Re: Bug 23159 - Inconsistent "length" property when generating keys (bits vs bytes)

Fair enough on the WebIDL RTT and ambiguity around "append".

The critical thing I was trying to get at is that IIUC the UA is supposed
to convert |r| and |s| to *byte* strings, in the same form as for
BigInteger, then return the concatenation of those two byte strings.


On Wed, Mar 5, 2014 at 2:38 AM, Ryan Sleevi <sleevi@google.com> wrote:

> Then it's as equally unspecified - what does it mean to "append an
> ArrayBuffer[View] to result".
>
> I find the original definition clearer because it doesn't require a
> roundtrip through a WebIDL type (that someone may have messed with)
>
> On the question of HMAC bits vs bytes, this is an example of a change that
> would be disastrous to make if any implementations had shipped (eg: sans
> prefix). Microsoft's choice of prefixing, along with the vastly different
> nature of the API, hopefully means that few people would run into the edge
> of using the same HmacParams dictionary for (previous version, current
> version) of the spec.
>
> I'm ambivalent on the change itself, and since no one has (to my
> knowledge) shipped, it's "probably" safe to make.
>
>
>
> On Tue, Mar 4, 2014 at 1:47 AM, Richard Barnes <rlb@ipv.sx> wrote:
>
>> After taking a look at how this shows up in the spec, I think I'm OK with
>> bits.  It seems to me like in all cases where it matters, the bit length is
>> constrained in the way you note.
>>
>> One related thing I noticed in the ECDSA definition:
>> "Convert r to a bitstring and append the sequence of bytes to result"
>>
>> Might be helpful to state this in a way that makes clear (by reference)
>> what happens for
>> "Convert r to a BigInteger and append it to result"
>>
>> And likewise for "s".
>>
>> --Richard
>>
>>
>> On Tue, Mar 4, 2014 at 12:13 AM, Mark Watson <watsonm@netflix.com> wrote:
>>
>>> Yes, well, since HMAC is the odd one out, it seemed changing that might
>>> be simpler from the point of view of all the existing implementations, test
>>> suites etc. Also, it's more common to refer to the number of bits in a key
>>> (e.g AES-128) than bytes. In those cases you have to check for specific
>>> values 128, 192, 256, not just for byte alignment.
>>>
>>> So, I still prefer bits.
>>>
>>> ...Mark
>>>
>>>
>>>
>>> Sent from my iPhone
>>>
>>> On Mar 3, 2014, at 3:59 PM, Richard Barnes <rlb@ipv.sx> wrote:
>>>
>>> Agree we should be uniform.  Typed arrays are all byte oriented, so it
>>> seems like aligning on BYTES (literally) would result in less ambiguity.
>>> Otherwise you have to say how you pack left over bits.
>>>
>>>
>>>
>>> On Monday, March 3, 2014, Mark Watson <watsonm@netflix.com> wrote:
>>>
>>>> https://www.w3.org/Bugs/Public/show_bug.cgi?id=23159
>>>>
>>>> The length property of an algorithm is everywhere specified as the
>>>> length in BITS, except HMAC which defines it as the length in BYTES.
>>>>
>>>> The proposal is to align on BITS.
>>>>
>>>> Any objections ?
>>>>
>>>> ...Mark
>>>>
>>>
>>
>

Received on Wednesday, 5 March 2014 07:35:19 UTC