Array constructor with length is not eliminated

[KimlikDAO-bot]

Consider the code:

/** @const {!Array<number>} */
const a = [1];

/** @const {!Array<number>} */
const b = Array(10);
b[0] = 2;

/** @const {!Array<number>} */
const c = [];
c.push(3);

/** @const {!Array<number>} */
const d = Array(10);
d.push(4);

When compiled with

yarn google-closure-compiler -O ADVANCED --js a.js --rewrite_polyfills=false 

we get

Array(10)[0]=2;Array(10).push(4);

Arrays created with the Array(len) constructor are not eliminated.

google-closure-compiler version: v20230103

[brad4d]

The peephole optimizations remove the cases of an array created with an array literal.

Presumably we could update whichever one does that so it also removes arrays created with the Array() constructor.

However, I don't think we're likely to prioritize fixing this.
It's just not likely to be worth the effort.

I'll mark this as "help wanted", since I think it would be a good small contribution that could be made from outside our team.

[KimlikDAO-bot]

Thank you! We'll look into fixing this

[jobedrony]

I want to work in this issue

[PTheocharis]

Is it possible I work on this issue?

[KimlikDAO-bot]

@brad4d Could you point to some relevant files in the code?

[brad4d]

I would start by adding a test case here:

closure-compiler/test/com/google/javascript/jscomp/PeepholeRemoveDeadCodeTest.java

Line 1476 in d77e3ad

public void testArrayLiteral() {

fold("new Array(10);", "");

Then investigate how to make PeepholeRemoveDeadCode pass this test.

[KimlikDAO-bot]

@brad4d Actually that test passes. The difference between [] and Array() constructor is only visible if we do something like [undefined].push(1) vs Array(1).push(1).

[brad4d]

In that case, you should start with a test case containing those two cases and use the debugger to investigate what it is that prevents the Array(1).push(1); from being removed but allows [undefined].push(1); to be removed.