Speed up minification

[tdewolff]

Speed up minification

[tdewolff]

17f97c5 has been pushed with 10% speed improvement.

These changes need to occur to make it faster, and I'm not sure if it's worth it:

• the HTML tokenizer returns strings, would save a realloc to return []byte
• the CSS parser builds a node tree which is heavy on allocs and the GC
• ignore io.Writer errors from Write (not sure if this is safe)
• the HTML minifier functions takes most time, but there isn't a way to improve it much...

[tdewolff]

e346a3c pushed with another 10% speed improvement.

It now takes []byte instead of string from the HTML tokenizer and uses atom.Atom enum for tags and attributes for fast comparison (instead of []byte equality tests).

Ignoring io.Writer errors will not increase performance significantly.

The only way to speed up is to reduce memory usage, but that is hardly possible!

[tdewolff]

Great improvement of 30% with e219023 and somewhat with 31ef4d6 by refactoring the code.

Speed could be improved if HTML tokenizer didn't unescape text tokens and attribute values, so the minifier wouldn't have to re-escape them.

[tdewolff]

HTML minification speed significantly increased by writing a new tokenizer 3b90fc7. The tokenizer is equal in speed, but more assumptions can be made to avoid memory copying and HTML entities re-escaping. Most of the speed improvement is actually from re-using the token buffer better and other things.

Avoiding allocations in 9831f46 also helped.

Additonally, the bytes.Equal is slightly slower for some reason than self-implementing it. And by using literal byte slices instead of literal strings to byte slice casts improved speed too: b36014e.

Furthermore, attribute escaping was a bottleneck: df03a65

Total speed increased by almost 50%!