Improve performance of n-quads parsing

[michaeladler]

This removes the use of regular expressions for parsing n-quads. As expected, this approach is significantly faster (over 30%) and results in cleaner code, thanks to the great work by the Cayley authors.

Motivation

Improve the performance of n-quads parsing.

Checks

• Passes make test

[kazarena]

@michaeladler, thank you for the PR. Will review asap.

[kazarena]

Hi @michaeladler, the proposed change makes sense but I'm a bit concerned it creates a circular dependency between github.com/cayleygraph/quad and github.com/piprate/json-gold through JSON-LD encoder/decoder. One alternative would be to bring in the necessary from into json-gold codebase (with attribution), given its static nature, but it's not ideal either.

What are your thoughts?

[michaeladler]

Hi @kazarena,

I noticed that dependency as well but I don't think it's a problem. In Golang, cyclic dependencies happen at the package-level, not at the repository-level.

json-gold:

$ go list -json -deps ./ld  | jq '.ImportPath' | grep cayley
"github.com/cayleygraph/quad/voc"
"github.com/cayleygraph/quad/voc/schema"
"github.com/cayleygraph/quad/voc/xsd"
"github.com/cayleygraph/quad"
"github.com/cayleygraph/quad/nquads"

cayleygraph:

$ go list -json -deps ./jsonld  | jq '.ImportPath' | grep json-gold
"github.com/piprate/json-gold/ld/internal/jsoncanonicalizer"
"github.com/piprate/json-gold/ld"

Also: if it actually was a problem, we would get a compile error. I just ran the tests of cayleygraph/quad with the go.mod pointing to my fork of json-gold and didn't get one :)