Imporve AppendHTMLEscape fast path

[zhangyunhao116]

The < is occurs more frequently in the HTML string.

Move <

name                 old time/op  new time/op  delta
AppendHTMLEscape-16  99.3ns ± 1%  93.8ns ± 1%  -5.50%  (p=0.000 n=10+10)

And I am wondering maybe we could disable the fast path? Looks like it is not used in most cases.

Disable fast path

name                 old time/op  new time/op  delta
AppendHTMLEscape-16  99.3ns ± 1%  82.3ns ± 3%  -17.10%  (p=0.000 n=10+10)

[erikdubbelboer]

How about using strings.IndexAny to find the first special character. If it returns -1 you can return append(dst, s...). Otherwise you can dst = append(dst, s[:i]...) and continue the for loop from there. I'm not 100% sure if strings.IndexAny doesn't allocate though.

[zhangyunhao116]

The strings.IndexByte actually use assembly directly(SSE AVX etc), but strings.IndexAny is more complicate (contains more Go codes), it will be slower in this case.

use strings.IndexAny

name                 old time/op  new time/op   delta
AppendHTMLEscape-16  99.3ns ± 1%  106.2ns ± 1%  +6.92%  (p=0.000 n=10+10)

[erikdubbelboer]

In that case I'm in favor of removing the fast path as well.

But in that case isn't it also possible to use strings.IndexByte to find the next index character to encode instead of looping though the whole string one by one?

[zhangyunhao116]

But in that case isn't it also possible to use strings.IndexByte to find the next index character to encode instead of looping though the whole string one by one?

I haven't come up with a good plan yet, the strategy seems complicated, maybe we could do it in the future.

This PR removes the fast path now.

name                 old time/op  new time/op  delta
AppendHTMLEscape-16  99.3ns ± 1%  82.3ns ± 3%  -17.10%  (p=0.000 n=10+10)

[erikdubbelboer]

Thanks. I like simplifying the code if possible so I think this is also a good solution.