1.0.0 #1

safareli · 2017-12-12T23:56:28Z

TODO

check if it's better to run operations on pure values directly add more optimization for pure node operations purescript-contrib/purescript-aff#133
performance issue on large number of nodes might be caused by v8 changing representation of js array from c array to hashmap after array becomes too big.

safareli · 2017-12-13T00:04:46Z

npm run bench:

bench	type	n	mean	stddev	min	max
bind assocR build	Eff	100	945.04 ns	10.50 μs	230.00 ns	319.94 μs
bind assocR build	Ef	100	548.83 ns	5.81 μs	273.00 ns	201.78 μs
bind assocR run	Eff	100	83.45 μs	81.41 μs	57.24 μs	924.19 μs
bind assocR run	Ef	100	51.51 μs	22.02 μs	43.43 μs	566.60 μs
bind assocR build	Eff	500	352.04 ns	1.33 μs	242.00 ns	46.88 μs
bind assocR build	Ef	500	524.96 ns	3.16 μs	282.00 ns	95.86 μs
bind assocR run	Eff	500	414.97 μs	298.30 μs	300.30 μs	2.93 ms
bind assocR run	Ef	500	238.37 μs	47.15 μs	213.71 μs	1.06 ms
bind assocR build	Eff	1000	391.90 ns	1.58 μs	283.00 ns	58.36 μs
bind assocR build	Ef	1000	514.28 ns	2.77 μs	281.00 ns	71.48 μs
bind assocR run	Eff	1000	881.69 μs	554.14 μs	600.23 μs	4.13 ms
bind assocR run	Ef	1000	479.24 μs	105.77 μs	424.65 μs	1.99 ms
bind assocR build	Eff	2000	369.02 ns	1.57 μs	242.00 ns	50.18 μs
bind assocR build	Ef	2000	533.73 ns	3.87 μs	280.00 ns	107.84 μs
bind assocR run	Eff	2000	2.02 ms	1.05 ms	1.20 ms	7.28 ms
bind assocR run	Ef	2000	1.00 ms	254.26 μs	848.04 μs	3.75 ms
bind assocR build	Eff	4000	365.82 ns	1.64 μs	241.00 ns	51.16 μs
bind assocR build	Ef	4000	484.45 ns	3.33 μs	276.00 ns	96.11 μs
bind assocR run	Eff	4000	4.91 ms	2.43 ms	2.50 ms	22.13 ms
bind assocR run	Ef	4000	2.29 ms	869.33 μs	1.69 ms	8.49 ms
bind assocR build	Eff	8000	414.06 ns	2.12 μs	240.00 ns	68.76 μs
bind assocR build	Ef	8000	515.46 ns	3.27 μs	281.00 ns	109.83 μs
bind assocR run	Eff	8000	38.28 ms	12.09 ms	9.59 ms	115.95 ms
bind assocR run	Ef	8000	6.56 ms	1.85 ms	4.30 ms	20.08 ms
bind assocR build	Eff	10000	355.31 ns	1.97 μs	240.00 ns	76.57 μs
bind assocR build	Ef	10000	573.82 ns	3.71 μs	311.00 ns	127.65 μs
bind assocR run	Eff	10000	79.64 ms	22.74 ms	36.48 ms	175.16 ms
bind assocR run	Ef	10000	7.19 ms	1.71 ms	5.24 ms	18.04 ms
bind assocL build	Eff	100	26.71 μs	51.22 μs	18.42 μs	1.07 ms
bind assocL build	Ef	100	11.38 μs	6.15 μs	9.26 μs	156.43 μs
bind assocL run	Eff	100	23.48 μs	69.28 μs	16.14 μs	1.55 ms
bind assocL run	Ef	100	14.85 μs	6.59 μs	13.92 μs	201.76 μs
bind assocL build	Eff	500	123.36 μs	103.25 μs	97.92 μs	1.36 ms
bind assocL build	Ef	500	53.55 μs	13.99 μs	47.51 μs	298.39 μs
bind assocL run	Eff	500	125.17 μs	192.44 μs	84.35 μs	2.31 ms
bind assocL run	Ef	500	86.09 μs	15.38 μs	79.21 μs	418.90 μs
bind assocL build	Eff	1000	264.39 μs	210.22 μs	195.13 μs	2.62 ms
bind assocL build	Ef	1000	105.60 μs	20.97 μs	94.77 μs	324.32 μs
bind assocL run	Eff	1000	245.24 μs	305.97 μs	162.13 μs	3.01 ms
bind assocL run	Ef	1000	195.82 μs	29.13 μs	183.39 μs	892.48 μs
bind assocL build	Eff	2000	575.85 μs	467.32 μs	386.96 μs	2.82 ms
bind assocL build	Ef	2000	207.73 μs	32.90 μs	187.73 μs	614.57 μs
bind assocL run	Eff	2000	527.22 μs	574.34 μs	317.74 μs	4.84 ms
bind assocL run	Ef	2000	501.78 μs	81.22 μs	460.96 μs	1.43 ms
bind assocL build	Eff	4000	1.16 ms	688.34 μs	785.20 μs	4.84 ms
bind assocL build	Ef	4000	536.46 μs	75.97 μs	482.31 μs	1.47 ms
bind assocL run	Eff	4000	1.22 ms	1.07 ms	658.75 μs	8.80 ms
bind assocL run	Ef	4000	1.46 ms	182.78 μs	1.34 ms	3.53 ms
bind assocL build	Eff	8000	3.16 ms	1.31 ms	1.78 ms	10.66 ms
bind assocL build	Ef	8000	1.11 ms	174.88 μs	960.70 μs	2.11 ms
bind assocL run	Eff	8000	2.59 ms	1.18 ms	1.69 ms	16.49 ms
bind assocL run	Ef	8000	7.75 ms	496.34 μs	7.23 ms	12.62 ms
map build	Eff	100	34.77 μs	97.12 μs	21.64 μs	1.90 ms
map build	Ef	100	13.59 μs	6.34 μs	11.69 μs	171.24 μs
map run	Eff	100	25.82 μs	76.17 μs	17.50 μs	2.16 ms
map run	Ef	100	15.89 μs	6.72 μs	14.30 μs	212.64 μs
map build	Eff	500	159.03 μs	195.17 μs	111.58 μs	1.72 ms
map build	Ef	500	68.44 μs	24.79 μs	60.14 μs	1.02 ms
map run	Eff	500	129.92 μs	171.99 μs	89.72 μs	2.01 ms
map run	Ef	500	90.90 μs	23.53 μs	82.72 μs	905.56 μs
map build	Eff	1000	330.00 μs	343.87 μs	221.16 μs	3.08 ms
map build	Ef	1000	129.78 μs	28.88 μs	119.38 μs	1.04 ms
map run	Eff	1000	252.79 μs	260.77 μs	173.81 μs	3.16 ms
map run	Ef	1000	210.18 μs	39.49 μs	190.60 μs	1.10 ms
map build	Eff	2000	718.99 μs	660.78 μs	441.72 μs	3.88 ms
map build	Ef	2000	257.94 μs	38.65 μs	237.90 μs	1.21 ms
map run	Eff	2000	506.02 μs	378.42 μs	349.51 μs	4.14 ms
map run	Ef	2000	518.35 μs	79.27 μs	473.70 μs	1.47 ms
map build	Eff	4000	1.42 ms	934.59 μs	878.94 μs	5.55 ms
map build	Ef	4000	525.22 μs	77.05 μs	473.96 μs	1.45 ms
map run	Eff	4000	1.04 ms	549.08 μs	722.80 μs	7.05 ms
map run	Ef	4000	1.47 ms	144.09 μs	1.37 ms	2.73 ms
map build	Eff	5000	1.90 ms	1.18 ms	1.11 ms	6.46 ms
map build	Ef	5000	668.66 μs	103.73 μs	595.40 μs	1.82 ms
map run	Eff	5000	1.56 ms	941.81 μs	973.49 μs	4.82 ms
map run	Ef	5000	2.57 ms	245.36 μs	2.27 ms	4.84 ms
apply build	Eff	100	531.69 ns	5.37 μs	235.00 ns	175.61 μs
apply build	Ef	100	509.32 ns	3.02 μs	331.00 ns	134.36 μs
apply run	Eff	100	173.57 μs	201.45 μs	122.28 μs	2.02 ms
apply run	Ef	100	136.74 μs	38.67 μs	116.69 μs	1.04 ms
apply build	Eff	500	373.04 ns	2.14 μs	243.00 ns	94.56 μs
apply build	Ef	500	484.92 ns	2.35 μs	337.00 ns	104.17 μs
apply run	Eff	500	999.63 μs	675.73 μs	629.89 μs	3.92 ms
apply run	Ef	500	764.32 μs	101.96 μs	683.45 μs	1.78 ms
apply build	Eff	1000	369.24 ns	1.80 μs	244.00 ns	79.92 μs
apply build	Ef	1000	478.92 ns	1.95 μs	333.00 ns	86.81 μs
apply run	Eff	1000	2.18 ms	1.12 ms	1.30 ms	6.93 ms
apply run	Ef	1000	1.73 ms	181.83 μs	1.56 ms	2.94 ms
apply build	Eff	2000	368.49 ns	1.43 μs	244.00 ns	62.42 μs
apply build	Ef	2000	471.56 ns	2.74 μs	328.00 ns	122.78 μs
apply run	Eff	2000	5.80 ms	1.82 ms	3.18 ms	14.53 ms
apply run	Ef	2000	4.61 ms	455.07 μs	4.07 ms	8.09 ms
apply build	Eff	4000	338.17 ns	1.45 μs	243.00 ns	64.89 μs
apply build	Ef	4000	457.02 ns	1.97 μs	334.00 ns	87.52 μs
apply run	Eff	4000	23.54 ms	5.64 ms	15.22 ms	47.46 ms
apply run	Ef	4000	18.73 ms	893.71 μs	17.42 ms	24.71 ms
apply build	Eff	5000	354.67 ns	1.35 μs	246.00 ns	60.00 μs
apply build	Ef	5000	467.07 ns	2.14 μs	336.00 ns	96.00 μs
apply run	Eff	5000	63.64 ms	24.42 ms	35.99 ms	149.47 ms
apply run	Ef	5000	29.13 ms	1.11 ms	27.65 ms	35.49 ms

npm run bench:run:extended

bench	type	n	mean	stddev	min	max
bind assocR	Ef	100000	226.42 ms	30.44 ms	183.64 ms	251.06 ms
bind assocR	Ef	1000000	2.22 s	58.55 ms	2.14 s	2.29 s
bind assocL	Ef	20000	145.42 ms	16.64 ms	123.93 ms	170.62 ms
bind assocL	Ef	40000	1.15 s	833.35 ms	517.74 ms	2.07 s
bind assocL	Ef	80000	19.65 s	3.03 s	14.90 s	23.20 s
map	Ef	10000	12.64 ms	1.01 ms	12.02 ms	14.43 ms
map	Ef	20000	50.15 ms	276.48 μs	49.77 ms	50.46 ms
map	Ef	40000	214.09 ms	3.09 ms	210.74 ms	217.56 ms
map	Ef	80000	15.36 s	77.80 ms	15.25 s	15.46 s
apply	Ef	10000	330.49 ms	273.96 ms	119.76 ms	634.09 ms
apply	Ef	20000	487.34 ms	5.83 ms	479.07 ms	494.23 ms
apply	Ef	40000	25.53 s	431.47 ms	25.14 s	26.11 s

safareli · 2017-12-13T01:00:24Z

npm run bench:

bench	type	n	mean	stddev	min	max
bind assocR build	Eff	100	1.07 μs	13.10 μs	225.00 ns	342.97 μs
bind assocR build	Ef	100	592.34 ns	6.56 μs	270.00 ns	224.92 μs
bind assocR run	Eff	100	78.08 μs	76.51 μs	53.94 μs	969.08 μs
bind assocR run	Ef	100	36.19 μs	21.14 μs	27.98 μs	451.29 μs
bind assocR build	Eff	500	330.31 ns	1.38 μs	236.00 ns	52.60 μs
bind assocR build	Ef	500	471.72 ns	3.17 μs	273.00 ns	90.08 μs
bind assocR run	Eff	500	402.77 μs	305.56 μs	280.04 μs	2.23 ms
bind assocR run	Ef	500	168.54 μs	59.26 μs	145.53 μs	1.13 ms
bind assocR build	Eff	1000	337.98 ns	1.70 μs	234.00 ns	66.74 μs
bind assocR build	Ef	1000	493.50 ns	3.23 μs	271.00 ns	94.04 μs
bind assocR run	Eff	1000	809.16 μs	441.44 μs	563.95 μs	3.42 ms
bind assocR run	Ef	1000	350.92 μs	114.66 μs	290.28 μs	1.85 ms
bind assocR build	Eff	2000	477.95 ns	2.08 μs	242.00 ns	79.33 μs
bind assocR build	Ef	2000	720.08 ns	5.75 μs	279.00 ns	154.55 μs
bind assocR run	Eff	2000	1.81 ms	764.95 μs	1.13 ms	6.91 ms
bind assocR run	Ef	2000	717.64 μs	238.34 μs	582.61 μs	3.24 ms
bind assocR build	Eff	4000	386.89 ns	1.96 μs	240.00 ns	58.98 μs
bind assocR build	Ef	4000	456.13 ns	2.59 μs	277.00 ns	79.25 μs
bind assocR run	Eff	4000	3.41 ms	920.06 μs	2.38 ms	16.35 ms
bind assocR run	Ef	4000	1.40 ms	376.05 μs	1.17 ms	6.81 ms
bind assocR build	Eff	8000	353.69 ns	1.95 μs	235.00 ns	59.56 μs
bind assocR build	Ef	8000	470.33 ns	2.60 μs	273.00 ns	76.77 μs
bind assocR run	Eff	8000	7.60 ms	2.32 ms	6.18 ms	51.94 ms
bind assocR run	Ef	8000	4.26 ms	1.04 ms	3.19 ms	16.76 ms
bind assocR build	Eff	10000	444.67 ns	2.33 μs	258.00 ns	78.46 μs
bind assocR build	Ef	10000	579.49 ns	3.76 μs	326.00 ns	136.65 μs
bind assocR run	Eff	10000	59.06 ms	18.74 ms	17.24 ms	133.24 ms
bind assocR run	Ef	10000	4.90 ms	1.09 ms	3.88 ms	11.25 ms
bind assocL build	Eff	100	26.55 μs	56.68 μs	18.17 μs	1.00 ms
bind assocL build	Ef	100	11.03 μs	7.04 μs	8.89 μs	148.55 μs
bind assocL run	Eff	100	27.39 μs	82.11 μs	16.53 μs	1.78 ms
bind assocL run	Ef	100	2.60 μs	7.62 μs	1.97 μs	228.19 μs
bind assocL build	Eff	500	125.83 μs	107.60 μs	98.89 μs	1.29 ms
bind assocL build	Ef	500	73.55 μs	40.79 μs	46.42 μs	1.05 ms
bind assocL run	Eff	500	176.11 μs	296.01 μs	85.26 μs	7.10 ms
bind assocL run	Ef	500	11.30 μs	9.38 μs	8.53 μs	250.18 μs
bind assocL build	Eff	1000	316.47 μs	287.27 μs	196.75 μs	6.48 ms
bind assocL build	Ef	1000	111.53 μs	53.30 μs	92.73 μs	1.33 ms
bind assocL run	Eff	1000	264.51 μs	319.44 μs	163.48 μs	3.14 ms
bind assocL run	Ef	1000	20.04 μs	23.65 μs	16.78 μs	902.12 μs
bind assocL build	Eff	2000	590.44 μs	476.66 μs	387.38 μs	3.33 ms
bind assocL build	Ef	2000	251.51 μs	102.59 μs	187.91 μs	1.25 ms
bind assocL run	Eff	2000	555.28 μs	580.38 μs	314.27 μs	4.50 ms
bind assocL run	Ef	2000	41.17 μs	32.79 μs	34.01 μs	971.66 μs
bind assocL build	Eff	4000	1.21 ms	747.20 μs	773.77 μs	6.08 ms
bind assocL build	Ef	4000	453.54 μs	177.13 μs	367.08 μs	2.98 ms
bind assocL run	Eff	4000	1.44 ms	1.23 ms	666.65 μs	9.56 ms
bind assocL run	Ef	4000	97.13 μs	98.26 μs	66.79 μs	2.38 ms
bind assocL build	Eff	8000	3.36 ms	1.46 ms	1.76 ms	12.17 ms
bind assocL build	Ef	8000	1.07 ms	205.74 μs	944.47 μs	3.28 ms
bind assocL run	Eff	8000	2.55 ms	1.17 ms	1.69 ms	17.08 ms
bind assocL run	Ef	8000	197.78 μs	91.12 μs	172.07 μs	1.77 ms
map build	Eff	100	32.87 μs	91.87 μs	20.98 μs	1.53 ms
map build	Ef	100	13.22 μs	6.29 μs	11.26 μs	174.25 μs
map run	Eff	100	25.76 μs	66.29 μs	16.93 μs	1.27 ms
map run	Ef	100	3.22 μs	6.60 μs	2.21 μs	199.89 μs
map build	Eff	500	156.63 μs	192.61 μs	110.74 μs	1.61 ms
map build	Ef	500	64.29 μs	22.29 μs	58.20 μs	927.55 μs
map run	Eff	500	128.05 μs	153.05 μs	87.62 μs	1.92 ms
map run	Ef	500	16.10 μs	9.05 μs	11.27 μs	254.84 μs
map build	Eff	1000	336.34 μs	366.75 μs	218.91 μs	2.99 ms
map build	Ef	1000	129.87 μs	30.97 μs	117.62 μs	1.10 ms
map run	Eff	1000	252.68 μs	234.18 μs	179.22 μs	2.35 ms
map run	Ef	1000	25.59 μs	29.67 μs	22.04 μs	913.62 μs
map build	Eff	2000	699.03 μs	651.74 μs	435.52 μs	3.73 ms
map build	Ef	2000	257.73 μs	45.95 μs	232.63 μs	1.18 ms
map run	Eff	2000	517.26 μs	361.48 μs	355.91 μs	4.20 ms
map run	Ef	2000	51.67 μs	33.56 μs	44.95 μs	995.56 μs
map build	Eff	4000	1.43 ms	950.64 μs	870.08 μs	5.01 ms
map build	Ef	4000	496.93 μs	65.05 μs	463.81 μs	1.55 ms
map run	Eff	4000	1.08 ms	563.89 μs	730.60 μs	7.41 ms
map run	Ef	4000	106.78 μs	60.04 μs	87.63 μs	1.07 ms
map build	Eff	5000	1.95 ms	1.24 ms	1.10 ms	6.78 ms
map build	Ef	5000	640.89 μs	98.72 μs	581.11 μs	1.70 ms
map run	Eff	5000	1.58 ms	926.79 μs	975.66 μs	4.80 ms
map run	Ef	5000	125.07 μs	65.76 μs	108.92 μs	1.50 ms
apply build	Eff	100	433.30 ns	4.51 μs	239.00 ns	143.75 μs
apply build	Ef	100	454.29 ns	2.12 μs	320.00 ns	94.01 μs
apply run	Eff	100	171.56 μs	190.27 μs	122.45 μs	2.87 ms
apply run	Ef	100	64.13 μs	36.84 μs	54.87 μs	1.12 ms
apply build	Eff	500	303.69 ns	1.17 μs	242.00 ns	52.59 μs
apply build	Ef	500	421.70 ns	1.70 μs	321.00 ns	74.92 μs
apply run	Eff	500	970.98 μs	583.27 μs	612.69 μs	3.86 ms
apply run	Ef	500	308.34 μs	54.92 μs	271.85 μs	1.28 ms
apply build	Eff	1000	319.60 ns	1.52 μs	242.00 ns	67.53 μs
apply build	Ef	1000	415.83 ns	1.69 μs	321.00 ns	74.76 μs
apply run	Eff	1000	2.18 ms	1.09 ms	1.27 ms	11.66 ms
apply run	Ef	1000	649.82 μs	137.22 μs	548.30 μs	2.23 ms
apply build	Eff	2000	344.88 ns	2.01 μs	245.00 ns	89.67 μs
apply build	Ef	2000	447.89 ns	2.35 μs	330.00 ns	105.16 μs
apply run	Eff	2000	5.61 ms	2.09 ms	2.91 ms	15.06 ms
apply run	Ef	2000	1.35 ms	238.45 μs	1.11 ms	3.28 ms
apply build	Eff	4000	312.94 ns	1.10 μs	242.00 ns	48.18 μs
apply build	Ef	4000	414.19 ns	1.67 μs	321.00 ns	74.88 μs
apply run	Eff	4000	26.05 ms	7.00 ms	13.72 ms	79.20 ms
apply run	Ef	4000	2.94 ms	761.51 μs	2.34 ms	22.61 ms
apply build	Eff	5000	322.27 ns	1.39 μs	239.00 ns	61.33 μs
apply build	Ef	5000	459.24 ns	2.31 μs	330.00 ns	103.42 μs
apply run	Eff	5000	65.66 ms	20.73 ms	37.99 ms	136.15 ms
apply run	Ef	5000	4.05 ms	906.83 μs	3.03 ms	11.31 ms

npm run bench:run:extended

bench	type	n	mean	stddev	min	max
bind assocR	Ef	20000	31.10 ms	11.50 ms	13.87 ms	42.72 ms
bind assocR	Ef	50000	76.90 ms	20.10 ms	59.85 ms	111.80 ms
bind assocR	Ef	100000	166.82 ms	16.35 ms	145.08 ms	188.14 ms
bind assocR	Ef	1000000	2.22 s	283.14 ms	1.84 s	2.59 s
bind assocL	Ef	20000	34.56 ms	14.32 ms	20.91 ms	53.50 ms
bind assocL	Ef	50000	93.31 ms	14.31 ms	77.26 ms	111.25 ms
bind assocL	Ef	100000	185.56 ms	24.32 ms	154.12 ms	218.03 ms
bind assocL	Ef	1000000	2.34 s	231.97 ms	2.02 s	2.57 s
map	Ef	10000	1.43 ms	437.95 μs	1.18 ms	2.21 ms
map	Ef	20000	2.15 ms	244.78 μs	1.95 ms	2.55 ms
map	Ef	50000	6.21 ms	888.34 μs	5.25 ms	7.22 ms
map	Ef	100000	11.38 ms	592.67 μs	10.65 ms	12.06 ms
map	Ef	1000000	122.13 ms	4.46 ms	115.44 ms	127.37 ms
map	Ef	10000000	2.57 s	618.45 ms	1.48 s	2.96 s
apply	Ef	10000	19.98 ms	6.64 ms	11.29 ms	28.79 ms
apply	Ef	20000	30.26 ms	5.03 ms	25.01 ms	37.17 ms
apply	Ef	50000	81.35 ms	13.06 ms	68.95 ms	100.05 ms
apply	Ef	100000	194.60 ms	32.15 ms	175.02 ms	251.35 ms
apply	Ef	1000000	2.07 s	72.61 ms	1.99 s	2.15 s

😬👆

natefaubion · 2017-12-13T03:53:09Z

I wonder if Effect could be an unlifted thunk (checking for it with typeof), then you could define it directly against the FFI. The FFI doesn't really need to use bind/map etc.

Some things to consider:

Eff do blocks get inlined into imperative statements, and in the 0.12 branch EffFn gets directly inlined, which means you can potentially write code with basically no overhead (though I'm sure it will be ugly).
These benchmarks benchmark the pure data structure aspect of Eff, but don't consider compiler optimizations since the trees are programmatically generated.
If you had an unlifted Effect constructor, you could have an unsafeCoerce implementation of liftEff as well which would make compatibility nice.
With an unlifted Effect you could probably do the same sort optimizations as Eff pretty easily.

safareli · 2017-12-13T15:32:12Z

Yes typeof x == "function" could totally work instead of Effect node, I was planning to do that.

With an unlifted Effect you could probably do the same sort optimizations as Eff pretty easily.

What you mean here? the optimizations are done compile time right? is the optimization specialized to Eff? If there are couple Ef statements I would still need to build Bind nodes.

safareli · 2017-12-13T16:03:44Z

This made it a bit faster in some cases

bench	type	n	mean	stddev	min	max
bind assocR build	Eff	100	1.54 μs	14.12 μs	231.00 ns	265.04 μs
bind assocR build	Ef	100	860.41 ns	10.79 μs	267.00 ns	307.18 μs
bind assocR run	Eff	100	82.18 μs	80.18 μs	52.94 μs	962.34 μs
bind assocR run	Ef	100	39.24 μs	28.98 μs	31.32 μs	605.14 μs
bind assocR build	Eff	500	565.36 ns	2.91 μs	259.00 ns	70.92 μs
bind assocR build	Ef	500	558.95 ns	3.16 μs	279.00 ns	78.20 μs
bind assocR run	Eff	500	404.03 μs	303.58 μs	276.04 μs	2.24 ms
bind assocR run	Ef	500	180.79 μs	76.61 μs	155.57 μs	1.19 ms
bind assocR build	Eff	1000	378.98 ns	2.07 μs	234.00 ns	55.07 μs
bind assocR build	Ef	1000	523.71 ns	2.94 μs	271.00 ns	67.02 μs
bind assocR run	Eff	1000	795.81 μs	434.58 μs	563.81 μs	3.44 ms
bind assocR run	Ef	1000	346.14 μs	109.81 μs	305.53 μs	1.79 ms
bind assocR build	Eff	2000	392.70 ns	2.34 μs	234.00 ns	61.28 μs
bind assocR build	Ef	2000	640.35 ns	3.86 μs	271.00 ns	94.31 μs
bind assocR run	Eff	2000	1.61 ms	605.90 μs	1.13 ms	6.45 ms
bind assocR run	Ef	2000	734.17 μs	216.11 μs	607.85 μs	2.52 ms
bind assocR build	Eff	4000	397.37 ns	2.16 μs	234.00 ns	53.88 μs
bind assocR build	Ef	4000	489.49 ns	2.75 μs	271.00 ns	65.92 μs
bind assocR run	Eff	4000	3.43 ms	1.02 ms	2.39 ms	15.92 ms
bind assocR run	Ef	4000	1.44 ms	364.73 μs	1.22 ms	4.03 ms
bind assocR build	Eff	8000	392.61 ns	2.23 μs	233.00 ns	56.89 μs
bind assocR build	Ef	8000	541.04 ns	3.16 μs	264.00 ns	70.48 μs
bind assocR run	Eff	8000	8.42 ms	3.33 ms	6.06 ms	47.50 ms
bind assocR run	Ef	8000	3.93 ms	947.32 μs	2.94 ms	13.17 ms
bind assocR build	Eff	10000	382.11 ns	1.97 μs	233.00 ns	44.81 μs
bind assocR build	Ef	10000	465.88 ns	2.59 μs	274.00 ns	69.81 μs
bind assocR run	Eff	10000	60.10 ms	18.69 ms	17.04 ms	136.96 ms
bind assocR run	Ef	10000	5.41 ms	1.58 ms	3.64 ms	16.28 ms
bind assocL build	Eff	100	25.53 μs	53.02 μs	17.96 μs	884.77 μs
bind assocL build	Ef	100	11.98 μs	8.34 μs	9.68 μs	178.49 μs
bind assocL run	Eff	100	27.51 μs	93.66 μs	15.83 μs	2.06 ms
bind assocL run	Ef	100	3.30 μs	10.59 μs	2.05 μs	219.43 μs
bind assocL build	Eff	500	120.82 μs	107.51 μs	95.88 μs	1.59 ms
bind assocL build	Ef	500	62.38 μs	22.78 μs	50.48 μs	287.69 μs
bind assocL run	Eff	500	119.65 μs	167.43 μs	85.30 μs	1.98 ms
bind assocL run	Ef	500	12.26 μs	26.70 μs	8.80 μs	778.41 μs
bind assocL build	Eff	1000	260.22 μs	211.02 μs	191.63 μs	1.83 ms
bind assocL build	Ef	1000	111.31 μs	27.18 μs	100.84 μs	415.88 μs
bind assocL run	Eff	1000	253.98 μs	304.55 μs	162.10 μs	2.82 ms
bind assocL run	Ef	1000	19.47 μs	29.81 μs	17.04 μs	813.45 μs
bind assocL build	Eff	2000	576.14 μs	483.90 μs	383.13 μs	3.46 ms
bind assocL build	Ef	2000	266.29 μs	95.36 μs	200.17 μs	1.09 ms
bind assocL run	Eff	2000	652.16 μs	776.47 μs	325.22 μs	6.84 ms
bind assocL run	Ef	2000	47.79 μs	47.87 μs	35.53 μs	930.58 μs
bind assocL build	Eff	4000	1.25 ms	806.64 μs	770.20 μs	5.74 ms
bind assocL build	Ef	4000	584.74 μs	114.68 μs	517.19 μs	1.92 ms
bind assocL run	Eff	4000	1.30 ms	1.08 ms	701.22 μs	9.20 ms
bind assocL run	Ef	4000	74.78 μs	69.70 μs	67.02 μs	1.37 ms
bind assocL build	Eff	8000	3.12 ms	1.33 ms	1.77 ms	13.67 ms
bind assocL build	Ef	8000	1.20 ms	245.79 μs	1.03 ms	2.93 ms
bind assocL run	Eff	8000	2.54 ms	1.27 ms	1.65 ms	16.99 ms
bind assocL run	Ef	8000	206.06 μs	131.35 μs	179.96 μs	3.08 ms
map build	Eff	100	40.07 μs	99.19 μs	21.63 μs	1.54 ms
map build	Ef	100	14.87 μs	7.17 μs	12.54 μs	195.28 μs
map run	Eff	100	26.26 μs	72.61 μs	16.71 μs	1.49 ms
map run	Ef	100	4.05 μs	11.85 μs	2.34 μs	278.85 μs
map build	Eff	500	171.53 μs	209.84 μs	112.81 μs	1.78 ms
map build	Ef	500	72.81 μs	16.28 μs	63.95 μs	211.12 μs
map run	Eff	500	129.44 μs	163.46 μs	87.17 μs	2.30 ms
map run	Ef	500	13.23 μs	12.19 μs	11.11 μs	247.87 μs
map build	Eff	1000	356.81 μs	393.99 μs	223.88 μs	4.08 ms
map build	Ef	1000	152.29 μs	45.67 μs	126.98 μs	498.26 μs
map run	Eff	1000	252.13 μs	238.58 μs	171.45 μs	2.50 ms
map run	Ef	1000	29.97 μs	30.76 μs	22.71 μs	790.31 μs
map build	Eff	2000	757.29 μs	745.33 μs	443.87 μs	6.76 ms
map build	Ef	2000	267.71 μs	34.91 μs	252.32 μs	539.37 μs
map run	Eff	2000	490.65 μs	331.30 μs	355.59 μs	3.97 ms
map run	Ef	2000	51.85 μs	37.46 μs	46.41 μs	1.00 ms
map build	Eff	4000	1.42 ms	948.94 μs	887.71 μs	6.12 ms
map build	Ef	4000	554.65 μs	85.19 μs	503.57 μs	1.09 ms
map run	Eff	4000	1.09 ms	585.87 μs	748.40 μs	5.18 ms
map run	Ef	4000	106.51 μs	72.34 μs	90.39 μs	1.39 ms
map build	Eff	5000	1.88 ms	1.18 ms	1.10 ms	7.90 ms
map build	Ef	5000	736.66 μs	188.45 μs	631.34 μs	2.82 ms
map run	Eff	5000	1.51 ms	953.11 μs	925.47 μs	7.52 ms
map run	Ef	5000	130.58 μs	67.27 μs	112.88 μs	1.65 ms
apply build	Eff	100	603.83 ns	6.53 μs	232.00 ns	149.49 μs
apply build	Ef	100	436.87 ns	2.18 μs	276.00 ns	66.54 μs
apply run	Eff	100	174.77 μs	199.01 μs	116.81 μs	1.96 ms
apply run	Ef	100	63.79 μs	33.04 μs	54.66 μs	912.83 μs
apply build	Eff	500	337.46 ns	1.56 μs	236.00 ns	48.66 μs
apply build	Ef	500	431.56 ns	2.62 μs	285.00 ns	82.55 μs
apply run	Eff	500	978.28 μs	595.58 μs	620.88 μs	3.80 ms
apply run	Ef	500	308.85 μs	74.92 μs	269.64 μs	1.15 ms
apply build	Eff	1000	342.55 ns	1.88 μs	234.00 ns	59.55 μs
apply build	Ef	1000	446.46 ns	2.90 μs	285.00 ns	89.73 μs
apply run	Eff	1000	2.18 ms	1.08 ms	1.25 ms	7.32 ms
apply run	Ef	1000	629.16 μs	214.26 μs	538.05 μs	4.37 ms
apply build	Eff	2000	355.54 ns	2.15 μs	244.00 ns	68.15 μs
apply build	Ef	2000	460.14 ns	3.05 μs	292.00 ns	96.15 μs
apply run	Eff	2000	5.48 ms	2.05 ms	2.91 ms	13.62 ms
apply run	Ef	2000	1.27 ms	229.82 μs	1.11 ms	3.39 ms
apply build	Eff	4000	343.87 ns	1.93 μs	241.00 ns	61.05 μs
apply build	Ef	4000	397.49 ns	1.94 μs	277.00 ns	60.76 μs
apply run	Eff	4000	27.31 ms	8.73 ms	14.03 ms	73.46 ms
apply run	Ef	4000	2.78 ms	594.66 μs	2.25 ms	6.82 ms
apply build	Eff	5000	333.29 ns	1.76 μs	232.00 ns	55.78 μs
apply build	Ef	5000	421.57 ns	2.56 μs	275.00 ns	80.51 μs
apply run	Eff	5000	67.62 ms	21.46 ms	35.71 ms	144.26 ms
apply run	Ef	5000	3.75 ms	505.63 μs	3.05 ms	7.09 ms

bench	type	n	mean	stddev	min	max
bind assocR	Ef	20000	33.47 ms	2.58 ms	31.19 ms	37.62 ms
bind assocR	Ef	50000	72.61 ms	19.12 ms	52.33 ms	103.82 ms
bind assocR	Ef	100000	175.08 ms	46.69 ms	133.62 ms	251.27 ms
bind assocR	Ef	1000000	1.91 s	240.86 ms	1.58 s	2.26 s
bind assocL	Ef	20000	22.34 ms	15.20 ms	7.99 ms	46.87 ms
bind assocL	Ef	50000	82.20 ms	22.12 ms	66.01 ms	119.93 ms
bind assocL	Ef	100000	187.54 ms	51.87 ms	139.80 ms	247.96 ms
bind assocL	Ef	1000000	2.31 s	305.29 ms	2.06 s	2.79 s
map	Ef	10000	2.57 ms	846.93 μs	1.67 ms	3.43 ms
map	Ef	20000	2.15 ms	265.70 μs	1.96 ms	2.60 ms
map	Ef	50000	6.92 ms	679.14 μs	6.16 ms	7.69 ms
map	Ef	100000	17.03 ms	3.51 ms	12.79 ms	21.19 ms
map	Ef	1000000	117.77 ms	1.42 ms	115.48 ms	119.07 ms
map	Ef	10000000	2.72 s	707.12 ms	1.51 s	3.27 s
apply	Ef	10000	10.00 ms	2.57 ms	7.49 ms	13.90 ms
apply	Ef	20000	26.25 ms	2.15 ms	23.74 ms	29.02 ms
apply	Ef	50000	74.68 ms	7.74 ms	66.42 ms	84.51 ms
apply	Ef	100000	154.23 ms	17.32 ms	133.69 ms	172.05 ms
apply	Ef	1000000	1.75 s	120.94 ms	1.65 s	1.92 s

natefaubion · 2017-12-13T17:01:14Z

src/Control/Monad/Ff.js

+  };
+};
+
+exports.toEff = function (inputEff) {


Maybe you could do a typeof inputEff === "function" check here so you can avoid creating another thunk to run the inner thunk.

so something like this:

if (typeof inputEff === "function") { return inputEff }

thanks!

natefaubion · 2017-12-13T17:06:29Z

What you mean here? the optimizations are done compile time right? is the optimization specialized to Eff? If there are couple Ef statements I would still need to build Bind nodes.

Since you have an unlifted thunk as a representation for Ef, you could directly inline the binds as imperative statements the same way as Eff does, and then wrap each "bind" with toEff. This optimization works for both Eff and Ef, since the core representation (an unlifted thunk) is still the same. Ie., calling toEff on an Ef that's just Eff is an identity function. This gives you the best of both current worlds, where syntactic do blocks are largely 0-overhead imperative statements, and combinator based transformations are stack safe and fast.

safareli · 2017-12-13T19:30:22Z

Since you have an unlifted thunk as a representation for Ef, you could directly inline the binds as imperative statements the same way as Eff does, and then wrap each "bind" with toEff

I don't understand how should I inline statements the same way? This is what I tired, but i'm not seeing same optimization happening on Ef, is there some way to force that? /cc @paf31:

test.purs

x :: Eff TestEff Unit
x = do
  ar <- mkArr
  _ <- pushToArr ar 1
  _ <- pushToArr ar 1
  _ <- pushToArr ar 1
  pure unit

x' :: Ef TestEff Unit
x' = do
  ar <- mkArr_
  _ <- pushToArr_ ar 1
  _ <- pushToArr_ ar 1
  _ <- pushToArr_ ar 1
  pure unit

foreign import mkArr :: forall e a. Eff e (Arr a)
foreign import mkArr_ :: forall e a. Ef e (Arr a)
foreign import pushToArr :: forall e a. Arr a -> a -> Eff e a
foreign import pushToArr_ :: forall e a. Arr a -> a -> Ef e a

out.js

var x$prime = Control_Bind.bind(Control_Monad_Ef.bindEf)($foreign.mkArr_)(function (v) {
    return Control_Bind.bind(Control_Monad_Ef.bindEf)($foreign.pushToArr_(v)(1))(function (v1) {
        return Control_Bind.bind(Control_Monad_Ef.bindEf)($foreign.pushToArr_(v)(1))(function (v2) {
            return Control_Bind.bind(Control_Monad_Ef.bindEf)($foreign.pushToArr_(v)(1))(function (v3) {
                return Control_Applicative.pure(Control_Monad_Ef.applicativeEf)(Data_Unit.unit);
            });
        });
    });
});

var x = function __do() {
    var v = $foreign.mkArr();
    var v1 = $foreign.pushToArr(v)(1)();
    var v2 = $foreign.pushToArr(v)(1)();
    var v3 = $foreign.pushToArr(v)(1)();
    return Data_Unit.unit;
};

test.js

exports.mkArr = function(){
  return [];
};

exports.pushToArr = function(xs) {
  return function(x) {
    return function() {
      xs.push(x);
      return x;
    };
  };
};

exports.mkArr_ = function(){
  return [];
};

exports.pushToArr_ = function(xs) {
  return function(x) {
    return function() {
      xs.push(x);
      return x;
    };
  };
};

paf31 · 2017-12-13T19:32:45Z

You would need to either

a) change the compiler, but this would need to be in a fork
b) write a corefn pass (but this isn't really a practical option just yet)
c) (horrible) trick the compiler by naming your module and instances to match Eff 😈

natefaubion · 2017-12-13T19:33:05Z

Oh sorry, I meant as a hypothetical 😄. Yes, it would still require built-in compiler optimizations. My point was that the optimizations can be made to be identical for both types due to the representation.

natefaubion · 2017-12-13T19:33:50Z

b) write a corefn pass (but this isn't really a practical option just yet)

I think it would have to be coreimp, as corefn has no way to write raw statements.

safareli · 2017-12-13T19:34:03Z

ah :d gotcha

safareli · 2017-12-13T20:20:38Z

Added Aff to benc for comparison, It was expected that Ef would be faster, as it's synchronous and has to dial with less staff. /cc @jdegoes

npm run bench:run

bench	type	n	mean	stddev	min	max
>>=R build	Eff	100	1.49 μs	16.23 μs	231.00 ns	442.54 μs
>>=R build	Aff	100	816.83 ns	7.55 μs	275.00 ns	151.82 μs
>>=R build	Ef	100	812.19 ns	9.26 μs	271.00 ns	225.59 μs
>>=R run	Eff	100	85.35 μs	94.94 μs	52.72 μs	1.45 ms
>>=R run	Aff	100	117.84 μs	362.92 μs	59.50 μs	4.27 ms
>>=R run	Ef	100	45.17 μs	66.93 μs	29.79 μs	898.93 μs
>>=R build	Eff	500	461.85 ns	2.48 μs	243.00 ns	51.89 μs
>>=R build	Aff	500	554.29 ns	2.85 μs	289.00 ns	70.86 μs
>>=R build	Ef	500	571.89 ns	4.03 μs	282.00 ns	110.71 μs
>>=R run	Eff	500	396.02 μs	306.05 μs	278.65 μs	3.11 ms
>>=R run	Aff	500	349.41 μs	216.55 μs	288.14 μs	2.14 ms
>>=R run	Ef	500	211.91 μs	214.10 μs	157.57 μs	3.05 ms
>>=R build	Eff	1000	451.43 ns	2.90 μs	240.00 ns	64.74 μs
>>=R build	Aff	1000	423.80 ns	1.96 μs	282.00 ns	48.23 μs
>>=R build	Ef	1000	484.42 ns	2.72 μs	271.00 ns	64.27 μs
>>=R run	Eff	1000	782.83 μs	447.02 μs	553.43 μs	4.33 ms
>>=R run	Aff	1000	698.41 μs	298.77 μs	575.56 μs	2.64 ms
>>=R run	Ef	1000	425.18 μs	319.81 μs	311.37 μs	2.47 ms
>>=R build	Eff	2000	467.45 ns	2.88 μs	241.00 ns	71.70 μs
>>=R build	Aff	2000	434.82 ns	2.11 μs	280.00 ns	51.48 μs
>>=R build	Ef	2000	617.60 ns	3.70 μs	309.00 ns	99.49 μs
>>=R run	Eff	2000	1.64 ms	711.80 μs	1.12 ms	7.69 ms
>>=R run	Aff	2000	1.50 ms	668.52 μs	1.14 ms	4.22 ms
>>=R run	Ef	2000	890.26 μs	556.44 μs	621.80 μs	3.61 ms
>>=R build	Eff	4000	414.43 ns	2.40 μs	239.00 ns	52.51 μs
>>=R build	Aff	4000	425.75 ns	2.14 μs	280.00 ns	48.56 μs
>>=R build	Ef	4000	560.42 ns	2.69 μs	288.00 ns	59.95 μs
>>=R run	Eff	4000	3.31 ms	817.00 μs	2.37 ms	12.42 ms
>>=R run	Aff	4000	3.14 ms	1.09 ms	2.31 ms	9.21 ms
>>=R run	Ef	4000	1.84 ms	863.40 μs	1.27 ms	6.14 ms
>>=R build	Eff	8000	436.05 ns	3.00 μs	234.00 ns	83.83 μs
>>=R build	Aff	8000	625.85 ns	2.10 μs	286.00 ns	52.33 μs
>>=R build	Ef	8000	489.29 ns	2.80 μs	271.00 ns	68.00 μs
>>=R run	Eff	8000	8.28 ms	2.12 ms	7.22 ms	44.06 ms
>>=R run	Aff	8000	7.82 ms	2.43 ms	4.68 ms	20.10 ms
>>=R run	Ef	8000	4.95 ms	1.92 ms	2.86 ms	14.96 ms
>>=R build	Eff	10000	434.23 ns	2.93 μs	243.00 ns	70.57 μs
>>=R build	Aff	10000	493.17 ns	3.74 μs	276.00 ns	110.81 μs
>>=R build	Ef	10000	610.73 ns	4.86 μs	260.00 ns	129.78 μs
>>=R run	Eff	10000	68.41 ms	17.47 ms	22.87 ms	145.99 ms
>>=R run	Aff	10000	12.33 ms	1.21 ms	6.52 ms	28.35 ms
>>=R run	Ef	10000	6.45 ms	2.33 ms	3.48 ms	13.76 ms
>>=L build	Eff	100	775.32 ns	9.66 μs	242.00 ns	249.20 μs
>>=L build	Aff	100	467.65 ns	2.55 μs	283.00 ns	66.55 μs
>>=L build	Ef	100	476.92 ns	2.40 μs	285.00 ns	64.84 μs
>>=L run	Eff	100	84.26 μs	145.72 μs	55.31 μs	1.86 ms
>>=L run	Aff	100	81.92 μs	89.33 μs	65.97 μs	1.17 ms
>>=L run	Ef	100	45.72 μs	83.40 μs	32.32 μs	1.55 ms
>>=L build	Eff	500	444.27 ns	3.07 μs	243.00 ns	79.22 μs
>>=L build	Aff	500	556.77 ns	2.39 μs	281.00 ns	54.30 μs
>>=L build	Ef	500	572.30 ns	3.64 μs	294.00 ns	105.45 μs
>>=L run	Eff	500	457.22 μs	427.23 μs	286.88 μs	3.45 ms
>>=L run	Aff	500	387.24 μs	198.47 μs	324.16 μs	1.59 ms
>>=L run	Ef	500	227.00 μs	197.79 μs	169.05 μs	1.73 ms
>>=L build	Eff	1000	461.00 ns	2.83 μs	250.00 ns	55.62 μs
>>=L build	Aff	1000	417.99 ns	1.86 μs	284.00 ns	45.41 μs
>>=L build	Ef	1000	505.28 ns	2.92 μs	291.00 ns	82.59 μs
>>=L run	Eff	1000	926.61 μs	616.93 μs	557.85 μs	5.31 ms
>>=L run	Aff	1000	787.18 μs	292.33 μs	648.23 μs	2.33 ms
>>=L run	Ef	1000	461.75 μs	302.23 μs	349.27 μs	2.23 ms
>>=L build	Eff	2000	418.84 ns	2.55 μs	245.00 ns	59.24 μs
>>=L build	Aff	2000	508.96 ns	3.20 μs	283.00 ns	72.93 μs
>>=L build	Ef	2000	483.00 ns	2.55 μs	286.00 ns	62.51 μs
>>=L run	Eff	2000	2.04 ms	1.04 ms	1.28 ms	9.27 ms
>>=L run	Aff	2000	1.62 ms	426.38 μs	1.29 ms	3.74 ms
>>=L run	Ef	2000	927.16 μs	396.91 μs	677.70 μs	3.07 ms
>>=L build	Eff	4000	377.22 ns	1.97 μs	245.00 ns	59.11 μs
>>=L build	Aff	4000	417.36 ns	1.98 μs	286.00 ns	60.09 μs
>>=L build	Ef	4000	477.56 ns	2.65 μs	287.00 ns	80.87 μs
>>=L run	Eff	4000	4.12 ms	1.39 ms	2.69 ms	17.59 ms
>>=L run	Aff	4000	3.83 ms	1.35 ms	2.67 ms	7.42 ms
>>=L run	Ef	4000	2.28 ms	1.23 ms	1.37 ms	5.84 ms
>>=L build	Eff	8000	339.91 ns	1.68 μs	246.00 ns	53.28 μs
>>=L build	Aff	8000	382.33 ns	1.77 μs	287.00 ns	56.21 μs
>>=L build	Ef	8000	472.56 ns	2.75 μs	291.00 ns	85.35 μs
>>=L run	Eff	8000	15.33 ms	3.14 ms	8.10 ms	50.90 ms
>>=L run	Aff	8000	8.31 ms	2.46 ms	5.71 ms	25.55 ms
>>=L run	Ef	8000	4.59 ms	1.38 ms	3.01 ms	8.39 ms
map build	Eff	100	1.04 μs	13.47 μs	267.00 ns	337.24 μs
map build	Aff	100	398.43 ns	1.72 μs	289.00 ns	54.45 μs
map build	Ef	100	450.71 ns	2.91 μs	294.00 ns	91.69 μs
map run	Eff	100	67.64 μs	148.88 μs	41.13 μs	2.39 ms
map run	Aff	100	34.45 μs	36.22 μs	29.09 μs	1.10 ms
map run	Ef	100	22.52 μs	29.50 μs	17.11 μs	850.09 μs
map build	Eff	500	336.66 ns	1.61 μs	246.00 ns	50.98 μs
map build	Aff	500	621.61 ns	4.27 μs	295.00 ns	83.55 μs
map build	Ef	500	476.36 ns	2.59 μs	318.00 ns	81.12 μs
map run	Eff	500	312.46 μs	318.26 μs	206.00 μs	2.57 ms
map run	Aff	500	162.39 μs	72.26 μs	138.00 μs	1.74 ms
map run	Ef	500	99.62 μs	52.59 μs	82.32 μs	1.10 ms
map build	Eff	1000	350.86 ns	1.83 μs	255.00 ns	58.04 μs
map build	Aff	1000	666.53 ns	4.48 μs	290.00 ns	75.19 μs
map build	Ef	1000	429.57 ns	2.05 μs	292.00 ns	63.74 μs
map run	Eff	1000	611.24 μs	458.89 μs	413.79 μs	3.10 ms
map run	Aff	1000	298.13 μs	79.09 μs	269.90 μs	1.96 ms
map run	Ef	1000	186.69 μs	58.39 μs	162.50 μs	1.10 ms
map build	Eff	2000	386.81 ns	2.29 μs	250.00 ns	72.65 μs
map build	Aff	2000	628.85 ns	5.01 μs	293.00 ns	135.23 μs
map build	Ef	2000	469.41 ns	2.96 μs	294.00 ns	92.62 μs
map run	Eff	2000	1.31 ms	729.52 μs	844.60 μs	5.33 ms
map run	Aff	2000	727.74 μs	150.59 μs	641.74 μs	2.74 ms
map run	Ef	2000	366.51 μs	73.29 μs	323.57 μs	1.35 ms
map build	Eff	4000	390.74 ns	2.15 μs	251.00 ns	63.23 μs
map build	Aff	4000	560.09 ns	3.09 μs	357.00 ns	93.56 μs
map build	Ef	4000	451.80 ns	2.89 μs	293.00 ns	91.13 μs
map run	Eff	4000	2.83 ms	1.16 ms	1.82 ms	11.20 ms
map run	Aff	4000	1.68 ms	220.56 μs	1.51 ms	4.29 ms
map run	Ef	4000	750.44 μs	148.78 μs	649.37 μs	2.15 ms
map build	Eff	5000	353.79 ns	2.22 μs	246.00 ns	70.36 μs
map build	Aff	5000	548.62 ns	3.26 μs	363.00 ns	102.53 μs
map build	Ef	5000	444.05 ns	2.41 μs	296.00 ns	75.04 μs
map run	Eff	5000	4.77 ms	2.18 ms	2.37 ms	13.35 ms
map run	Aff	5000	2.15 ms	315.96 μs	1.88 ms	5.28 ms
map run	Ef	5000	1.08 ms	482.85 μs	797.27 μs	5.99 ms
apply build	Eff	100	854.43 ns	8.09 μs	251.00 ns	176.91 μs
apply build	Aff	100	535.31 ns	2.70 μs	372.00 ns	85.08 μs
apply build	Ef	100	460.71 ns	2.50 μs	297.00 ns	78.60 μs
apply run	Eff	100	183.71 μs	248.92 μs	117.09 μs	2.91 ms
apply run	Aff	100	210.72 μs	113.11 μs	170.22 μs	1.67 ms
apply run	Ef	100	73.85 μs	78.71 μs	54.57 μs	958.33 μs
apply build	Eff	500	351.47 ns	2.15 μs	247.00 ns	68.35 μs
apply build	Aff	500	553.04 ns	3.13 μs	363.00 ns	98.30 μs
apply build	Ef	500	455.55 ns	2.73 μs	292.00 ns	85.24 μs
apply run	Eff	500	1.02 ms	655.54 μs	629.29 μs	4.14 ms
apply run	Aff	500	1.08 ms	360.78 μs	848.25 μs	3.42 ms
apply run	Ef	500	346.69 μs	160.20 μs	273.48 μs	1.33 ms
apply build	Eff	1000	348.80 ns	1.93 μs	247.00 ns	61.13 μs
apply build	Aff	1000	516.73 ns	2.41 μs	363.00 ns	75.88 μs
apply build	Ef	1000	502.20 ns	4.05 μs	294.00 ns	126.36 μs
apply run	Eff	1000	2.30 ms	1.17 ms	1.29 ms	8.21 ms
apply run	Aff	1000	2.18 ms	562.34 μs	1.73 ms	4.95 ms
apply run	Ef	1000	697.71 μs	220.07 μs	549.14 μs	2.11 ms
apply build	Eff	2000	352.05 ns	1.89 μs	256.00 ns	59.94 μs
apply build	Aff	2000	538.76 ns	2.97 μs	362.00 ns	93.42 μs
apply build	Ef	2000	479.01 ns	3.43 μs	291.00 ns	107.71 μs
apply run	Eff	2000	5.53 ms	1.98 ms	3.00 ms	15.27 ms
apply run	Aff	2000	4.65 ms	1.07 ms	3.49 ms	8.77 ms
apply run	Ef	2000	1.45 ms	367.98 μs	1.12 ms	4.70 ms
apply build	Eff	4000	353.27 ns	1.71 μs	245.00 ns	53.51 μs
apply build	Aff	4000	670.78 ns	2.88 μs	371.00 ns	87.53 μs
apply build	Ef	4000	507.00 ns	3.63 μs	294.00 ns	109.41 μs
apply run	Eff	4000	28.25 ms	6.09 ms	14.86 ms	53.36 ms
apply run	Aff	4000	9.73 ms	1.71 ms	7.08 ms	15.10 ms
apply run	Ef	4000	3.21 ms	769.60 μs	2.47 ms	8.24 ms
apply build	Eff	5000	355.53 ns	1.81 μs	250.00 ns	57.46 μs
apply build	Aff	5000	797.64 ns	5.99 μs	369.00 ns	189.10 μs
apply build	Ef	5000	476.37 ns	3.53 μs	292.00 ns	111.36 μs
apply run	Eff	5000	79.21 ms	26.47 ms	36.88 ms	188.51 ms
apply run	Aff	5000	13.59 ms	2.04 ms	11.89 ms	29.44 ms
apply run	Ef	5000	4.04 ms	729.38 μs	3.08 ms	8.37 ms

npm run bench:run:extended

bench	type	n	mean	stddev	min	max
>>=R build	Aff	20000	63.59 μs	111.81 μs	3.10 μs	231.14 μs
>>=R build	Ef	20000	46.44 μs	79.11 μs	757.00 ns	164.38 μs
>>=R run	Aff	20000	53.51 ms	12.01 ms	40.71 ms	66.93 ms
>>=R run	Ef	20000	26.01 ms	12.07 ms	8.79 ms	36.01 ms
>>=R build	Aff	50000	15.56 μs	29.81 μs	518.00 ns	60.27 μs
>>=R build	Ef	50000	23.09 μs	38.01 μs	568.00 ns	79.59 μs
>>=R run	Aff	50000	143.41 ms	34.44 ms	115.83 ms	193.77 ms
>>=R run	Ef	50000	67.65 ms	14.46 ms	48.13 ms	79.33 ms
>>=R build	Aff	100000	51.05 μs	96.89 μs	573.00 ns	196.33 μs
>>=R build	Ef	100000	21.01 μs	32.37 μs	841.00 ns	69.12 μs
>>=R run	Aff	100000	224.61 ms	42.60 ms	192.33 ms	285.06 ms
>>=R run	Ef	100000	143.07 ms	14.66 ms	128.64 ms	157.06 ms
>>=R build	Aff	1000000	15.97 μs	29.76 μs	896.00 ns	60.61 μs
>>=R build	Ef	1000000	27.82 μs	44.41 μs	758.00 ns	93.58 μs
>>=R run	Aff	1000000	2.53 s	194.15 ms	2.38 s	2.81 s
>>=R run	Ef	1000000	2.22 s	325.62 ms	1.74 s	2.44 s
>>=L build	Aff	20000	38.49 μs	70.62 μs	583.00 ns	144.29 μs
>>=L build	Ef	20000	38.12 μs	62.43 μs	618.00 ns	130.78 μs
>>=L run	Aff	20000	53.55 ms	5.38 ms	46.70 ms	59.74 ms
>>=L run	Ef	20000	22.72 ms	10.96 ms	10.18 ms	32.41 ms
>>=L build	Aff	50000	14.97 μs	28.53 μs	582.00 ns	57.77 μs
>>=L build	Ef	50000	33.43 μs	48.62 μs	979.00 ns	104.18 μs
>>=L run	Aff	50000	246.69 ms	77.03 ms	146.78 ms	332.67 ms
>>=L run	Ef	50000	113.16 ms	42.66 ms	68.15 ms	166.69 ms
>>=L build	Aff	100000	157.46 μs	295.74 μs	1.52 μs	600.77 μs
>>=L build	Ef	100000	41.53 μs	63.13 μs	2.10 μs	134.95 μs
>>=L run	Aff	100000	403.56 ms	219.39 ms	235.79 ms	707.30 ms
>>=L run	Ef	100000	189.61 ms	43.81 ms	158.03 ms	254.45 ms
>>=L build	Aff	1000000	31.65 μs	57.03 μs	2.60 μs	117.20 μs
>>=L build	Ef	1000000	42.08 μs	54.68 μs	1.35 μs	118.22 μs
>>=L run	Aff	1000000	3.04 s	520.98 ms	2.65 s	3.79 s
>>=L run	Ef	1000000	2.12 s	240.13 ms	1.88 s	2.44 s
map build	Aff	10000	66.50 μs	117.12 μs	2.90 μs	242.09 μs
map build	Ef	10000	33.23 μs	53.26 μs	716.00 ns	112.08 μs
map run	Aff	10000	11.91 ms	6.50 ms	6.28 ms	21.22 ms
map run	Ef	10000	2.61 ms	1.65 ms	1.43 ms	4.99 ms
map build	Aff	20000	30.53 μs	51.70 μs	884.00 ns	107.87 μs
map build	Ef	20000	18.35 μs	29.02 μs	550.00 ns	61.32 μs
map run	Aff	20000	23.16 ms	12.29 ms	10.52 ms	39.54 ms
map run	Ef	20000	4.69 ms	1.48 ms	3.13 ms	6.25 ms
map build	Aff	50000	23.79 μs	40.19 μs	763.00 ns	83.77 μs
map build	Ef	50000	27.86 μs	41.34 μs	1.35 μs	89.08 μs
map run	Aff	50000	64.11 ms	40.20 ms	35.95 ms	123.66 ms
map run	Ef	50000	15.03 ms	5.24 ms	10.19 ms	20.75 ms
map build	Aff	100000	24.35 μs	41.58 μs	578.00 ns	86.40 μs
map build	Ef	100000	22.57 μs	37.04 μs	563.00 ns	77.62 μs
map run	Aff	100000	139.77 ms	66.61 ms	96.66 ms	238.15 ms
map run	Ef	100000	48.34 ms	15.66 ms	33.24 ms	68.65 ms
map build	Aff	1000000	31.40 μs	55.62 μs	622.00 ns	114.59 μs
map build	Ef	1000000	70.68 μs	129.09 μs	532.00 ns	263.95 μs
map run	Aff	1000000	1.42 s	839.80 ms	874.74 ms	2.66 s
map run	Ef	1000000	474.92 ms	54.45 ms	405.97 ms	535.28 ms
map	Ef	10000000	8.76 s	143.23 ms	8.51 s	8.87 s
apply build	Aff	10000	75.54 μs	133.12 μs	901.00 ns	274.51 μs
apply build	Ef	10000	34.96 μs	60.99 μs	545.00 ns	126.07 μs
apply run	Aff	10000	47.83 ms	11.93 ms	36.43 ms	61.47 ms
apply run	Ef	10000	9.84 ms	1.52 ms	8.18 ms	11.69 ms
apply build	Aff	20000	25.97 μs	45.47 μs	493.00 ns	93.91 μs
apply build	Ef	20000	24.17 μs	38.85 μs	396.00 ns	81.69 μs
apply run	Aff	20000	114.99 ms	22.36 ms	96.32 ms	147.39 ms
apply run	Ef	20000	28.27 ms	4.86 ms	22.32 ms	34.13 ms
apply build	Aff	50000	25.58 μs	44.63 μs	459.00 ns	92.25 μs
apply build	Ef	50000	24.69 μs	41.43 μs	448.00 ns	86.39 μs
apply run	Aff	50000	252.50 ms	58.62 ms	220.01 ms	340.35 ms
apply run	Ef	50000	79.21 ms	11.33 ms	70.81 ms	95.53 ms
apply build	Aff	100000	26.68 μs	44.40 μs	535.00 ns	93.05 μs
apply build	Ef	100000	25.22 μs	42.96 μs	387.00 ns	89.25 μs
apply run	Aff	100000	512.79 ms	51.73 ms	451.08 ms	561.61 ms
apply run	Ef	100000	166.15 ms	17.98 ms	147.43 ms	186.04 ms
apply build	Aff	1000000	67.80 μs	120.27 μs	1.32 μs	247.80 μs
apply build	Ef	1000000	32.53 μs	56.14 μs	526.00 ns	116.34 μs
apply run	Aff	1000000	5.88 s	554.09 ms	5.21 s	6.42 s
apply run	Ef	1000000	1.83 s	108.59 ms	1.73 s	1.92 s

safareli · 2017-12-13T20:26:56Z

The implementation is pretty fast and stack safe.
By accepting function as effect node doing FFI is also trivial.

So what you think how could we "integrate" work done in Ef in ps-eff/in PS/etc?

/cc @paf31 @natefaubion @garyb @jdegoes

natefaubion · 2017-12-13T20:27:19Z

These are great benchmarks! It's nice to know that Aff is at least faster than Eff 😈

How about this:

function makeEff(tag, _1, _2) {
  var eff = function () {
    return runEff(eff);
  };
  eff.tag = tag;
  eff._1 = _1;
  eff._2 = _2;
  return eff;
}

And then every Eff node is also a unary function that can interpret itself. This means it can be a drop in replacement.

natefaubion · 2017-12-13T20:30:10Z

That may result in too much heap due to the closure. I wonder if you could do something with this? I guess you'd need to do a function bind.

paf31 · 2017-12-13T20:32:36Z

I think this is interesting, and makes a useful library, but as I said above, not a good default for the core libraries. The default should be as close to the JS as possible, so that the FFI can be simple and the optimizations straightforward.

natefaubion · 2017-12-13T21:56:11Z

Maybe you could take a self reference:

function makeEff(tag, _1, _2) {
  var eff = function thisEff() {
    return runEff(thisEff);
  };
  eff.tag = tag;
  eff._1 = _1;
  eff._2 = _2;
  return eff;
}

I'd be interested to see the difference.

Mapping was happening on pure node, such operation was optimised explicitly in Aff, so switched to do map on effect so it's more fair

safareli · 2017-12-13T23:19:03Z

1. store `meta` prop with `EfMeta` object on function

on timed ["map", "Ef", "10000000"] $ testMap 10000000 test Ef no runs out of memory as Aff did

diff:

Iraklis-MacBook-Pro:purescript-ef safareli$ cat asEff.patch 
diff --git a/bower.json b/bower.json
index f20d14e..1dfcde0 100644
--- a/bower.json
+++ b/bower.json
@@ -18,7 +18,8 @@
   ],
   "dependencies": {
     "purescript-prelude": "^3.0.0",
-    "purescript-eff": "^3.1.0"
+    "purescript-eff": "^3.1.0",
+    "purescript-unsafe-coerce": "^3.0.0"
   },
   "devDependencies": {
     "purescript-foldable-traversable": "^3.6.1",
diff --git a/src/Control/Monad/Ff.js b/src/Control/Monad/Ff.js
index e6b3aee..5101bde 100644
--- a/src/Control/Monad/Ff.js
+++ b/src/Control/Monad/Ff.js
@@ -1,7 +1,9 @@
 "use strict";
 
 
-// Ef a 
+// Ef a = { Eff a, meta :: EfMeta a }
+
+// EfMeta a
 // = () -> a
 // | { tag: "PURE",   _0 :: a,         _1 :: Void  }
 // | { tag: "MAP",    _0 :: b -> a,    _1 :: Ef b }
@@ -15,7 +17,7 @@
 // | { tag: "BIND",       _0 :: a -> Ef b }
 
 
-function Ef(tag, _0, _1) {
+function EfMeta(tag, _0, _1) {
   this.tag = tag;
   this._0 = _0;
   this._1 = _1;
@@ -27,52 +29,53 @@ var APPLY = "APPLY";
 var BIND = "BIND";
 var APPLY_FUNC = "APPLY_FUNC";
 
-exports.liftEffE = function (eff) {
-  return eff;
-};
-
 exports.pureE = function (x) {
-  return new Ef(PURE, x);
+  return asEff(new EfMeta(PURE, x));
 };
 
 exports.mapE = function (f) {
   return function (eff) {
-    return new Ef(MAP, f, eff);
+    return asEff(new EfMeta(MAP, f, eff));
   };
 };
 
 exports.applyE = function (effF) {
   return function (eff) {
-    return new Ef(APPLY, eff, effF);
+    return asEff(new EfMeta(APPLY, eff, effF));
   };
 };
 
 exports.bindE = function (eff) {
   return function (f) {
-    return new Ef(BIND, f, eff);
+    return asEff(new EfMeta(BIND, f, eff));
   };
 };
 
-exports.toEff = function (inputEff) {
-  if (typeof inputEff === "function") {
-    return inputEff;
-  }
+var asEff = function (meta) {
+  var eff = toEff(meta)
+  eff.meta = meta
+  return eff
+}
+
+var toEff = function (inputMeta) {
   return function() {
     var operations = [];
-    var eff = inputEff;
+    var eff = { meta: inputMeta };
+    var meta;
     var res;
     var op;
     var tag;
     effLoop: for (;;) {
-      tag = eff.tag;
-      if (tag !== undefined) {
+      meta = eff.meta
+      if (meta !== undefined) {
+        tag = meta.tag
         if (tag === MAP || tag === BIND || tag === APPLY) {
-          operations.push(eff);
-          eff = eff._1;
+          operations.push(meta);
+          eff = meta._1;
           continue;
         }
         // here `tag === PURE`
-        res = eff._0;
+        res = meta._0;
       } else {
         // here `typeof eff == "function"`
         res = eff();
@@ -85,7 +88,7 @@ exports.toEff = function (inputEff) {
           res = op._0(res);
         } else if (op.tag === APPLY) {
           eff = op._0;
-          operations.push(new Ef(APPLY_FUNC, res));
+          operations.push(new EfMeta(APPLY_FUNC, res));
           continue effLoop;
         } else { // op.tag === BIND
           eff = op._0(res);
diff --git a/src/Control/Monad/Ff.purs b/src/Control/Monad/Ff.purs
index bf890e2..8c2f216 100644
--- a/src/Control/Monad/Ff.purs
+++ b/src/Control/Monad/Ff.purs
@@ -1,7 +1,4 @@
-module Control.Monad.Ef
-  ( Ef
-  , toEff
-  ) where
+module Control.Monad.Ef (Ef) where
 
 import Control.Applicative (class Applicative)
 -- import Control.Applicative (class Applicative, liftA1)
@@ -12,6 +9,7 @@ import Control.Monad (class Monad)
 import Control.Monad.Eff (Eff, kind Effect)
 import Data.Functor (class Functor)
 import Control.Monad.Eff.Class (class MonadEff)
+import Unsafe.Coerce (unsafeCoerce)
 
 
 foreign import data Ef :: # Effect -> Type -> Type
@@ -33,11 +31,8 @@ instance bindEf :: Bind (Ef e) where
 instance monadEf :: Monad (Ef e)
 
 instance monadEEFff :: MonadEff eff (Ef eff) where
-  liftEff = liftEffE
+  liftEff = unsafeCoerce
 
-foreign import toEff :: forall e a. Ef e a -> Eff e a
-
-foreign import liftEffE :: forall e a. Eff e a -> Ef e a
 foreign import mapE :: forall e a b. (a -> b) -> Ef e a -> Ef e b
 foreign import applyE :: forall e a b. Ef e (a -> b) -> Ef e a-> Ef e b
 foreign import pureE :: forall e a. a -> Ef e a
diff --git a/src/Control/Monad/Ff/Class.purs b/src/Control/Monad/Ff/Class.purs
index 0e898e2..2a398a1 100644
--- a/src/Control/Monad/Ff/Class.purs
+++ b/src/Control/Monad/Ff/Class.purs
@@ -1,9 +1,13 @@
-module Control.Monad.Ef.Class where
+module Control.Monad.Ef.Class
+  ( class MonadEf
+  , liftEf
+  ) where
 
 import Control.Category (id)
 import Control.Monad (class Monad)
-import Control.Monad.Ef (Ef, toEff)
+import Control.Monad.Ef (Ef)
 import Control.Monad.Eff (Eff)
+import Unsafe.Coerce (unsafeCoerce)
 
 class Monad m <= MonadEf eff m | m -> eff where
   liftEf :: forall a. Ef eff a -> m a
@@ -12,4 +16,4 @@ instance monadEfEf :: MonadEf eff (Ef eff) where
   liftEf = id
 
 instance monadEfEff :: MonadEf eff (Eff eff) where
-  liftEf = toEff
+  liftEf = unsafeCoerce
\ No newline at end of file

bench:

bench	type	n	mean	stddev	min	max
>>=R build	Eff	100	1.71 μs	15.70 μs	260.00 ns	310.62 μs
>>=R build	Aff	100	785.61 ns	7.02 μs	279.00 ns	149.53 μs
>>=R build	Ef	100	1.12 μs	14.36 μs	360.00 ns	368.61 μs
>>=R run	Eff	100	78.59 μs	83.71 μs	52.63 μs	1.03 ms
>>=R run	Aff	100	114.23 μs	353.41 μs	60.22 μs	3.85 ms
>>=R run	Ef	100	74.37 μs	99.36 μs	55.50 μs	1.06 ms
>>=R build	Eff	500	423.09 ns	2.56 μs	239.00 ns	68.21 μs
>>=R build	Aff	500	498.83 ns	2.66 μs	287.00 ns	57.79 μs
>>=R build	Ef	500	674.51 ns	3.76 μs	366.00 ns	95.76 μs
>>=R run	Eff	500	389.44 μs	297.95 μs	279.99 μs	2.09 ms
>>=R run	Aff	500	343.42 μs	205.56 μs	288.75 μs	1.94 ms
>>=R run	Ef	500	329.51 μs	247.80 μs	261.15 μs	2.74 ms
>>=R build	Eff	1000	472.30 ns	3.28 μs	241.00 ns	74.95 μs
>>=R build	Aff	1000	488.50 ns	2.49 μs	288.00 ns	61.29 μs
>>=R build	Ef	1000	693.37 ns	3.77 μs	373.00 ns	94.18 μs
>>=R run	Eff	1000	758.93 μs	425.89 μs	558.46 μs	3.76 ms
>>=R run	Aff	1000	670.60 μs	255.59 μs	575.84 μs	2.82 ms
>>=R run	Ef	1000	680.97 μs	451.23 μs	517.09 μs	3.50 ms
>>=R build	Eff	2000	387.29 ns	1.98 μs	243.00 ns	49.06 μs
>>=R build	Aff	2000	477.95 ns	2.47 μs	285.00 ns	47.09 μs
>>=R build	Ef	2000	645.97 ns	3.69 μs	370.00 ns	80.18 μs
>>=R run	Eff	2000	1.57 ms	599.66 μs	1.13 ms	6.77 ms
>>=R run	Aff	2000	1.47 ms	647.71 μs	1.15 ms	4.64 ms
>>=R run	Ef	2000	1.39 ms	660.42 μs	1.04 ms	4.83 ms
>>=R build	Eff	4000	405.36 ns	2.38 μs	244.00 ns	62.05 μs
>>=R build	Aff	4000	817.42 ns	2.97 μs	480.00 ns	69.95 μs
>>=R build	Ef	4000	711.11 ns	4.36 μs	380.00 ns	118.39 μs
>>=R run	Eff	4000	3.39 ms	845.87 μs	2.41 ms	12.32 ms
>>=R run	Aff	4000	3.27 ms	1.22 ms	2.32 ms	10.56 ms
>>=R run	Ef	4000	3.26 ms	1.41 ms	2.09 ms	9.02 ms
>>=R build	Eff	8000	413.41 ns	2.50 μs	242.00 ns	59.22 μs
>>=R build	Aff	8000	465.01 ns	2.09 μs	289.00 ns	51.52 μs
>>=R build	Ef	8000	604.56 ns	3.03 μs	369.00 ns	73.41 μs
>>=R run	Eff	8000	8.34 ms	1.85 ms	6.99 ms	43.05 ms
>>=R run	Aff	8000	7.38 ms	2.27 ms	4.69 ms	21.16 ms
>>=R run	Ef	8000	11.09 ms	918.20 μs	6.68 ms	27.11 ms
>>=R build	Eff	10000	426.69 ns	2.65 μs	236.00 ns	68.14 μs
>>=R build	Aff	10000	501.91 ns	3.54 μs	287.00 ns	104.13 μs
>>=R build	Ef	10000	830.32 ns	5.52 μs	357.00 ns	122.04 μs
>>=R run	Eff	10000	67.56 ms	16.43 ms	17.30 ms	121.04 ms
>>=R run	Aff	10000	12.14 ms	1.15 ms	6.43 ms	29.68 ms
>>=R run	Ef	10000	11.13 ms	1.03 ms	7.87 ms	26.05 ms
>>=L build	Eff	100	688.34 ns	8.74 μs	242.00 ns	222.75 μs
>>=L build	Aff	100	479.38 ns	2.86 μs	283.00 ns	80.35 μs
>>=L build	Ef	100	614.15 ns	3.33 μs	374.00 ns	90.97 μs
>>=L run	Eff	100	86.19 μs	162.23 μs	54.91 μs	2.32 ms
>>=L run	Aff	100	83.65 μs	111.48 μs	66.41 μs	2.28 ms
>>=L run	Ef	100	75.63 μs	139.61 μs	56.95 μs	2.44 ms
>>=L build	Eff	500	409.27 ns	2.24 μs	250.00 ns	56.94 μs
>>=L build	Aff	500	463.48 ns	2.29 μs	289.00 ns	61.53 μs
>>=L build	Ef	500	586.78 ns	2.80 μs	374.00 ns	71.66 μs
>>=L run	Eff	500	424.87 μs	404.61 μs	289.11 μs	2.84 ms
>>=L run	Aff	500	385.35 μs	209.29 μs	321.56 μs	2.47 ms
>>=L run	Ef	500	339.54 μs	251.37 μs	271.77 μs	2.59 ms
>>=L build	Eff	1000	440.45 ns	2.88 μs	243.00 ns	75.65 μs
>>=L build	Aff	1000	450.12 ns	2.34 μs	284.00 ns	63.00 μs
>>=L build	Ef	1000	597.32 ns	3.46 μs	375.00 ns	100.11 μs
>>=L run	Eff	1000	878.17 μs	619.18 μs	561.76 μs	4.61 ms
>>=L run	Aff	1000	776.17 μs	311.28 μs	640.45 μs	3.03 ms
>>=L run	Ef	1000	687.25 μs	360.89 μs	538.21 μs	2.92 ms
>>=L build	Eff	2000	387.05 ns	2.11 μs	243.00 ns	53.99 μs
>>=L build	Aff	2000	442.71 ns	2.17 μs	283.00 ns	50.22 μs
>>=L build	Ef	2000	595.97 ns	2.76 μs	373.00 ns	70.33 μs
>>=L run	Eff	2000	1.87 ms	880.22 μs	1.28 ms	8.48 ms
>>=L run	Aff	2000	1.56 ms	433.59 μs	1.29 ms	4.66 ms
>>=L run	Ef	2000	1.90 ms	712.17 μs	1.46 ms	4.82 ms
>>=L build	Eff	4000	393.94 ns	2.70 μs	247.00 ns	83.52 μs
>>=L build	Aff	4000	410.92 ns	2.04 μs	285.00 ns	62.55 μs
>>=L build	Ef	4000	550.97 ns	2.68 μs	377.00 ns	81.39 μs
>>=L run	Eff	4000	4.12 ms	1.43 ms	2.71 ms	19.78 ms
>>=L run	Aff	4000	3.79 ms	1.36 ms	2.65 ms	7.87 ms
>>=L run	Ef	4000	3.85 ms	940.81 μs	2.98 ms	6.82 ms
>>=L build	Eff	8000	350.50 ns	1.74 μs	245.00 ns	54.50 μs
>>=L build	Aff	8000	413.31 ns	2.10 μs	286.00 ns	65.69 μs
>>=L build	Ef	8000	521.19 ns	2.52 μs	380.00 ns	78.53 μs
>>=L run	Eff	8000	14.89 ms	2.89 ms	8.47 ms	53.77 ms
>>=L run	Aff	8000	7.94 ms	2.25 ms	5.73 ms	22.69 ms
>>=L run	Ef	8000	12.69 ms	629.11 μs	9.06 ms	16.57 ms
map build	Eff	100	603.87 ns	6.99 μs	242.00 ns	165.81 μs
map build	Aff	100	393.05 ns	1.89 μs	284.00 ns	59.30 μs
map build	Ef	100	524.01 ns	2.32 μs	383.00 ns	71.62 μs
map run	Eff	100	66.05 μs	156.95 μs	41.84 μs	3.04 ms
map run	Aff	100	34.61 μs	32.64 μs	29.93 μs	979.52 μs
map run	Ef	100	43.46 μs	31.86 μs	36.85 μs	939.45 μs
map build	Eff	500	377.58 ns	2.33 μs	254.00 ns	73.18 μs
map build	Aff	500	766.19 ns	5.03 μs	290.00 ns	99.31 μs
map build	Ef	500	544.15 ns	2.53 μs	383.00 ns	77.67 μs
map run	Eff	500	291.56 μs	301.25 μs	204.47 μs	2.42 ms
map run	Aff	500	152.43 μs	59.56 μs	138.00 μs	1.60 ms
map run	Ef	500	191.71 μs	43.82 μs	173.74 μs	1.15 ms
map build	Eff	1000	366.88 ns	2.04 μs	249.00 ns	64.72 μs
map build	Aff	1000	590.20 ns	4.10 μs	286.00 ns	87.98 μs
map build	Ef	1000	560.40 ns	3.35 μs	380.00 ns	104.16 μs
map run	Eff	1000	591.60 μs	461.28 μs	407.66 μs	3.17 ms
map run	Aff	1000	307.60 μs	76.74 μs	277.95 μs	1.96 ms
map run	Ef	1000	381.15 μs	63.60 μs	348.81 μs	1.30 ms
map build	Eff	2000	352.88 ns	1.62 μs	248.00 ns	51.23 μs
map build	Aff	2000	550.17 ns	3.41 μs	290.00 ns	76.72 μs
map build	Ef	2000	615.11 ns	3.44 μs	382.00 ns	105.38 μs
map run	Eff	2000	1.27 ms	727.35 μs	829.96 μs	5.44 ms
map run	Aff	2000	705.77 μs	139.14 μs	636.44 μs	2.68 ms
map run	Ef	2000	758.22 μs	86.64 μs	690.01 μs	1.74 ms
map build	Eff	4000	379.64 ns	1.89 μs	247.00 ns	58.91 μs
map build	Aff	4000	610.16 ns	4.25 μs	359.00 ns	130.67 μs
map build	Ef	4000	884.60 ns	6.79 μs	381.00 ns	212.99 μs
map run	Eff	4000	2.63 ms	1.11 ms	1.79 ms	9.18 ms
map run	Aff	4000	1.65 ms	200.53 μs	1.54 ms	4.51 ms
map run	Ef	4000	1.57 ms	258.31 μs	1.40 ms	3.23 ms
map build	Eff	5000	402.61 ns	2.72 μs	247.00 ns	85.80 μs
map build	Aff	5000	576.00 ns	3.46 μs	361.00 ns	108.35 μs
map build	Ef	5000	569.57 ns	4.04 μs	380.00 ns	126.99 μs
map run	Eff	5000	4.59 ms	2.13 ms	2.33 ms	11.31 ms
map run	Aff	5000	2.12 ms	283.92 μs	1.90 ms	5.19 ms
map run	Ef	5000	2.00 ms	324.50 μs	1.74 ms	4.25 ms
apply build	Eff	100	753.18 ns	7.39 μs	242.00 ns	163.91 μs
apply build	Aff	100	623.71 ns	3.77 μs	357.00 ns	116.91 μs
apply build	Ef	100	591.71 ns	2.71 μs	414.00 ns	82.43 μs
apply run	Eff	100	174.97 μs	232.80 μs	115.41 μs	1.97 ms
apply run	Aff	100	204.46 μs	119.15 μs	171.84 μs	1.61 ms
apply run	Ef	100	143.56 μs	82.78 μs	121.26 μs	1.06 ms
apply build	Eff	500	436.73 ns	2.30 μs	247.00 ns	67.09 μs
apply build	Aff	500	571.61 ns	2.94 μs	359.00 ns	91.65 μs
apply build	Ef	500	599.88 ns	4.19 μs	382.00 ns	130.64 μs
apply run	Eff	500	964.92 μs	679.49 μs	615.91 μs	5.04 ms
apply run	Aff	500	1.04 ms	393.97 μs	849.20 μs	3.16 ms
apply run	Ef	500	716.98 μs	208.73 μs	617.84 μs	2.01 ms
apply build	Eff	1000	360.04 ns	2.08 μs	246.00 ns	66.11 μs
apply build	Aff	1000	532.21 ns	2.60 μs	363.00 ns	81.88 μs
apply build	Ef	1000	576.99 ns	2.91 μs	381.00 ns	89.14 μs
apply run	Eff	1000	2.11 ms	1.10 ms	1.28 ms	7.17 ms
apply run	Aff	1000	2.14 ms	644.01 μs	1.70 ms	5.10 ms
apply run	Ef	1000	1.47 ms	270.63 μs	1.26 ms	2.88 ms
apply build	Eff	2000	383.30 ns	2.18 μs	256.00 ns	68.31 μs
apply build	Aff	2000	549.86 ns	2.60 μs	364.00 ns	80.79 μs
apply build	Ef	2000	590.85 ns	4.25 μs	383.00 ns	132.21 μs
apply run	Eff	2000	5.42 ms	1.99 ms	2.93 ms	14.68 ms
apply run	Aff	2000	4.57 ms	1.12 ms	3.46 ms	8.16 ms
apply run	Ef	2000	2.98 ms	434.08 μs	2.52 ms	5.39 ms
apply build	Eff	4000	359.07 ns	1.70 μs	246.00 ns	53.87 μs
apply build	Aff	4000	632.66 ns	3.66 μs	373.00 ns	115.70 μs
apply build	Ef	4000	584.80 ns	3.62 μs	388.00 ns	113.48 μs
apply run	Eff	4000	26.81 ms	5.19 ms	15.51 ms	50.33 ms
apply run	Aff	4000	9.41 ms	1.72 ms	6.91 ms	15.40 ms
apply run	Ef	4000	5.99 ms	526.85 μs	5.39 ms	9.36 ms
apply build	Eff	5000	342.75 ns	1.87 μs	245.00 ns	59.27 μs
apply build	Aff	5000	530.40 ns	3.14 μs	353.00 ns	98.75 μs
apply build	Ef	5000	552.04 ns	3.11 μs	380.00 ns	97.23 μs
apply run	Eff	5000	72.99 ms	23.32 ms	36.56 ms	158.01 ms
apply run	Aff	5000	12.65 ms	1.35 ms	11.74 ms	31.80 ms
apply run	Ef	5000	9.81 ms	2.06 ms	6.95 ms	19.70 ms

extended bench:

bench	type	n	mean	stddev	min	max
>>=R build	Aff	20000	59.24 μs	103.16 μs	3.06 μs	213.81 μs
>>=R build	Ef	20000	69.12 μs	119.96 μs	781.00 ns	248.26 μs
>>=R run	Aff	20000	58.35 ms	6.37 ms	49.21 ms	63.38 ms
>>=R run	Ef	20000	67.08 ms	21.99 ms	45.10 ms	95.20 ms
>>=R build	Aff	50000	25.13 μs	47.41 μs	1.11 μs	96.23 μs
>>=R build	Ef	50000	26.24 μs	40.09 μs	652.00 ns	85.16 μs
>>=R run	Aff	50000	133.90 ms	25.01 ms	115.01 ms	168.75 ms
>>=R run	Ef	50000	137.69 ms	38.34 ms	113.59 ms	194.91 ms
>>=R build	Aff	100000	19.27 μs	36.70 μs	663.00 ns	74.32 μs
>>=R build	Ef	100000	37.30 μs	58.93 μs	1.23 μs	124.89 μs
>>=R run	Aff	100000	220.46 ms	52.33 ms	192.06 ms	298.87 ms
>>=R run	Ef	100000	289.22 ms	18.41 ms	263.64 ms	303.31 ms
>>=R build	Aff	1000000	47.52 μs	87.66 μs	648.00 ns	178.85 μs
>>=R build	Ef	1000000	32.69 μs	51.68 μs	594.00 ns	109.09 μs
>>=R run	Aff	1000000	2.43 s	102.44 ms	2.30 s	2.55 s
>>=R run	Ef	1000000	3.08 s	433.87 ms	2.82 s	3.73 s
>>=L build	Aff	20000	43.12 μs	79.90 μs	560.00 ns	162.85 μs
>>=L build	Ef	20000	36.79 μs	59.48 μs	680.00 ns	124.98 μs
>>=L run	Aff	20000	56.54 ms	7.98 ms	47.12 ms	63.54 ms
>>=L run	Ef	20000	62.14 ms	27.68 ms	41.02 ms	100.93 ms
>>=L build	Aff	50000	16.64 μs	31.09 μs	854.00 ns	63.27 μs
>>=L build	Ef	50000	25.03 μs	36.60 μs	646.00 ns	78.30 μs
>>=L run	Aff	50000	147.03 ms	39.37 ms	113.96 ms	199.06 ms
>>=L run	Ef	50000	148.35 ms	33.50 ms	116.78 ms	191.63 ms
>>=L build	Aff	100000	56.91 μs	107.77 μs	1.62 μs	218.55 μs
>>=L build	Ef	100000	33.46 μs	52.53 μs	877.00 ns	111.08 μs
>>=L run	Aff	100000	290.46 ms	59.60 ms	241.86 ms	371.72 ms
>>=L run	Ef	100000	307.69 ms	31.11 ms	271.62 ms	347.12 ms
>>=L build	Aff	1000000	68.43 μs	128.18 μs	632.00 ns	260.54 μs
>>=L build	Ef	1000000	30.22 μs	47.15 μs	626.00 ns	99.75 μs
>>=L run	Aff	1000000	3.02 s	217.05 ms	2.84 s	3.33 s
>>=L run	Ef	1000000	3.41 s	384.64 ms	3.12 s	3.95 s
map build	Aff	10000	65.59 μs	115.13 μs	3.15 μs	238.10 μs
map build	Ef	10000	65.39 μs	114.29 μs	675.00 ns	236.10 μs
map run	Aff	10000	15.10 ms	9.14 ms	6.85 ms	27.63 ms
map run	Ef	10000	7.62 ms	3.64 ms	3.64 ms	11.95 ms
map build	Aff	20000	25.98 μs	40.96 μs	732.00 ns	86.48 μs
map build	Ef	20000	41.60 μs	65.87 μs	681.00 ns	138.89 μs
map run	Aff	20000	25.72 ms	11.23 ms	11.42 ms	38.88 ms
map run	Ef	20000	15.47 ms	6.87 ms	9.05 ms	25.01 ms
map build	Aff	50000	30.67 μs	52.96 μs	652.00 ns	109.75 μs
map build	Ef	50000	26.61 μs	37.17 μs	693.00 ns	79.90 μs
map run	Aff	50000	63.48 ms	36.50 ms	35.91 ms	117.19 ms
map run	Ef	50000	60.90 ms	18.52 ms	41.93 ms	80.25 ms
map build	Aff	100000	37.18 μs	62.22 μs	1.07 μs	129.88 μs
map build	Ef	100000	24.45 μs	35.40 μs	677.00 ns	75.98 μs
map run	Aff	100000	145.08 ms	73.58 ms	95.31 ms	253.62 ms
map run	Ef	100000	116.41 ms	28.09 ms	98.57 ms	158.30 ms
map build	Aff	1000000	23.51 μs	38.48 μs	563.00 ns	80.67 μs
map build	Ef	1000000	31.83 μs	49.70 μs	637.00 ns	105.11 μs
map run	Aff	1000000	1.44 s	810.81 ms	877.74 ms	2.63 s
map run	Ef	1000000	1.24 s	182.04 ms	1.01 s	1.39 s
apply build	Aff	10000	61.05 μs	109.44 μs	638.00 ns	224.79 μs
apply build	Ef	10000	41.17 μs	64.33 μs	703.00 ns	136.05 μs
apply run	Aff	10000	47.19 ms	12.73 ms	36.01 ms	63.88 ms
apply run	Ef	10000	31.67 ms	11.11 ms	17.72 ms	41.68 ms
apply build	Aff	20000	29.84 μs	52.05 μs	663.00 ns	107.62 μs
apply build	Ef	20000	26.61 μs	40.97 μs	647.00 ns	86.91 μs
apply run	Aff	20000	112.90 ms	25.53 ms	91.10 ms	149.44 ms
apply run	Ef	20000	75.06 ms	11.76 ms	66.09 ms	91.28 ms
apply build	Aff	50000	22.61 μs	38.23 μs	600.00 ns	79.62 μs
apply build	Ef	50000	40.43 μs	66.50 μs	655.00 ns	139.23 μs
apply run	Aff	50000	253.01 ms	63.15 ms	217.79 ms	347.65 ms
apply run	Ef	50000	212.55 ms	33.33 ms	194.51 ms	262.46 ms
apply build	Aff	100000	38.94 μs	60.60 μs	3.37 μs	129.58 μs
apply build	Ef	100000	44.52 μs	71.15 μs	617.00 ns	149.68 μs
apply run	Aff	100000	497.15 ms	61.48 ms	438.15 ms	560.08 ms
apply run	Ef	100000	434.29 ms	47.56 ms	392.31 ms	495.25 ms
apply build	Aff	1000000	107.85 μs	191.44 μs	1.24 μs	394.36 μs
apply build	Ef	1000000	67.12 μs	117.97 μs	689.00 ns	243.41 μs
apply run	Aff	1000000	5.97 s	442.12 ms	5.52 s	6.41 s
apply run	Ef	1000000	5.23 s	433.15 ms	4.64 s	5.64 s

natefaubion · 2017-12-13T23:23:04Z

So in some cases, it's worse off, but you are allocating an extra meta object. The reason why I suggested adding properties to the function itself is to avoid the allocation of the meta object.

safareli · 2017-12-13T23:28:44Z

2. store `tag`, `_0` and `_1` props on function directly

diff

diff --git a/src/Control/Monad/Ff.js b/src/Control/Monad/Ff.js
index e6b3aee..1fb94ac 100644
--- a/src/Control/Monad/Ff.js
+++ b/src/Control/Monad/Ff.js
@@ -1,12 +1,12 @@
 "use strict";
 
 
-// Ef a 
+// Ef a
 // = () -> a
-// | { tag: "PURE",   _0 :: a,         _1 :: Void  }
-// | { tag: "MAP",    _0 :: b -> a,    _1 :: Ef b }
-// | { tag: "APPLY",  _0 :: Ef b,      _1 :: Ef (b -> a) }
-// | { tag: "BIND",   _0 :: b -> Ef a, _1 :: Ef b }
+// | { Ef a, tag: "PURE",   _0 :: a,         _1 :: Void  }
+// | { Ef a, tag: "MAP",    _0 :: b -> a,    _1 :: Ef b }
+// | { Ef a, tag: "APPLY",  _0 :: Ef b,      _1 :: Ef (b -> a) }
+// | { Ef a, tag: "BIND",   _0 :: b -> Ef a, _1 :: Ef b }
 
 // Operation a b
 // = { tag: "MAP",        _0 :: a -> b }
@@ -15,57 +15,51 @@
 // | { tag: "BIND",       _0 :: a -> Ef b }
 
 
-function Ef(tag, _0, _1) {
-  this.tag = tag;
-  this._0 = _0;
-  this._1 = _1;
-}
-
 var PURE = "PURE";
 var MAP = "MAP";
 var APPLY = "APPLY";
 var BIND = "BIND";
 var APPLY_FUNC = "APPLY_FUNC";
 
-exports.liftEffE = function (eff) {
-  return eff;
-};
-
 exports.pureE = function (x) {
-  return new Ef(PURE, x);
+  return mkEf(PURE, x);
 };
 
 exports.mapE = function (f) {
   return function (eff) {
-    return new Ef(MAP, f, eff);
+    return mkEf(MAP, f, eff);
   };
 };
 
 exports.applyE = function (effF) {
   return function (eff) {
-    return new Ef(APPLY, eff, effF);
+    return mkEf(APPLY, eff, effF);
   };
 };
 
 exports.bindE = function (eff) {
   return function (f) {
-    return new Ef(BIND, f, eff);
+    return mkEf(BIND, f, eff);
   };
 };
 
-exports.toEff = function (inputEff) {
-  if (typeof inputEff === "function") {
-    return inputEff;
-  }
-  return function() {
+
+var mkEf = function (tag, _0, _1) {
+  var eff = function eff_() { return toEff(eff_) }
+  eff.tag = tag
+  eff._0 = _0
+  eff._1 = _1
+  return eff
+}
+
+var toEff = function (eff) {
     var operations = [];
-    var eff = inputEff;
     var res;
     var op;
     var tag;
     effLoop: for (;;) {
-      tag = eff.tag;
-      if (tag !== undefined) {
+      if (eff.tag !== undefined) {
+        tag = eff.tag
         if (tag === MAP || tag === BIND || tag === APPLY) {
           operations.push(eff);
           eff = eff._1;
@@ -94,5 +88,4 @@ exports.toEff = function (inputEff) {
       }
       return res;
     }
-  };
 };
\ No newline at end of file

bench

bench	type	n	mean	stddev	min	max
>>=R build	Eff	100	1.56 μs	15.28 μs	231.00 ns	330.73 μs
>>=R build	Aff	100	804.35 ns	7.47 μs	275.00 ns	155.00 μs
>>=R build	Ef	100	844.88 ns	9.19 μs	279.00 ns	233.30 μs
>>=R run	Eff	100	79.80 μs	78.14 μs	52.77 μs	1.02 ms
>>=R run	Aff	100	115.69 μs	349.13 μs	60.63 μs	3.84 ms
>>=R run	Ef	100	68.76 μs	65.05 μs	55.41 μs	801.61 μs
>>=R build	Eff	500	429.48 ns	2.42 μs	239.00 ns	62.72 μs
>>=R build	Aff	500	453.56 ns	2.48 μs	282.00 ns	58.27 μs
>>=R build	Ef	500	513.77 ns	2.85 μs	282.00 ns	60.86 μs
>>=R run	Eff	500	387.53 μs	308.36 μs	274.28 μs	2.49 ms
>>=R run	Aff	500	344.03 μs	210.55 μs	287.03 μs	1.82 ms
>>=R run	Ef	500	296.10 μs	186.94 μs	245.79 μs	1.53 ms
>>=R build	Eff	1000	446.37 ns	2.70 μs	239.00 ns	67.54 μs
>>=R build	Aff	1000	477.74 ns	2.38 μs	280.00 ns	57.73 μs
>>=R build	Ef	1000	521.51 ns	3.23 μs	281.00 ns	61.50 μs
>>=R run	Eff	1000	771.10 μs	438.47 μs	553.35 μs	3.79 ms
>>=R run	Aff	1000	693.87 μs	285.45 μs	580.55 μs	3.16 ms
>>=R run	Ef	1000	526.83 μs	356.63 μs	397.50 μs	2.16 ms
>>=R build	Eff	2000	402.33 ns	2.28 μs	236.00 ns	56.57 μs
>>=R build	Aff	2000	531.05 ns	2.46 μs	280.00 ns	51.87 μs
>>=R build	Ef	2000	535.98 ns	3.40 μs	282.00 ns	63.77 μs
>>=R run	Eff	2000	1.63 ms	703.93 μs	1.11 ms	6.49 ms
>>=R run	Aff	2000	1.50 ms	667.88 μs	1.14 ms	4.18 ms
>>=R run	Ef	2000	998.89 μs	594.59 μs	704.75 μs	3.24 ms
>>=R build	Eff	4000	420.46 ns	2.34 μs	237.00 ns	53.21 μs
>>=R build	Aff	4000	456.66 ns	2.11 μs	281.00 ns	46.66 μs
>>=R build	Ef	4000	534.09 ns	3.04 μs	283.00 ns	63.99 μs
>>=R run	Eff	4000	3.32 ms	801.06 μs	2.37 ms	12.29 ms
>>=R run	Aff	4000	3.17 ms	1.16 ms	2.33 ms	8.35 ms
>>=R run	Ef	4000	1.89 ms	820.67 μs	1.33 ms	4.17 ms
>>=R build	Eff	8000	423.48 ns	2.66 μs	237.00 ns	67.36 μs
>>=R build	Aff	8000	459.44 ns	2.26 μs	280.00 ns	55.02 μs
>>=R build	Ef	8000	548.23 ns	3.78 μs	283.00 ns	83.15 μs
>>=R run	Eff	8000	7.89 ms	1.72 ms	6.96 ms	38.57 ms
>>=R run	Aff	8000	7.27 ms	2.24 ms	4.70 ms	19.45 ms
>>=R run	Ef	8000	3.94 ms	944.80 μs	2.92 ms	6.40 ms
>>=R build	Eff	10000	428.80 ns	2.45 μs	240.00 ns	59.21 μs
>>=R build	Aff	10000	443.61 ns	2.43 μs	280.00 ns	65.63 μs
>>=R build	Ef	10000	669.84 ns	5.22 μs	268.00 ns	122.86 μs
>>=R run	Eff	10000	64.64 ms	15.12 ms	18.11 ms	105.76 ms
>>=R run	Aff	10000	12.01 ms	1.21 ms	6.54 ms	27.84 ms
>>=R run	Ef	10000	5.44 ms	1.68 ms	3.59 ms	8.89 ms
>>=L build	Eff	100	762.68 ns	8.91 μs	243.00 ns	221.87 μs
>>=L build	Aff	100	426.06 ns	2.00 μs	280.00 ns	44.38 μs
>>=L build	Ef	100	473.54 ns	2.47 μs	286.00 ns	60.47 μs
>>=L run	Eff	100	82.15 μs	153.77 μs	53.50 μs	2.58 ms
>>=L run	Aff	100	83.57 μs	94.82 μs	66.51 μs	1.18 ms
>>=L run	Ef	100	69.68 μs	103.69 μs	53.90 μs	1.57 ms
>>=L build	Eff	500	407.37 ns	2.40 μs	243.00 ns	63.91 μs
>>=L build	Aff	500	437.43 ns	2.09 μs	285.00 ns	46.83 μs
>>=L build	Ef	500	501.64 ns	2.94 μs	288.00 ns	67.99 μs
>>=L run	Eff	500	430.74 μs	426.52 μs	276.61 μs	3.23 ms
>>=L run	Aff	500	382.30 μs	199.80 μs	324.92 μs	1.57 ms
>>=L run	Ef	500	301.45 μs	214.48 μs	238.89 μs	1.88 ms
>>=L build	Eff	1000	455.65 ns	2.76 μs	245.00 ns	63.94 μs
>>=L build	Aff	1000	427.56 ns	2.20 μs	281.00 ns	57.41 μs
>>=L build	Ef	1000	521.37 ns	3.22 μs	286.00 ns	83.16 μs
>>=L run	Eff	1000	888.88 μs	617.73 μs	570.93 μs	5.32 ms
>>=L run	Aff	1000	768.97 μs	291.08 μs	647.39 μs	2.82 ms
>>=L run	Ef	1000	496.86 μs	282.55 μs	389.00 μs	1.97 ms
>>=L build	Eff	2000	411.58 ns	2.30 μs	244.00 ns	52.92 μs
>>=L build	Aff	2000	425.66 ns	1.87 μs	282.00 ns	44.74 μs
>>=L build	Ef	2000	490.73 ns	2.72 μs	288.00 ns	60.70 μs
>>=L run	Eff	2000	1.88 ms	848.57 μs	1.27 ms	7.59 ms
>>=L run	Aff	2000	1.54 ms	378.53 μs	1.31 ms	2.89 ms
>>=L run	Ef	2000	1.02 ms	611.88 μs	717.28 μs	3.18 ms
>>=L build	Eff	4000	389.58 ns	2.00 μs	244.00 ns	60.37 μs
>>=L build	Aff	4000	425.28 ns	2.10 μs	283.00 ns	63.38 μs
>>=L build	Ef	4000	470.53 ns	2.71 μs	291.00 ns	81.62 μs
>>=L run	Eff	4000	3.97 ms	1.33 ms	2.71 ms	18.20 ms
>>=L run	Aff	4000	3.78 ms	1.31 ms	2.68 ms	7.55 ms
>>=L run	Ef	4000	2.07 ms	844.54 μs	1.44 ms	4.78 ms
>>=L build	Eff	8000	351.05 ns	1.60 μs	245.00 ns	50.67 μs
>>=L build	Aff	8000	396.03 ns	1.91 μs	285.00 ns	59.90 μs
>>=L build	Ef	8000	424.15 ns	2.14 μs	295.00 ns	66.03 μs
>>=L run	Eff	8000	14.18 ms	2.62 ms	7.32 ms	51.36 ms
>>=L run	Aff	8000	7.78 ms	2.04 ms	5.77 ms	13.86 ms
>>=L run	Ef	8000	4.26 ms	951.65 μs	3.14 ms	7.86 ms
map build	Eff	100	811.38 ns	8.59 μs	314.00 ns	189.12 μs
map build	Aff	100	388.66 ns	1.46 μs	284.00 ns	46.29 μs
map build	Ef	100	417.27 ns	1.92 μs	294.00 ns	59.88 μs
map run	Eff	100	65.52 μs	169.18 μs	41.48 μs	3.54 ms
map run	Aff	100	34.76 μs	37.12 μs	29.39 μs	1.11 ms
map run	Ef	100	30.70 μs	12.90 μs	26.91 μs	289.64 μs
map build	Eff	500	406.55 ns	2.33 μs	269.00 ns	73.86 μs
map build	Aff	500	632.89 ns	4.70 μs	317.00 ns	125.04 μs
map build	Ef	500	480.03 ns	3.67 μs	299.00 ns	115.07 μs
map run	Eff	500	308.57 μs	325.95 μs	202.75 μs	2.79 ms
map run	Aff	500	159.48 μs	65.32 μs	138.09 μs	1.70 ms
map run	Ef	500	143.59 μs	38.31 μs	123.87 μs	1.02 ms
map build	Eff	1000	374.13 ns	2.34 μs	246.00 ns	74.22 μs
map build	Aff	1000	591.15 ns	3.75 μs	292.00 ns	84.37 μs
map build	Ef	1000	418.45 ns	2.09 μs	293.00 ns	65.14 μs
map run	Eff	1000	606.13 μs	485.10 μs	404.96 μs	3.37 ms
map run	Aff	1000	301.25 μs	78.10 μs	271.71 μs	1.97 ms
map run	Ef	1000	266.60 μs	45.69 μs	246.73 μs	1.23 ms
map build	Eff	2000	400.07 ns	2.26 μs	250.00 ns	69.92 μs
map build	Aff	2000	595.01 ns	4.38 μs	286.00 ns	108.34 μs
map build	Ef	2000	456.99 ns	3.12 μs	294.00 ns	98.00 μs
map run	Eff	2000	1.26 ms	728.33 μs	818.29 μs	5.47 ms
map run	Aff	2000	715.17 μs	140.31 μs	637.98 μs	2.88 ms
map run	Ef	2000	444.46 μs	99.49 μs	359.40 μs	1.52 ms
map build	Eff	4000	379.40 ns	2.41 μs	246.00 ns	76.38 μs
map build	Aff	4000	561.97 ns	2.82 μs	356.00 ns	84.31 μs
map build	Ef	4000	506.10 ns	2.73 μs	296.00 ns	79.10 μs
map run	Eff	4000	2.70 ms	1.15 ms	1.76 ms	8.80 ms
map run	Aff	4000	1.73 ms	396.25 μs	1.54 ms	6.99 ms
map run	Ef	4000	772.77 μs	188.27 μs	677.22 μs	2.44 ms
map build	Eff	5000	347.42 ns	1.83 μs	250.00 ns	58.20 μs
map build	Aff	5000	618.04 ns	2.85 μs	369.00 ns	87.33 μs
map build	Ef	5000	465.33 ns	3.01 μs	295.00 ns	94.03 μs
map run	Eff	5000	5.00 ms	2.58 ms	2.31 ms	19.12 ms
map run	Aff	5000	2.12 ms	283.83 μs	1.92 ms	5.29 ms
map run	Ef	5000	1.02 ms	247.71 μs	862.65 μs	3.08 ms
apply build	Eff	100	723.69 ns	7.80 μs	244.00 ns	184.79 μs
apply build	Aff	100	521.72 ns	2.61 μs	356.00 ns	82.21 μs
apply build	Ef	100	418.62 ns	1.97 μs	294.00 ns	61.65 μs
apply run	Eff	100	171.71 μs	230.24 μs	114.77 μs	2.29 ms
apply run	Aff	100	211.13 μs	145.15 μs	173.34 μs	2.86 ms
apply run	Ef	100	118.80 μs	82.30 μs	95.33 μs	1.03 ms
apply build	Eff	500	384.12 ns	2.86 μs	250.00 ns	90.72 μs
apply build	Aff	500	580.11 ns	3.87 μs	369.00 ns	122.00 μs
apply build	Ef	500	487.41 ns	3.85 μs	293.00 ns	121.05 μs
apply run	Eff	500	1.14 ms	838.02 μs	609.47 μs	6.60 ms
apply run	Aff	500	1.22 ms	483.01 μs	892.31 μs	4.60 ms
apply run	Ef	500	431.78 μs	218.46 μs	318.22 μs	2.39 ms
apply build	Eff	1000	373.86 ns	2.46 μs	246.00 ns	77.60 μs
apply build	Aff	1000	551.04 ns	2.85 μs	366.00 ns	89.89 μs
apply build	Ef	1000	527.46 ns	3.66 μs	304.00 ns	113.54 μs
apply run	Eff	1000	2.14 ms	1.07 ms	1.25 ms	6.54 ms
apply run	Aff	1000	2.19 ms	643.58 μs	1.74 ms	5.09 ms
apply run	Ef	1000	734.70 μs	307.66 μs	554.76 μs	3.38 ms
apply build	Eff	2000	378.94 ns	2.55 μs	249.00 ns	80.17 μs
apply build	Aff	2000	549.62 ns	3.64 μs	354.00 ns	114.86 μs
apply build	Ef	2000	459.91 ns	3.49 μs	294.00 ns	109.33 μs
apply run	Eff	2000	5.48 ms	2.06 ms	2.98 ms	14.43 ms
apply run	Aff	2000	4.80 ms	1.16 ms	3.57 ms	9.39 ms
apply run	Ef	2000	1.51 ms	570.43 μs	1.10 ms	5.05 ms
apply build	Eff	4000	341.78 ns	1.53 μs	245.00 ns	47.56 μs
apply build	Aff	4000	537.14 ns	3.24 μs	356.00 ns	102.04 μs
apply build	Ef	4000	446.32 ns	2.40 μs	294.00 ns	74.13 μs
apply run	Eff	4000	28.57 ms	6.45 ms	15.18 ms	68.81 ms
apply run	Aff	4000	10.27 ms	2.03 ms	7.15 ms	16.72 ms
apply run	Ef	4000	3.05 ms	791.76 μs	2.25 ms	7.27 ms
apply build	Eff	5000	565.76 ns	3.19 μs	247.00 ns	96.82 μs
apply build	Aff	5000	571.16 ns	3.89 μs	367.00 ns	122.49 μs
apply build	Ef	5000	490.75 ns	3.90 μs	292.00 ns	122.65 μs
apply run	Eff	5000	75.52 ms	24.65 ms	36.48 ms	200.52 ms
apply run	Aff	5000	13.04 ms	1.12 ms	12.04 ms	19.40 ms
apply run	Ef	5000	3.92 ms	794.87 μs	3.04 ms	7.80 ms

extended

bench	type	n	mean	stddev	min	max
>>=R build	Aff	20000	58.54 μs	101.60 μs	3.42 μs	210.77 μs
>>=R build	Ef	20000	42.49 μs	71.77 μs	648.00 ns	149.42 μs
>>=R run	Aff	20000	54.26 ms	9.77 ms	46.26 ms	66.60 ms
>>=R run	Ef	20000	24.71 ms	9.76 ms	15.62 ms	34.52 ms
>>=R build	Aff	50000	14.31 μs	27.20 μs	563.00 ns	55.11 μs
>>=R build	Ef	50000	20.52 μs	32.27 μs	542.00 ns	68.18 μs
>>=R run	Aff	50000	119.38 ms	22.72 ms	94.43 ms	147.93 ms
>>=R run	Ef	50000	95.20 ms	19.55 ms	77.47 ms	120.74 ms
>>=R build	Aff	100000	15.84 μs	30.26 μs	506.00 ns	61.23 μs
>>=R build	Ef	100000	26.09 μs	40.38 μs	955.00 ns	86.09 μs
>>=R run	Aff	100000	207.89 ms	44.52 ms	182.30 ms	274.56 ms
>>=R run	Ef	100000	169.70 ms	16.15 ms	149.40 ms	185.54 ms
>>=R build	Aff	1000000	18.58 μs	35.87 μs	505.00 ns	72.39 μs
>>=R build	Ef	1000000	24.89 μs	38.34 μs	620.00 ns	81.27 μs
>>=R run	Aff	1000000	2.64 s	408.88 ms	2.24 s	3.19 s
>>=R run	Ef	1000000	2.01 s	47.67 ms	1.95 s	2.06 s
>>=L build	Aff	20000	43.62 μs	78.62 μs	560.00 ns	161.29 μs
>>=L build	Ef	20000	34.85 μs	55.36 μs	516.00 ns	116.63 μs
>>=L run	Aff	20000	55.00 ms	7.67 ms	45.90 ms	64.43 ms
>>=L run	Ef	20000	36.97 ms	14.11 ms	17.88 ms	49.43 ms
>>=L build	Aff	50000	14.06 μs	26.83 μs	547.00 ns	54.30 μs
>>=L build	Ef	50000	28.13 μs	37.51 μs	814.00 ns	83.41 μs
>>=L run	Aff	50000	140.71 ms	33.94 ms	112.37 ms	186.78 ms
>>=L run	Ef	50000	98.82 ms	26.63 ms	77.65 ms	137.68 ms
>>=L build	Aff	100000	17.62 μs	32.71 μs	1.03 μs	66.69 μs
>>=L build	Ef	100000	20.40 μs	31.39 μs	519.00 ns	66.60 μs
>>=L run	Aff	100000	259.10 ms	57.47 ms	226.01 ms	345.05 ms
>>=L run	Ef	100000	194.13 ms	46.25 ms	162.57 ms	262.90 ms
>>=L build	Aff	1000000	12.11 μs	22.95 μs	514.00 ns	46.53 μs
>>=L build	Ef	1000000	25.70 μs	41.27 μs	554.00 ns	86.82 μs
>>=L run	Aff	1000000	2.74 s	167.06 ms	2.59 s	2.98 s
>>=L run	Ef	1000000	2.26 s	54.35 ms	2.20 s	2.31 s
map build	Aff	10000	37.48 μs	67.87 μs	1.68 μs	139.25 μs
map build	Ef	10000	40.36 μs	66.80 μs	591.00 ns	139.63 μs
map run	Aff	10000	12.17 ms	6.61 ms	6.40 ms	21.65 ms
map run	Ef	10000	2.55 ms	1.18 ms	1.65 ms	4.17 ms
map build	Aff	20000	24.15 μs	39.53 μs	758.00 ns	83.19 μs
map build	Ef	20000	19.43 μs	30.59 μs	560.00 ns	64.65 μs
map run	Aff	20000	23.14 ms	11.62 ms	10.26 ms	38.15 ms
map run	Ef	20000	5.05 ms	1.70 ms	3.18 ms	6.56 ms
map build	Aff	50000	28.70 μs	49.63 μs	611.00 ns	102.82 μs
map build	Ef	50000	43.64 μs	75.95 μs	532.00 ns	157.06 μs
map run	Aff	50000	64.76 ms	35.34 ms	40.80 ms	117.17 ms
map run	Ef	50000	21.46 ms	7.10 ms	14.11 ms	30.92 ms
map build	Aff	100000	23.60 μs	37.89 μs	703.00 ns	79.76 μs
map build	Ef	100000	20.55 μs	31.35 μs	604.00 ns	66.64 μs
map run	Aff	100000	139.06 ms	67.94 ms	94.38 ms	239.31 ms
map run	Ef	100000	56.95 ms	11.70 ms	44.71 ms	69.94 ms
map build	Aff	1000000	43.82 μs	72.40 μs	1.10 μs	151.51 μs
map build	Ef	1000000	99.95 μs	182.43 μs	578.00 ns	373.21 μs
map run	Aff	1000000	1.40 s	826.01 ms	869.20 ms	2.61 s
map run	Ef	1000000	681.45 ms	67.42 ms	614.27 ms	768.07 ms
apply build	Aff	10000	55.90 μs	98.57 μs	724.00 ns	203.21 μs
apply build	Ef	10000	27.77 μs	45.44 μs	609.00 ns	95.25 μs
apply run	Aff	10000	45.76 ms	13.47 ms	33.63 ms	63.49 ms
apply run	Ef	10000	8.08 ms	1.30 ms	7.26 ms	9.99 ms
apply build	Aff	20000	25.53 μs	44.07 μs	541.00 ns	91.34 μs
apply build	Ef	20000	26.08 μs	43.35 μs	407.00 ns	90.53 μs
apply run	Aff	20000	107.59 ms	23.25 ms	86.21 ms	140.48 ms
apply run	Ef	20000	34.28 ms	4.33 ms	29.25 ms	38.88 ms
apply build	Aff	50000	76.23 μs	141.23 μs	531.00 ns	287.81 μs
apply build	Ef	50000	33.94 μs	59.22 μs	445.00 ns	122.40 μs
apply run	Aff	50000	252.15 ms	58.81 ms	207.85 ms	336.16 ms
apply run	Ef	50000	98.93 ms	17.99 ms	83.20 ms	118.65 ms
apply build	Aff	100000	31.95 μs	52.73 μs	855.00 ns	110.38 μs
apply build	Ef	100000	25.37 μs	41.29 μs	400.00 ns	86.60 μs
apply run	Aff	100000	494.45 ms	58.20 ms	424.30 ms	544.51 ms
apply run	Ef	100000	220.36 ms	51.25 ms	177.40 ms	284.08 ms
apply build	Aff	1000000	69.65 μs	122.89 μs	839.00 ns	253.57 μs
apply build	Ef	1000000	33.79 μs	52.03 μs	822.00 ns	110.38 μs
apply run	Aff	1000000	5.84 s	427.10 ms	5.28 s	6.23 s
apply run	Ef	1000000	2.10 s	16.31 ms	2.08 s	2.12 s

safareli · 2017-12-13T23:31:50Z

@natefaubion yah did that in 2. aproach. which is better for sure. Ef is still faster then Eff and Aff in run. in terms of build time it's a bit slow but build time is nothing compared to running it.
And it's safe as before.

@paf31 what you think on this?

This version will make FFI still trivial, and making this changes on Eff will be non braking.

I guess the optimizations done on Eff now will optimize this implementation as well.

natefaubion · 2017-12-13T23:33:16Z

@safareli I think you should try to work it into Eff proper in a non-breaking way (which should be possible to do by only touching the FFI), and then try testing it against optimizations.

safareli · 2017-12-14T00:20:10Z

here you go 🎉 purescript-deprecated/purescript-eff#31

initial version

3a86afe

safareli force-pushed the init branch from 7b4b550 to 3a86afe Compare December 13, 2017 00:04

safareli mentioned this pull request Dec 13, 2017

free-ish representation of Eff in JS purescript-deprecated/purescript-eff#29

Open

safareli added 2 commits December 13, 2017 01:30

use push/pop instead of unshift/shift

fc36978

use property mutation instead of array push as base effect

637e922

safareli force-pushed the init branch from 955c06c to 637e922 Compare December 13, 2017 00:31

safareli added 2 commits December 13, 2017 01:32

update benchmar log formating

a1c4cec

add more bench

119508d

safareli force-pushed the init branch from 27dd423 to 119508d Compare December 13, 2017 00:59

use effect function directly without wraping in a node

f96efe8

natefaubion reviewed Dec 13, 2017

View reviewed changes

in toEff, optimise case when Ef is effect node

d089f04

add tests to travis

487ff16

add Aff to benchamrks

fdf9d6e

Make benchmar more fair

ad5675e

Mapping was happening on pure node, such operation was optimised explicitly in Aff, so switched to do map on effect so it's more fair

Use same representation as Eff

c5a83e3

safareli force-pushed the init branch from b6969db to c5a83e3 Compare December 13, 2017 23:41

fix build

d0d3a5f

safareli mentioned this pull request Dec 14, 2017

Makes Eff Harder, Better, Faster, Stronger purescript-deprecated/purescript-eff#31

Open

5 tasks

safareli closed this Dec 14, 2017

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

1.0.0 #1

1.0.0 #1

safareli commented Dec 12, 2017 •

edited

Loading

safareli commented Dec 13, 2017

safareli commented Dec 13, 2017

natefaubion commented Dec 13, 2017

safareli commented Dec 13, 2017

safareli commented Dec 13, 2017 •

edited

Loading

natefaubion Dec 13, 2017

safareli Dec 13, 2017

natefaubion commented Dec 13, 2017

safareli commented Dec 13, 2017 •

edited

Loading

paf31 commented Dec 13, 2017

natefaubion commented Dec 13, 2017

natefaubion commented Dec 13, 2017

safareli commented Dec 13, 2017

safareli commented Dec 13, 2017 •

edited

Loading

safareli commented Dec 13, 2017

natefaubion commented Dec 13, 2017

natefaubion commented Dec 13, 2017

paf31 commented Dec 13, 2017

natefaubion commented Dec 13, 2017

safareli commented Dec 13, 2017

natefaubion commented Dec 13, 2017 •

edited

Loading

safareli commented Dec 13, 2017

safareli commented Dec 13, 2017

natefaubion commented Dec 13, 2017

safareli commented Dec 14, 2017

1.0.0 #1

1.0.0 #1

Conversation

safareli commented Dec 12, 2017 • edited Loading

safareli commented Dec 13, 2017

safareli commented Dec 13, 2017

natefaubion commented Dec 13, 2017

safareli commented Dec 13, 2017

safareli commented Dec 13, 2017 • edited Loading

natefaubion Dec 13, 2017

Choose a reason for hiding this comment

safareli Dec 13, 2017

Choose a reason for hiding this comment

natefaubion commented Dec 13, 2017

safareli commented Dec 13, 2017 • edited Loading

paf31 commented Dec 13, 2017

natefaubion commented Dec 13, 2017

natefaubion commented Dec 13, 2017

safareli commented Dec 13, 2017

safareli commented Dec 13, 2017 • edited Loading

safareli commented Dec 13, 2017

natefaubion commented Dec 13, 2017

natefaubion commented Dec 13, 2017

paf31 commented Dec 13, 2017

natefaubion commented Dec 13, 2017

safareli commented Dec 13, 2017

1. store meta prop with EfMeta object on function

natefaubion commented Dec 13, 2017 • edited Loading

safareli commented Dec 13, 2017

2. store tag, _0 and _1 props on function directly

safareli commented Dec 13, 2017

natefaubion commented Dec 13, 2017

safareli commented Dec 14, 2017

safareli commented Dec 12, 2017 •

edited

Loading

safareli commented Dec 13, 2017 •

edited

Loading

safareli commented Dec 13, 2017 •

edited

Loading

safareli commented Dec 13, 2017 •

edited

Loading

1. store `meta` prop with `EfMeta` object on function

natefaubion commented Dec 13, 2017 •

edited

Loading

2. store `tag`, `_0` and `_1` props on function directly