More improved performance

[Code-Hex]

This PR will be improving performance when called GraphemeClusterCount function.

benchmark code.

func BenchmarkCountBefore(b *testing.B) {
    for i := 0; i < b.N; i++ {
        for _, bcase := range testCases {
            g := beforeNewGraphemes(bcase.original)
            var n int
            for g.Next() {
                n++
            }
        }
    }
}

func beforeNewGraphemes(s string) *Graphemes {
    g := &Graphemes{}
    for index, codePoint := range s {
        g.codePoints = append(g.codePoints, codePoint)
        g.indices = append(g.indices, index)
    }
    g.indices = append(g.indices, len(s))
    g.Next()
    return g
}

func BenchmarkCountAfter(b *testing.B) {
    for i := 0; i < b.N; i++ {
        for _, bcase := range testCases {
            g := NewGraphemes(bcase.original)
            var n int
            for g.Next() {
                n++
            }
        }
    }
}

Result

go test -benchmem -run="^$" github.com/rivo/uniseg -bench .
goos: darwin
goarch: amd64
pkg: github.com/rivo/uniseg
BenchmarkCountBefore-8              5000            247797 ns/op           96392 B/op       3458 allocs/op
BenchmarkCountAfter-8              10000            199063 ns/op          105544 B/op       1865 allocs/op

[Code-Hex]

@rivo Review please~~~~ 🙏🙏 🙏 🙏 🙏 🙏 🙏

[System-Glitch]

Any news on this PR? It is a nice optimization.

[dolmen]

Could be improved more: use local variables for slices and create the Grapheme object only after the loop.

[dolmen]

This performance optimization uses more memory. To allocate the right amount (loosing some speed), use utf8.RuneCountInString instead of len.

[rivo]

I merged #8 instead. Hope that's ok with you.