Add a JSON.parse function which can receive re-usable objects from an object pool

[jdgriffith]

In certain high ingestion scenarios from a websocket, it might be advantageous to use an object pool of pre-initialized objects which can be used for JSON parsing where a new object does not need to be created. This could help in several ways as it could help tell the JSON parser the shape of the data and also no new memory needs to be allocated for the object. Once an object is finished being utilized by processing, it could be given back to the object pool. I'm not asking for Bun to create an API for an object pool (but I'm not opposed!) but I am asking for more logic on the JSON processing side - the ability to pass an object along with the JSON string.

JSON.parse(jsonString, objectToBePopulated)

or

JSON.parse(jsonString, objectPool)

I'm curious if my thought about this helping with performance would hold true in this scenario but it's an interesting thought. I wonder what others think of this proposal.

[guest271314]

You can already do that by spreading data into an object then using JSON.stringify() on the object before/at passing to JSON.parse(). Note, the second parameter to JSON.parse() is a "reviver" function and with JSON.stringify() is a "replacer" function, both of which are recursive where you can mutate the result.

[jdgriffith]

I'm not sure how to communicate this to you - it's not about streams or the surrounding interfaces for transmitting or piping the data. Even your example above probably uses JSON.parse under the hood in new Response(v).json(), if that returns an object. I use streams all the time for large amounts of data and they are wonderful for that - super efficient and fast.

My post and performance concerns are only about JSON.parse and that's it. Plenty of other projects have explored the concept of super efficient JSON parsers, including using SIMD, and I'm not the first to explore more efficiency as it relates to JSON processing.

[guest271314]

What do you suggest? How would you specify the algorithm?

I'm curious why JSON.parse is important for your application?

I've used various means to parse JSON, including RegExp.

No matter what you do you are going to have to parse everything between opening and closing ". I just don't see how you can come up with something faster than what exists, given the mandatory condition above.

[guest271314]

For some context I day traded 20 years ago so I know those seconds mean something. I just don't see how you can reliably speed up JSON parsing when the parser has to handle nested objects and arrays.

[guest271314]

Maybe farm out JSON parsing to WebAssembly or something?

If you are proposing a new feature to the language I would write out the steps of the algorithm, perhaps including spreading into an existing object and branching then breaking the flow when specific properties or values are found during parsing, which JSON.parse() implements, e.g., https://stackoverflow.com/a/39263212

var [title, innerData] = JSON.parse(JSON.stringify([data.title, data.innerData], ["title"]));

otherwise you are going to have to parse the entire string, it's mandatory.

[guest271314]

A for loop is faster starting from the end than from the beginning, so in theory and practice you can use a for loop on the JSON string and break when specific property names of objects are matched.

Using arrays is far simpler, if you have control over the input and the data does not include {"property":"value"}.