I was looking at text/template/parse/lex.go and noticed it seemed to be
very inefficient in how it searched for {{ in its input. I rewrote lexText
to us strings.Index rather than calling l.next() for every single rune in
the string. The results were up to a 9x speed improvement, depending on
the density of {{'s in the text.
My results were:
BenchmarkOldR512-4 100000 12300 ns/op
BenchmarkNewR512-4 300000 4513 ns/op
BenchmarkOldT512-4 100000 23600 ns/op
BenchmarkNewT512-4 100000 13898 ns/op
BenchmarkOldR8k-4 10000 133261 ns/op
BenchmarkNewR8k-4 100000 14852 ns/op
BenchmarkOldT8k-4 5000 264654 ns/op
BenchmarkNewT8k-4 10000 135056 ns/op
BenchmarkOldMany-4 300 5577909 ns/op
BenchmarkNewMany-4 300 5632914 ns/op
BenchmarkOldR512-4 100000 12100 ns/op
BenchmarkNewR512-4 300000 4507 ns/op
BenchmarkOldT512-4 100000 23297 ns/op
BenchmarkNewT512-4 100000 14699 ns/op
BenchmarkOldR8k-4 10000 136474 ns/op
BenchmarkNewR8k-4 100000 16566 ns/op
BenchmarkOldT8k-4 5000 266256 ns/op
BenchmarkNewT8k-4 10000 137201 ns/op
BenchmarkOldMany-4 300 5606573 ns/op
BenchmarkNewMany-4 300 5596660 ns/op
BenchmarkOldR512-4 100000 12150 ns/op
BenchmarkNewR512-4 300000 4405 ns/op
BenchmarkOldT512-4 100000 21539 ns/op
BenchmarkNewT512-4 100000 13735 ns/op
BenchmarkOldR8k-4 10000 133586 ns/op
BenchmarkNewR8k-4 100000 14076 ns/op
BenchmarkOldT8k-4 5000 264374 ns/op
BenchmarkNewT8k-4 10000 132618 ns/op
BenchmarkOldMany-4 300 5622040 ns/op
BenchmarkNewMany-4 300 5638461 ns/op
R512 was 512 bytes of Lorem Ipsum text with no {{}}'s
R8K was 8,192 bytes of Lorem Ipsum text with no {{}}'s
The T versions are the same, but with all occurrences of ipsum changed to
{{.}}
(3 times in T512 and 13 times in T8K).
The Many version was the string "{{.}}" repeated 1000 times.
The old code:
for {
delim, trimSpace := l.atLeftDelim()
if delim {
trimLength := Pos(0)
if trimSpace {
trimLength =
rightTrimLength(l.input[l.start:l.pos])
}
l.pos -= trimLength
if l.pos > l.start {
l.emit(itemText)
}
l.pos += trimLength
l.ignore()
return lexLeftDelim
}
if l.next() == eof {
break
}
}
The new code:
l.width = 0
if x := strings.Index(l.input[l.pos:], l.leftDelim); x >= 0 {
ldn := Pos(len(l.leftDelim))
l.pos += Pos(x)
trimLength := Pos(0)
if strings.HasPrefix(l.input[l.pos+ldn:], leftTrimMarker) {
trimLength = rightTrimLength(l.input[l.start:l.pos])
}
l.pos -= trimLength
if l.pos > l.start {
l.emit(itemText)
}
l.pos += trimLength
l.ignore()
return lexLeftDelim
} else {
l.pos = Pos(len(l.input))
}
Is this something that would be accepted into the standard library?
Thanks,
-Paul
--
You received this message because you are subscribed to the Google Groups
"golang-nuts" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to golang-nuts+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
