Re: [AVTCORE] Review of draft-ietf-avtcore-rfc5764-mux-fixes-02

[Magnus Westerlund <[e-mail]>]

Den 2015-09-25 kl. 16:25, skrev Paul Kyzivat:
> On 9/25/15 6:33 AM, Marc Petit-Huguenin wrote:
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA256
>>
>> Hi Paul,
>>
>> On 09/18/2015 10:17 AM, Paul Kyzivat wrote:
>>> On 9/17/15 4:10 PM, Marc Petit-Huguenin wrote:
>>>
>>>>> K. Section 3:
>>>>>
>>>>> When new values or ranges are added, they MUST be tested in
>>>>> ascending order.
>>>>>
>>>>
>>>> This text should not talk about adding new values/ranges. Changed
>>>> the text to:
>>>>
>>>> "The various range values for the first byte MUST be tested in
>>>> ascending order."
>>>
>>> I just reviewed this, and I have no idea what it means. I see nothing
>>> about Figure 3 that requires testing to be done in a particular
>>> order. The *text* suggests an if/then/else or switch statement
>>> implementation. But the code generated for a switch statement may
>>> well be a lookup table and so have no particular ordering.
>>>
>>> The only time ordering would matter would be if the ranges
>>> overlapped. And IIUC the goal here that the ranges *not* overlap.
>>
>> The reason we added this was to plan when a new overlapping range is
>> added (mostly because someone ignored the advices provided by this
>> draft).  If the algorithm says today that range must be checked in
>> ascending order, then *all* implementations will work (or, most
>> likely, fail) in a similar way.  That sentence is meant to explicitly
>> prevent using a lookup table or perfect hashing or whatever technique
>> that would prevent things to fail identically everywhere.
>>
>> That said the sentence can probably be improved.
>
> Yes.
>
> Do you consider overlapping ranges to be an error, and want the error to
> be handled deterministically? Or are you thinking that there may be
> overt changes that legally define overlapping ranges and want to make
> the meaning deterministic?
>
> If there is a deterministic choice when ranges overlap, then that seems
> to effectively update one of the ranges so that they don't overlap.
>

I think this discussion is going into the comment of my issue J. It 
really is to write down in the document why it would benefit to have a 
particular order of checking, and thus the implication of that for 
implementations. A reader in the future must have the chance of 
understanding why it should do what you suggest and why it matters. That 
was missing before. And as Paul comments what to actually do need to be 
more clearly described.

Cheers

Magnus Westerlund

----------------------------------------------------------------------
Services, Media and Network features, Ericsson Research EAB/TXM
----------------------------------------------------------------------
Ericsson AB                 | Phone  [phone]
Färögatan 6                 | Mobile [phone]
SE-164 80 Stockholm, Sweden | mailto: [e-mail]
----------------------------------------------------------------------