perf: don't process entire string for every variable

[bomgar]

Use replacer to run replace

[bomgar]

There is a lot going on. Especially alpha and $(1|2|3). Why did you choose simple regex instead of a template engine? It would be a lot easier to extend go templates (granted the current syntax is a lot nicer).

[mvllow]

There is a lot going on. Especially alpha and $(1|2|3). Why did you choose simple regex instead of a template engine? It would be a lot easier to extend go templates (granted the current syntax is a lot nicer).

At least in JavaScript, the regex felt the easiest when we started the build tool. The surface area was smaller and it was quicker than a .includes. When moving Go, I was/am not aware of alternatives like the one you implemented. Do you have any benchmarks for this new method?

[bomgar]

I pushed a benchmark. The new version is slightly worse with very small templates. It gets faster the larger the template becomes.

I meant this as an alternative: https://pkg.go.dev/text/template
I know it is probably to late to change this stuff. I was just curious because I think features like blending would benefit from registering functions {{ blend .Pine .Base 20 }}

[mvllow]

I don't mind changing things if it makes them better, the risk is fairly low here. I'm thinking about user ergonomics, though, and $pine/20 feels nice.

[bomgar]

I agree. The current syntax is a lot nicer. It is probably sufficient, too. As I said I was mostly curious. Go templates are powerful but the syntax is kind of ugly.

[mvllow]

Rebased and force pushed. I will try to add more tests to this PR before merging.