Consider adding a fast path for encode where indentation = 0

[jakub-nlx]

V8 recently landed a significant improvement to JSON serialization of "simple" objects (which is I think what we'll be 99% using). However, the gotcha being that you need to call JSON.stringify with only a single argument. Using the defaults of null, 0 doesn't trigger the fast path.

See https://news.ycombinator.com/item?id=44796648.

So in https://github.com/elm/json/blob/1.1.3/src/Elm/Kernel/Json.js#L415C24-L415C62, we'd need to do something like

var _Json_encode = F2(function(indentLevel, value)
{
       
	return indentLevel === 0 ? JSON.stringify(_Json_unwrap(value)) : JSON.stringify(_Json_unwrap(value), null, indentLevel) + '';
});

to opt into this optimization.

[jfmengels]

Slight correction to the suggested code, as the + '' needs to be on the whole operation and not just on the latter branch:

var _Json_encode = F2(function(indentLevel, value)
{
	return (indentLevel === 0 ? JSON.stringify(_Json_unwrap(value)) : JSON.stringify(_Json_unwrap(value), null, indentLevel)) + '';
});

(To clarify: JSON.stringify returns undefined and not a string if value contains anything that is not stringifiable)

[robinheghan]

I experimented with this in Gren.

Here's the repo I used: https://github.com/gren-lang/gren-json-benchmark

Please look it over in case I've done something wrong.

The optimized version is running 3% faster on my machine when encoding a list of 5000 records.
Not a massive difference. It could be that the JSON Encoder API itself is a bigger source of overhead.

[robinheghan]

Nevermind. Just realized that the V8 version which has this optimization hasn't made it into NodeJS yet. 🤦

At least we now have a benchmark to check once it lands :)