[JavaScript runtime] Implementation of ArrayPredictionContext.prototype.equals is incorrect

[ota-meshi]

I checked that the JavaScript runtime is very slow compared to the Java runtime.
And I noticed that ArrayPredictionContext.prototype.equals rarely returns true.

Looking at the Java and JavaScript implementations, the Java implementation is comparing the array elements correctly, while the JavaScript implementation is comparing the array instances.

antlr4/runtime/JavaScript/src/antlr4/PredictionContext.js

Lines 252 to 253 in a4c66dc

	return this.returnStates === other.returnStates &&
	this.parents === other.parents;

antlr4/runtime/Java/src/org/antlr/v4/runtime/atn/ArrayPredictionContext.java

Lines 76 to 78 in be881fa

	ArrayPredictionContext a = (ArrayPredictionContext)o;
	return Arrays.equals(returnStates, a.returnStates) &&
	Arrays.equals(parents, a.parents);

I think perhaps JS needs to be modified as follows:

    return equalArrays(this.returnStates, other.returnStates) &&
        equalArrays(this.parents, other.parents);

We can also use equalArrays in Utils.js, but we also need to change equalArrays in Utils.js because equalArrays in Utils.js doesn't allow arrays of primitives.

[ota-meshi]

FYI, on my PC, the source code that took 10 seconds to parse was improved to less than 1 second after making this change.

[ericvergnaud]

Sounds promising, I’ll look into this next week

…

Le 8 sept. 2020 à 04:10, Yosuke Ota ***@***.***> a écrit : FYI, on my PC, the source code that took 10 seconds to parse was improved to less than 1 second after making this change. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#2902 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAZNQJCCZ2LVTIM2MLQM64DSEWHBNANCNFSM4Q7GVF2Q>.

[ota-meshi]

Thanks for your comment! I'm looking forward to the results you look into this.

As additional information, I think that the cache existence judgment failed, the cache was not used, and it was always recalculated, but this change made it use the cache, so it was faster.