extend fuzzy RHS folding

[alexlamsl]

• a = []; if (1) x(); ➡️ if (a = []) x();

[kzc]

a = []; if (1) x(); ➡️ if (a = []) x();

Will this de-opt some code if the if could otherwise be simplified?

Or can this form:

if (a = []) x();

get to this form:

a=[];x();

with or without passes?

[alexlamsl]

I have yet to observe lost opportunities like that with the samples I'm testing, but theoretically a = [] is not being treated as truthy at the moment.

Shouldn't be too hard to make it happen in a separate PR.

[kzc]

Optimal code for this transform is dependent on the order of optimizations. Does default minify show an improvement generally?

[alexlamsl]

For test/benchmark.js, the major improvement is observed with lodash.js:

if (isArray(objValue)) {
    newValue = objValue;
}
else if (isArrayLikeObject(objValue)) {
    newValue = copyArray(objValue);
}
else if (isBuff) {
    isCommon = false;
    newValue = cloneBuffer(srcValue, true);
}
else if (isTyped) {
    isCommon = false;
    newValue = cloneTypedArray(srcValue, true);
}
else {
    newValue = [];
}

whereby the folding of isCommon = false; into true allows for all the newValue = to be extracted as a single assignment instead of multiple ones buried in a chain of conditionals.

[kzc]

That would be a good test for this PR.

[alexlamsl]

That would be a good test for this PR.

Done in https://github.com/mishoo/UglifyJS2/pull/3006/files#diff-37dcf1d17b8a2b3c980e12edef17fdc8R5036