1639898 - More cleanups of NaN in JS::Value

Status: RESOLVED FIXED

(Core :: JavaScript Engine, task, P3)

Description

[Ted Campbell [:tcampbell]]

Misc cleanup around handling of NaN in JS::Value.

Comment 1

[Ted Campbell [:tcampbell]]

Attachment: Bug 1639898 - Cleanup NaN code for JS::Value. r?iain

• Add more comments
• Simplify the JSVAL_SHIFTED_TAG_MAX_DOUBLE
• Add ValueToDouble helper methods
• Assert GenericNaN is compatible
• Fixup the WrappedPtrOperations for JS::Value

Comment 2

[Ted Campbell [:tcampbell]]

Attachment: Bug 1639898 - Change polarity of CanonicalNaN. r?iain

This is arbitrary, but since the negative NaN is the dividing point between
Double and boxed values, it seems like a better canonical value.

Depends on D76339

Comment 3

[Ted Campbell [:tcampbell]]

Attachment: Bug 1639898 - Use volatile in CheckCanonicalNaN. r?iain

This ensures we compute the NaN at runtime instead of compile time since this
is this test is aimed at JIT compatibility. In practice, these should be the
same, plus/minus the sign bit.

Depends on D76340

Comment 4

[Ted Campbell [:tcampbell]]

Attachment: Bug 1639898 - Decouple the GenericNaN encoding from StructuredClone. r?iain

The StructuredClone algorithm has its own form of NaN-boxing that does not
need to strictly match JS::Value. Add a SCTAG_NAN type that matches current
value of GenericNaN and replace the CanonicalizeNaN call.

When converting to JS::Value, we still explicitly canonicalize. I put back
the JS:: namespace qualifier since this isn't a universal canonicalization.
Also remove the extraneous canonicalization.

Comment 5

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/ed3c28387f3a
Cleanup NaN code for JS::Value. r=iain
https://hg.mozilla.org/integration/autoland/rev/156314339700
Use volatile in CheckCanonicalNaN. r=iain

Comment 6

[Atila Butkovits]

https://hg.mozilla.org/mozilla-central/rev/ed3c28387f3a
https://hg.mozilla.org/mozilla-central/rev/156314339700