack safe if the compiler does not optimize it, which is a very strange/unintuitive condition, and I think is a showstopper for including it as part of the core library. I'm inclined to say that this should exist as a separate library with a separate opaque type with a MonadEffect instance.

[safareli]

That strange/unintuitive condition can be "fixed" by removing that compiler optimization. So choice is between having compiler optimization and no stack safety OR removing compiler optimization and having library level stack safety. But as this PR was hanging around for over 2 years, I guess stack safety at the cost of the compiler optimization is not acceptable so closing.

[natefaubion]

I do not consider removing the optimization acceptable. It's the only reason why things like halogen-vdom can even be written with acceptable performance.