Revert "Add support for Bun (#20)"

[AaronDewes]

This reverts commit 5bb4f50.

The latest Bun works with this package again (oven-sh/bun#5802).

[sindresorhus]

Is there any reason to revert though? The current version works too and is less fragile.

[AaronDewes]

I thought the original change would make performance worse, so I did some benchmarks...

function callsites_current() {
        const _prepareStackTrace = Error.prepareStackTrace;
        try {
                let result = [];
                Error.prepareStackTrace = (_, callSites) => {
                        const callSitesWithoutCurrent = callSites.slice(1);
                        result = callSitesWithoutCurrent;
                        return callSitesWithoutCurrent;
                };

                new Error().stack; // eslint-disable-line unicorn/error-message, no-unused-expressions
                return result;
        } finally {
                Error.prepareStackTrace = _prepareStackTrace;
        }
}

function callsites_original() {
        const _prepareStackTrace = Error.prepareStackTrace;
        Error.prepareStackTrace = (_, stack) => stack;
        const stack = new Error().stack.slice(1); // eslint-disable-line unicorn/error-message
        Error.prepareStackTrace = _prepareStackTrace;
        return stack;
}


function callsites_with_try_finally() {
        const _prepareStackTrace = Error.prepareStackTrace;
        try {
            const stack = new Error().stack.slice(1);
                return stack;
        } finally {
                Error.prepareStackTrace = _prepareStackTrace;
        }
}

function callsites_with_try_catch() {
        const _prepareStackTrace = Error.prepareStackTrace;
        try {
            const stack = new Error().stack.slice(1);
                Error.prepareStackTrace = _prepareStackTrace;
                return stack;
        } catch {
                Error.prepareStackTrace = _prepareStackTrace;
        }
}



console.time('Original callsites (pre-bun)');
for (let i = 0; i < 100000; i++) {
  callsites_original();
}
console.timeEnd('Original callsites (pre-bun)');

console.time('Callsites with Bun');
for (let i = 0; i < 100000; i++) {
  callsites_current();
}
console.timeEnd('Callsites with Bun');

console.time('Callsites with try-finally block');
for (let i = 0; i < 100000; i++) {
  callsites_with_try_finally();
}
console.timeEnd('Callsites with try-finally block');


console.time('Callsites with try-catch block');
for (let i = 0; i < 100000; i++) {
  callsites_with_try_finally();
}
console.timeEnd('Callsites with try-catch block');

In Node:

Original callsites (pre-bun): 371.756ms
Callsites with Bun: 391.198ms
Callsites with try-finally block: 1.095s
Callsites with try-catch block: 1.146s

In Bun:

[122.09ms] Original callsites (pre-bun)
[112.69ms] Callsites with Bun
[100.88ms] Callsites with try-finally block
[99.62ms] Callsites with try-catch block

I have no idea why the "cleaner" implementation is so much worse in Node 🤔

In Deno, it's the same...:

Original callsites (pre-bun): 225ms
Callsites with Bun: 234ms
Callsites with try-finally block: 910ms
Callsites with try-catch block: 929ms

But in Chrome, this does not happen:

Original callsites (pre-bun): 160.52001953125 ms
Callsites with Bun: 172.321044921875 ms
Callsites with try-finally block: 172.614990234375 ms
Callsites with try-catch block: 181.050048828125 ms

So I wonder whether maybe V8 recently optimized performance for that and Chrome has that improvement, but Node & Deno don't 🤷‍♂️