Bug 101783 – WebIDL: number type conversions should throw TypeError for Infinity and NaN

• Status: RESOLVED
• Resolution: LATER
• Component: DOM

[Erik Arvidsson]

When a JS value is supposed to be converted to one of the following IDL types

byte
octet
short
unsigned short
long
unsigned long
long long
unsigned long long
float
double

we should throw a TypeError if the value is NaN or Infinity.

The only ones that accept NaN are the "restricted float" and "restricted double" (which we are not using anywhere)

[Joshua Bell]

I linked this to https://bugs.webkit.org/show_bug.cgi?id=96798 which has at least part of this logic implemented, but is blocked on a V8 bug that is, in turn, blocked on a Chromium bug (which is getting worked on now).

It's been a while since I looked at the patch so I'm not sure if I tackled all the 32/64/signed/unsigned permutations or just the subset I needed at the time.

Feel free to steal some/all of the patch.

[Joshua Bell]

(In reply to comment #1)
> I linked this to https://bugs.webkit.org/show_bug.cgi?id=96798

... which is focused on [EnforceRange] specifically, not on NaNs/Infinities, but the conversion for all cases could use an overhaul.

[Erik Arvidsson]

Created attachment 173355 [details]
WIP

[Adam Barth]

How does this affect performance?

[Erik Arvidsson]

(In reply to comment #4)
> How does this affect performance?

Adding more tests and branching is going to be unavoidable. Once this gets closer to complete I will have to do some perf tests. I expect some slow down, hopefully it will be small enough.

[Ahmad Saleem]

[e-mail] - Is this needed because attached WIP seems to have v8 (JS engine of Chrome) specific bits I think? Is this needed in JSC? Thanks!

[Ryosuke Niwa]

This is Later at this point.