feat(minifier): constant addition expression folding

[DonIsaac]

Fold constant addition expressions. Handles string concatenation and addition, both with implicit casting.

For example,

let x = 1 + 1
let y = "hello " + "world"

now becomes

let x = 2
let y = "hello world"

Extra Goodies

• test(minifier): add test_snapshot helper to perform snapshot tests with insta
• up(hir): implement std::ops::Add for NumericValue
• up(span): impl TryFrom<Cow<'_, &str>> for Atom

[codspeed-hq]

CodSpeed Performance Report

Merging #882 will not alter performance

_{Comparing don/feat/minifier/const-add-folding (75a9b3b) with main (a9d36f1)}

Summary

✅ 18 untouched benchmarks

[DonIsaac]

This PR causes several Test262 benchmarks to fail since messages in Test262Error are getting concatinated. The minifier is just checked by minifying twice, then comparing the output. Not sure what to do about this one.

Example:

expected:
if(isNaN(Number.NaN+1)!==!0)throw new Test262Error('#1: NaN + 1 === Not-a-Number. Actual: '+NaN+1);if(isNaN(1+Number.NaN)!==!0)throw new Test262Error('#2: 1 + NaN === Not-a-Number. Actual: '+1+NaN);if(isNaN(Number.NaN+Number.POSITIVE_INFINITY)!==!0)throw new Test262Error('#3: NaN + Infinity === Not-a-Number. Actual: '+NaN+Infinity);if(isNaN(Number.POSITIVE_INFINITY+Number.NaN)!==!0)throw new Test262Error('#4: Infinity + NaN === Not-a-Number. Actual: '+Infinity+NaN);if(isNaN(Number.NaN+Number.NEGATIVE_INFINITY)!==!0)throw new Test262Error('#5: NaN + Infinity === Not-a-Number. Actual: '+NaN+Infinity);if(isNaN(Number.NEGATIVE_INFINITY+Number.NaN)!==!0)throw new Test262Error('#6: Infinity + NaN === Not-a-Number. Actual: '+Infinity+NaN)

actual:
if(isNaN(Number.NaN+1)!==!0)throw new Test262Error('#1: NaN + 1 === Not-a-Number. Actual: NaN1');if(isNaN(1+Number.NaN)!==!0)throw new Test262Error('#2: 1 + NaN === Not-a-Number. Actual: 1NaN');if(isNaN(Number.NaN+Number.POSITIVE_INFINITY)!==!0)throw new Test262Error('#3: NaN + Infinity === Not-a-Number. Actual: NaNInfinity');if(isNaN(Number.POSITIVE_INFINITY+Number.NaN)!==!0)throw new Test262Error('#4: Infinity + NaN === Not-a-Number. Actual: InfinityNaN');if(isNaN(Number.NaN+Number.NEGATIVE_INFINITY)!==!0)throw new Test262Error('#5: NaN + Infinity === Not-a-Number. Actual: NaNInfinity');if(isNaN(Number.NEGATIVE_INFINITY+Number.NaN)!==!0)throw new Test262Error('#6: Infinity + NaN === Not-a-Number. Actual: InfinityNaN')

When both outputs are run through prettier, I get

// expected
if (isNaN(Number.NaN + 1) !== !0)
    throw new Test262Error('#1: NaN + 1 === Not-a-Number. Actual: ' + NaN + 1)
if (isNaN(1 + Number.NaN) !== !0)
    throw new Test262Error('#2: 1 + NaN === Not-a-Number. Actual: ' + 1 + NaN)
if (isNaN(Number.NaN + Number.POSITIVE_INFINITY) !== !0)
    throw new Test262Error(
        '#3: NaN + Infinity === Not-a-Number. Actual: ' + NaN + Infinity
    )
if (isNaN(Number.POSITIVE_INFINITY + Number.NaN) !== !0)
    throw new Test262Error(
        '#4: Infinity + NaN === Not-a-Number. Actual: ' + Infinity + NaN
    )
if (isNaN(Number.NaN + Number.NEGATIVE_INFINITY) !== !0)
    throw new Test262Error(
        '#5: NaN + Infinity === Not-a-Number. Actual: ' + NaN + Infinity
    )
if (isNaN(Number.NEGATIVE_INFINITY + Number.NaN) !== !0)
    throw new Test262Error(
        '#6: Infinity + NaN === Not-a-Number. Actual: ' + Infinity + NaN
    )

// actual
if (isNaN(Number.NaN + 1) !== !0)
    throw new Test262Error('#1: NaN + 1 === Not-a-Number. Actual: NaN1')
if (isNaN(1 + Number.NaN) !== !0)
    throw new Test262Error('#2: 1 + NaN === Not-a-Number. Actual: 1NaN')
if (isNaN(Number.NaN + Number.POSITIVE_INFINITY) !== !0)
    throw new Test262Error(
        '#3: NaN + Infinity === Not-a-Number. Actual: NaNInfinity'
    )
if (isNaN(Number.POSITIVE_INFINITY + Number.NaN) !== !0)
    throw new Test262Error(
        '#4: Infinity + NaN === Not-a-Number. Actual: InfinityNaN'
    )
if (isNaN(Number.NaN + Number.NEGATIVE_INFINITY) !== !0)
    throw new Test262Error(
        '#5: NaN + Infinity === Not-a-Number. Actual: NaNInfinity'
    )
if (isNaN(Number.NEGATIVE_INFINITY + Number.NaN) !== !0)
    throw new Test262Error(
        '#6: Infinity + NaN === Not-a-Number. Actual: InfinityNaN'
    )

This is, however, expected behavior.

[Boshen]

Terser has the evaluate option for the compressor https://github.com/terser/terser#compress-options

To get this PR merged, we can add this option and turn if off for the conformance test for now.

We can figure out how to make the algorithm pass the test, or change the test later.

[Boshen]

The code itself is good.

I'm not sure about adding snapshot testing because this deviates from our current testing strategy, which will cause future contributors confused on how to test the minifier. But I can fix this later.

The only blocker is adding the evaluate option to get the conformance test passing.

[Boshen]

The test is still failing 🤔

[mysteryven]

Just two nitpicks.