Re: [Cbor] draft-ietf-cbor-network-addresses: tag validity; deterministic encoding

On 09-Oct-21 11:57, Carsten Bormann wrote:
> Hi Brian,
> 
> The weird thing about the current definition of the prefix is that trailing zero bytes are not sent, even if they are part of the prefix.
> 
>    As an example, ::/128 is encoded as	
>  		
>    54([128, h’’])
> 
> (::/128 is 16 zero bytes).
> 
> So don’t forget to zero-expand the prefix you get (which is another way to do the “recommendation”).
> 
> I’m not fluent in Python after midnight, so here’s Ruby:
> 
>    fail if prefix[-1] == “\0” # may need to check for prefix != ‘’ in your language
>    l = prefix.bytesize
>    fail if l > 16
>    prefix << “\0”.b * (16 - l)
>    [now do the trailing bit check in the document[

I was tidying up my Python code so that it would work with either the cbor or cbor2 library, and was distracted by a mysterious segfault in the Python interpreter, but only on Linux. It was a nightmare to find the offending statement because the GRASP code is threaded and gdb has no way to tell you which thread crashed.

Anyway, there was of course a user error *and* an oversight in the cbor2 library (now fixed: it wasn't reporting my user error as an exception).

So, my ugly code is in the functions build5254() and detag5254() in https://github.com/becarpenter/graspy/blob/master/pfxm4.py

   Brian

> 
> Grüße, Carsten
> 
> 
>> On 9. Oct 2021, at 00:40, Brian E Carpenter <brian.e.carpenter@gmail.com> wrote:
>>
>> On 09-Oct-21 00:52, Carsten Bormann wrote:
>>> There are various admonitions scattered in draft-ietf-cbor-network-addresses about doing the right thing when encoding and decoding the tags.
>>> We do have a concept in CBOR to handle this properly:
>>> Tag validity (Section 5.3.2 of RFC8949).
>>>
>>> In response to various observations about those admonitions during IESG 
>> processing, I went ahead and consolidated the actual content to a proper section about Tag Validity.
>>>
>>> https://github.com/cbor-wg/cbor-network-address/pull/15
>>>
>>> I also wrote:
>>>
>>>> The tag validity rules, combined with the rules in {{Section 4.2.1 of
>>>> -cbor}}, lead to deterministic encoding for tags 54 and 52 and require
>>>> no further Additional Deterministic Encoding Considerations as per
>>>> {{Section 4.2.2 of -cbor}}.
>>>
>>> … which might come in handy when deterministic encoding is actually needed.
>>>
>>> Comments welcome before we merge.
>>
>> I glanced at my Python code while checking the PR and noticed this:
>>
>>>> A recommendation for prefix decoder implementations is to first create
>>>> an array of 16 (or 4) zero bytes.
>>
>> That's exactly what I don't do in my code, although I'm sure it's exactly
>> what I would do in C. I think 'recommendation' is a bit strong.
>>
>>>> check that any unused
>>>> bits in the byte string are zero:
>>
>> I came up with this:
>>
>>    if prefix[length//8] & bytemasks[length%8][0] != prefix[length//8]:
>>        #extra bits in prefix, signal an error
>>
>> I also came up with a corner use case - suppose all I want to convey is
>> a prefix length but no prefix? At the moment, I simply send an all-zeroes
>> prefix, but those are wasted bits. Probably, this isn't worth fixing.
>>
>>   Brian
>>
>>
>>>
>>> Grüße, Carsten
>>>
>>> _______________________________________________
>>> CBOR mailing list
>>> CBOR@ietf.org
>>> https://www.ietf.org/mailman/listinfo/cbor
> 