Re: scheme-specific length limits (issue 48) (was: Re: other trac issues against 4395bis)

On Thu, Mar 31, 2011 at 1:49 AM, "Martin J. Dürst"
<duerst@it.aoyama.ac.jp>wrote:

>
>
> On 2011/03/30 18:26, Larry Masinter wrote:
>
>> Opinions about open issues:
>>
>> tracker issue 48:     Can schemes set specific length limits? Should
>> RFC4395bis say something about this?
>> Ted: They can, in my opinion.  We should add text that says that they can
>> and how to say that.
>>
>
> The only place where I can see this making sense is where the underlying
> protocol/format that the data in the scheme is referring to has some length
> limitations (the practical cases I can come up with are domain names
> (overall 255 octet limit) and DOS file names (8.3). In such cases, the
> length limit is intrinsic, it may not even have to be mentioned in the
> scheme definition. On the other hand, it would be completely foolish to
> define a scheme that imposes arbitrary limits that do not exist in the
> underlying protocol. As an example, it would be total nonsense to define
> something like the http: scheme to have a maximum length of 512 (or any
> other number you prefer).
>
> So given that there was no actual practical case where this was brought up
> (at least not that I can remember), my proposal would be to just leave this
> out.
>
>
If it had been permited, Vidya and I would have used it in:

http://tools.ietf.org/id/draft-hardie-p2psip-p2p-pointers-01.txt

to limit the reg-name portion of the authority section to the 128 bits that
were actually permitted by DHTs.  I think it is also pretty likely in
proprietary protocols that register URI schemes.

It's possible, of course, that these are corner cases and that the draft
doesn't need to actually have them as scheme limitations, but use-case
limitations in the protocol documents.

In other words, I think scheme limitations of length, like scheme
limitations of permitted characters, should be permitted.  If it is left out
of the doc, though, I will not scream.


>  Larry: OK.   "IRI schemes can set specific length limits or other
>> constraints, as long as the limits are implementable."
>>
>
> I'm not sure what is meant by "implementable". The software components that
> check scheme-specific constraints on URIs/IRIs are few and far between
> anyway. And I don't think we need to point out that it's a bad idea to
> specify a length restriction in terms of a function that is hard to
> implement.
>
>     Given issue #63, length limit must not depend on whether it is IRI vs.
>> hex-encoded into URI.
>>
>
> What exactly does this mean? Does it mean that it's e.g. 512 octets,
> whether as an IRI or as an URI? Does it mean that it's e.g. 512 characters,
> whether as an IRI or as an URI? Or does it mean that it's 512 characters
> when written as an IRI with actual Unicode characters, but it can be longer
> for URIs that result from converting IRIs to percent-encoding?
>
> I guess the later is desirable, but then we should actually say so. On the
> other hand, may be are situations where it's not as easy as that. An example
> would be the length limits on IDNs, which are given by using punycode and
> checking the length of the result. (Because punycode in a clever compression
> algorithm, the results are not exactly straightforward.)
>
> So in summary: Close the issue without any changes (preferred) or be
> precise.
>
> Regards,    Martin.
>