Performance improvements

[cfvescovo]

I have tried to improve parsing performance by replacing the standard hash function with fnv and by replacing LocalNames with strings (in node.rs) as suggested in #45.
@teymour-aldridge are there any other changes we could make to improve parsing speed? Could someone that needs to parse multiple/huge HTMLs perform some benchmarks and report back?

[cfvescovo]

I had to amend my commit because cargo fmt was complaining the imports were not sorted...

[adamreichold]

I don't think using a unkeyed hash like FNV is a reasonable choice here as it opens programs using this library up to denial of service attacks. If better hashing speed is wanted, a keyed hash like ahash (default hasher used by hashbrown) would seem more appropriate IMHO.

[cfvescovo]

You are right. Good catch! I did not think of this as an attack surface but it could definitely be exploited. Will switch to ahash ASAP

[adamreichold]

If we include #101, it might be preferable to keep this as LocalName as class names should be highly redundant seen over one or even multiple documents. Lazy initialization would then potentially allow avoiding some of the cost of deduplicating these strings when they are never matched for a given element.

[cfvescovo]

You are right. Maybe we should consider merging #91 and #101 at the same time after #101 is ready. However, this would require a major update to the crate since #101 causes breaking changes to be made to the API.

[teymour-aldridge]

Hi, sorry I haven't been responsive as of late - don't really have time to look at this, but if both of you think that the changes are worthwhile then very happy to approve so that we can merge this.

[cfvescovo]

See #101