fs: improve error performance for fsyncSync

[pluris]

 confidence improvement accuracy (*) (**) (***)
fs/bench-fsyncSync.js n=10000 type='existing' *** 3.37 % ±1.49% ±1.99% ±2.60%
fs/bench-fsyncSync.js n=10000 type='non-existing' *** 116.95 % ±6.64% ±8.87% ±11.64%
Be aware that when doing many comparisons the risk of a false-positive result increases.
In this case, there are 2 comparisons, you can thus expect the following amount of false-positive results:
0.10 false positives, when considering a 5% risk acceptance (*, **, ***),
0.02 false positives, when considering a 1% risk acceptance (**, ***),
0.00 false positives, when considering a 0.1% risk acceptance (***)

Refs: nodejs/performance#106

[anonrig]

Nit: If you moved this function to c++, it would improve the happy path (existing file)

[pluris]

That's a good opinion!
I think you just need to add CHECK_GE(fd,0) to C++.
Do you have a better idea?

[anonrig]

We need to replicate the behavior of getValidatedFd in C++ and throw the correct error.

const getValidatedFd = hideStackFrames((fd, propName = 'fd') => {
if (ObjectIs(fd, -0)) {
return 0;
}
validateInt32(fd, propName, 0);
return fd;
});

[pluris]

I added the implementation in C++.
I'm not sure if this implementation is correct.
also I updated the benchmark.
When I check it locally, the performance is strange.

Please review again.

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/54293/

[joyeecheung]

I think the diff can be a lot simpler if you just change the sync branch of the original implementations instead of repeating the code in a new binding..(and if you only introduce new bindings, the original sync branch would be dead code..) also I think this breaks --trace-sync-io?

[anonrig]

Hey @pluris can you rebase this pull request? You'll see that sync.js file is now removed and merged into lib/fs.js

[joyeecheung]

I think this should be a ValidateInt32 if we want to start moving the validation to C++. Also this is not entirely on-par with the original implementation - the original implementation would print the value in a readable format in the case it's not a number, and this does not give any information about the invalid argument, which would make it harder for users to fix the error. We could also just don't move the error validation code now and leave it in JS if it's not ready to take on implementing proper argument validation + printing in C++.

[pluris]

@joyeecheung Thank you for your comment.
You are right. As you said, I think we need to think more about the implementation of GetValidateFd(). First, I will remove this addition to C++ and change it to the original implementation.

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/54530/

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/54533/

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/54637/

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/54732/

[nodejs-github-bot]

Landed in fbd08ec