continuing on the same idea from article 5, here i wanted to investigate the complexity of those two cases.

for the int type this benchmark is easier as i only vary the number of cases inside the switch, and the map keys are just the numbers.

the benchmark was done in like this:

given a number *n* of cases.

- generate the switch with
*n*cases - generate the map with
*n*entries - measure the time to retrieve an item

i've done that for values from 5 to 10000 using a shell script to hardcode all values on the source file. the values are random strings of size 16.

here is one example:

package switchvsmap func switchCaseInt5(input int) string { res := "" switch input { case 0: res = "9U5mScxLZLbQfj7y" case 1: res = "P0CoRrj0x7PwOZTO" case 2: res = "jAbhtkEBmEXlmQ15" case 3: res = "2GJV7FBmYemqF7VB" case 4: res = "AflqdquuhoUBqYT0" } return res } func mapCaseInt5(input int) string { return mapInt5[input] } var mapInt5 = map[int]string{ 0: "T5Nlajp4dWNsdAxq", 1: "VLXyeeQVNP8LNEfS", 2: "GfjlCRLOiv5qrj3m", 3: "yySDbcA0Mf7J90qA", 4: "GdKSg4pR8xpRb6Lx", } func switchCaseInt10(input int) string { res := "" switch input { case 0: res = "wylNCLjLylMZgjo9" case 1: res = "e6X044Vj9OyNliBC" case 2: res = "MfSoK3jCk3LujGUM" case 3: res = "FY8Z9owHtWAiA3eJ" case 4: res = "sof0DLcmDeR1nxGf" case 5: res = "Rny0iTpLbh79Tfl4" case 6: res = "XyFaFrCKpOLCIL6D" case 7: res = "OvYkV1NmdUpQA0q8" case 8: res = "cL7RWN0aSCEno37m" case 9: res = "AMgSpv4t7BN1Lj8l" } return res } func mapCaseInt10(input int) string { return mapInt10[input] } var mapInt10 = map[int]string{ 0: "KlRf6TsIkvoaOf01", 1: "IdBDDIIjLKnULDUA", 2: "MmqxCLC4ssC8AoJo", 3: "FJSFb2ozxQn0QRhu", 4: "ISx7BqtOGsgcCODS", 5: "vzHJ0gAhtj2Ejx4m", 6: "Wm24QTHb4fVA34jt", 7: "T6o64gYAr3N5Ppxr", 8: "gspkuIF7rHYUySma", 9: "SJAJSFdYKhWp7ZNH", } . . .

the complete file has 149,112 lines.

the test file was generated using the same strategy:

package switchvsmap import ( "testing" ) func BenchmarkSwitchCaseInt5(b *testing.B) { for n := 0; n < b.N; n++ { switchCaseInt5(n % 5) } } func BenchmarkMapCaseInt5(b *testing.B) { for n := 0; n < b.N; n++ { mapCaseInt5(n % 5) } } func BenchmarkSwitchCaseInt10(b *testing.B) { for n := 0; n < b.N; n++ { switchCaseInt10(n % 10) } } func BenchmarkMapCaseInt10(b *testing.B) { for n := 0; n < b.N; n++ { mapCaseInt10(n % 10) } } . . .

then i just ran the go test command and ploted it.

x axis is the size of the switch/map, and y axis is the ns/op value that go test outputs.

for big number of cases both methods perform very close, due to the big variation of speed i cannot clearly see a faster one. one thing that suprised me is that the complexity looks more logarithmic, but my expectation was to see a linear growth for the switch case.

we see that the switch statement is faster than the map up to 3000 cases, after that it is not waranteed. i don't think anyone would hardcode that much cases in a real life example.

the advantage of the map is that it is a dynamic structure, i mean, it can me modified on runtime, so that is not a surprise that the switch case is faster, since it is a fully determined at compile time, and thus can be otimized.

my conclusion is that if you know your cases at compile time, then use the switch case, otherwise use the map.