The extra step isn't needed and this makes the code simpler.
And slightly faster:
benchmark old ns/op new ns/op delta
BenchmarkAbsUrl 19987 17498 -12.45%
BenchmarkXmlAbsUrl 10606 9503 -10.40%
benchmark old allocs new allocs delta
BenchmarkAbsUrl 28 24 -14.29%
BenchmarkXmlAbsUrl 14 12 -14.29%
benchmark old bytes new bytes delta
BenchmarkAbsUrl 3512 3297 -6.12%
BenchmarkXmlAbsUrl 2059 1963 -4.66%
And looks even faster:
Compared to previous attempt:
benchmark old ns/op new ns/op delta
BenchmarkAbsUrl 30902 27206 -11.96%
BenchmarkXmlAbsUrl 15389 14216 -7.62%
benchmark old allocs new allocs delta
BenchmarkAbsUrl 33 28 -15.15%
BenchmarkXmlAbsUrl 16 14 -12.50%
benchmark old bytes new bytes delta
BenchmarkAbsUrl 4167 3504 -15.91%
BenchmarkXmlAbsUrl 2057 2048 -0.44%
Compared to before I started all of this:
benchmark old ns/op new ns/op delta
BenchmarkAbsUrl 36219 27206 -24.88%
BenchmarkXmlAbsUrl 23903 14216 -40.53%
benchmark old allocs new allocs delta
BenchmarkAbsUrl 60 28 -53.33%
BenchmarkXmlAbsUrl 30 14 -53.33%
benchmark old bytes new bytes delta
BenchmarkAbsUrl 5842 3504 -40.02%
BenchmarkXmlAbsUrl 3754 2048 -45.44%
Fixes#816
This commit replaces the multuple `bytes.Containts` and `bytes.Replace` with a custom replacer that does one pass through the document and exploits the fact that there are two common prefixes we search for, `src=` and `href=`.
This is both faster and consumes less memory. There may be even better algos to use here, but we must leave some room for improvements for future versions.
This should also make it possible to solve #816.
```
benchmark old ns/op new ns/op delta
BenchmarkAbsUrl 25795 22597 -12.40%
BenchmarkXmlAbsUrl 17187 11166 -35.03%
benchmark old allocs new allocs delta
BenchmarkAbsUrl 60 33 -45.00%
BenchmarkXmlAbsUrl 30 16 -46.67%
benchmark old bytes new bytes delta
BenchmarkAbsUrl 5844 4167 -28.70%
BenchmarkXmlAbsUrl 3754 2069 -44.89%
```
Fixes#894
