807697 - Enable integer to float type coercion in the Web IDL parser

Status: RESOLVED FIXED

(Core :: DOM: Core & HTML, defect)

Description

[(no longer active)]

The aim is to make the following work:

interface Foo {
  void bar(optional float baz = 1);
};

Comment 1

[(no longer active)]

Attachment: Patch (v1)

Comment 2

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 677450 [details] [diff] [review]
Patch (v1)

s/intreger/integer/ and r=me

Comment 3

[Boris Zbarsky [:bzbarsky]]

Hmm.  Actually, we should check that the int is exactly representable as a float and raise an error if it's not.

Comment 4

[(no longer active)]

Attachment: Patch (v2)

Comment 5

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 677470 [details] [diff] [review]
Patch (v2)

This should still check that the value is not too large when type is a float instead of a double, right?  And it still has the comment typo.

r=me with those addressed.

Comment 6

[(no longer active)]

(In reply to Boris Zbarsky (:bz) from comment #5)
> This should still check that the value is not too large when type is a float
> instead of a double, right?

OK, I don't think I know how to do that in python.  The best I can think of is to make sure that -2**24 <= intValue <= 2*24, which guarantees an exact representation for 32-bit floats.  Is that good enough?  If not, I need guidance on how to do the python magic to detect this case.

Comment 7

[Boris Zbarsky [:bzbarsky]]

> Is that good enough? 

I think so, yes.

Comment 8

[Nathan Froyd [:froydnj]]

(In reply to Ehsan Akhgari [:ehsan] from comment #6)
> (In reply to Boris Zbarsky (:bz) from comment #5)
> > This should still check that the value is not too large when type is a float
> > instead of a double, right?
> 
> OK, I don't think I know how to do that in python.  The best I can think of
> is to make sure that -2**24 <= intValue <= 2*24, which guarantees an exact
> representation for 32-bit floats.  Is that good enough?  If not, I need
> guidance on how to do the python magic to detect this case.

bz has approved the obvious cases, but if you wanted to be pedantic, this is a dirty C way of doing it:

bool
exactly_convertable_p(int32_t x)
{
  if ((-1 << 24) <= x && x <= (1 << 24))
    return true;

  int first_low_bit = __builtin_ctz(x);
  if (first_low_bit == 0) {
    return false;
  }

  return exactly_convertable_p(x >> first_low_bit);
}

Comment 9

[(no longer active)]

(In reply to Nathan Froyd (:froydnj) from comment #8)
> (In reply to Ehsan Akhgari [:ehsan] from comment #6)
> > (In reply to Boris Zbarsky (:bz) from comment #5)
> > > This should still check that the value is not too large when type is a float
> > > instead of a double, right?
> > 
> > OK, I don't think I know how to do that in python.  The best I can think of
> > is to make sure that -2**24 <= intValue <= 2*24, which guarantees an exact
> > representation for 32-bit floats.  Is that good enough?  If not, I need
> > guidance on how to do the python magic to detect this case.
> 
> bz has approved the obvious cases, but if you wanted to be pedantic, this is
> a dirty C way of doing it:
> 
> bool
> exactly_convertable_p(int32_t x)
> {
>   if ((-1 << 24) <= x && x <= (1 << 24))
>     return true;
> 
>   int first_low_bit = __builtin_ctz(x);
>   if (first_low_bit == 0) {
>     return false;
>   }
> 
>   return exactly_convertable_p(x >> first_low_bit);
> }

If you rewrite this in python, and file another bug I may look into doing that, but for now all I need is to coerce 1 into 1.0.  :-)

Comment 10

[(no longer active)]

http://hg.mozilla.org/integration/mozilla-inbound/rev/a2351bf10a09

Comment 11

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/a2351bf10a09