lib: improve debuglog() performance

[mscdex]

Example benchmark results:

 confidence improvement accuracy (*) (**) (***) streams/pipe-object-mode.js n=5000000 *** 27.10 % ±1.46% ±1.95% ±2.55% streams/pipe.js n=5000000 *** 23.04 % ±1.54% ±2.08% ±2.76% streams/readable-boundaryread.js type='buffer' n=2000 ** 3.22 % ±2.19% ±2.92% ±3.83% streams/readable-readall.js n=5000 *** 9.04 % ±2.51% ±3.35% ±4.39% streams/readable-unevenread.js n=1000 *** 6.38 % ±0.63% ±0.84% ±1.10% 

Improvements come from optimizing the existing function returned by debuglog() but also by adding an optional callback that can be used to get a reference to the actual underlying debug output function to avoid calling into the unnecessary wrapper function.

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• tests and/or benchmarks are included
• commit message follows commit guidelines

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/29830/

[himself65]

What about this?

Suggested change

	functiondebuglog(set,cb){
	letdebug=(...args)=>{
	// Only invokes debuglogImpl() when the debug function is
	// called for the first time.
	debug=debuglogImpl(set);
	if(typeofcb==='function')
	cb(debug);
	debug(...args);
	};
	return(...args)=>{
	debug(...args);
	};
	functiondebuglog(set){
	letdebug=(...args)=>{
	// Only invokes debuglogImpl() when the debug function is
	// called for the first time.
	debug=debuglogImpl(set);
	debug(...args);
	};
	conststate={called: false};
	return((...args)=>{
	if(!this.called){
	debug(...args);
	this.called=true;
	}else{
	debug(...args);
	}
	}).bind(state);

just like

node/lib/events.js

Lines 426 to 432 in ec204d8

	function_onceWrap(target,type,listener){
	conststate={fired: false,wrapFn: undefined, target, type, listener };
	constwrapped=onceWrapper.bind(state);
	wrapped.listener=listener;
	state.wrapFn=wrapped;
	returnwrapped;
	}

[himself65]

because I don't think let debug = fn('xxx', v => debug = v) is a good code style

[mscdex]

The suggested change is basically what the code was doing before. Having a callback is the only way to be able to avoid the wrapper function and utilizing it provides a further, noticeable performance improvement.

[himself65]

Thanks, I understand

[jasnell]

Fairly certain this is because of the way V8 optimizes but that's only a guess. It's interesting that it's such a big perf difference.

[mscdex]

Results again with current master:

 confidence improvement accuracy (*) (**) (***) streams/pipe-object-mode.js n=5000000 *** 40.88 % ±2.68% ±3.57% ±4.65% streams/pipe.js n=5000000 *** 31.51 % ±1.83% ±2.44% ±3.18% streams/readable-boundaryread.js type='buffer' n=2000 *** 4.53 % ±0.41% ±0.55% ±0.72% streams/readable-readall.js n=5000 *** 10.17 % ±2.23% ±2.97% ±3.88% streams/readable-unevenread.js n=1000 *** 4.25 % ±0.82% ±1.09% ±1.43% 

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/31662/