752223 - NumberValue/DoubleValue are footguns

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Boris Zbarsky [:bzbarsky]]

They don't canonicalize NaN, so it's easy for consumers to screw up if they end up with a NaN somehow.  And if I understand right, such a screwup could lead to security bugs...

I'd love an API that has the ergonomics of NumberValue()/DoubleValue() (including the inlining, ideally!) but the safety of JS_NewNumberValue.

Comment 1

[Boris Zbarsky [:bzbarsky]]

Oh, and this is keeping new bindings on JS_NewNumberValue for now.

Comment 2

[Luke Wagner [:luke]]

Given that JS_NewNumberValue is no longer fallible, we could just change/rename it to

  jsval JS_NumberValue(double d);

Comment 3

[Boris Zbarsky [:bzbarsky]]

and inline?  ;)

Comment 4

[Luke Wagner [:luke]]

I was hoping for a JS_NEVER_INLINE, JS_PUBLIC_API, CrossCompartmentWrapper::call, but I guess that's fine too.

Comment 5

[Jeff Walden [:Waldo]]

Don't we need a runtime-wide lock to make sure the memory location isn't being changed by another thread or by incremental GC?

Comment 6

[Boris Zbarsky [:bzbarsky]]

Runtime-wide, pah.  You need a process-wide lock, under which you do network I/O!

Comment 7

[Jeff Walden [:Waldo]]

Oh, right, you need e10s to load an SVG 1.2 document to do the socket access.  How could I have forgotten that?

Comment 8

[Boris Zbarsky [:bzbarsky]]

So... DOUBLE_TO_JSVAL is exactly the API I want, actually.  ;)  Except it doesn't produce int values for ints, of course.

Comment 9

[Eric Faust [:efaust]]

Attachment: Implement JS_NumberValue()

Comment 10

[Luke Wagner [:luke]]

Comment on attachment 647798 [details] [diff] [review]
Implement JS_NumberValue()

Review of attachment 647798 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/src/jsapi.h
@@ +2241,5 @@
> +JS_ALWAYS_INLINE JS_PUBLIC_API(jsval)
> +JS_NumberValue(double d)
> +{
> +    d = JS_CANONICALIZE_NAN(d);
> +    int32_t i;

Since this is technically C code, you'll need to hoist this decl.  (Building js/jsd should have caught this; I recommend doing a try build to make sure you've caught it all.)

Comment 11

[Eric Faust [:efaust]]

On try after hoist and fixing linker bustage at https://tbpl.mozilla.org/?tree=Try&rev=9ed00d0fe3da

Comment 12

[Eric Faust [:efaust]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/70d749a0e1ff

Comment 13

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/70d749a0e1ff