Apologies for my tardy response, I've now had a chance to consider this
thread a little more and review the relevant discussion (other POWDER WG
members may want to chime in too of course!).
It goes without saying that there are important differences in what WAF
Access Control and POWDER are trying to do. It's not surprising
therefore that we're taking different approaches.
POWDER is trying to make it easy (and desirable) for content providers
to make claims and assertions about their material. That sentence alone
has the lawyers salivating. "You mean we're going to say in public that
all our content is accessible? What's our liability if some of it isn't?
Will some people try and pin other people's mistakes on us?" and so on.
Therefore, step one to making this acceptable is making sure that even a
lawyer can understand that it is easy to restrict any and all claims
made to something that their client explicitly owns. So, if you're in
the legal team at example.com, you want to be sure that claims are only
about that domain.
From a POWDER perspective therefore, the default position is that
example.org matches both www.example.org and example.org. What this
discussion has made me realise is that we need to be explicit in our
documentation that it DOES NOT match badexample.org - i.e. a different
domain altogether.
Like WAF, we have an exclude method as well so we can say:
<wdr:includeHosts>example.org example.com</wdr:includeHosts>
<wdr:excludeHosts>private.example.org</wdr:excludeHosts>
(which means everything on example.org and example.com except resources
on private.example.org). The assumption being that someone describing a
load of content would know what they wanted to leave out.
So the general approach has been, as ever, that simple things will be
simple (we own example.com so that's the scope of this description) but
that complex situations can also be handled (you can write a Reg Ex if
you need to)
I wonder whether some earlier work that Mark Nottingham did might be
adapted, see section 3.22. of URISpace [1]. There he uses both ? and *
as different wildcard characters and posits that:
?.example.com
matches
www.example.com
foo.example.com
but not
example.com
www.foo.example.com
while
*.example.com
will match
foo.example.com
www.foo.example.com
but still not
example.com
It's that last point that, for me anyway, makes this more correct, yes,
but also more difficult to use. Would WAF be able to specify that:
?.example.com meant 'example.com and all its subdomains'
*.example.com meant all subdomains of example.com but not example.com
itself??
This is obviously a change from what Mark did originally but might it
solve the problem??
As for Jonas' other point - what else could/should we share. Well,
access control is clearly an application of what we're doing, whether
that's in terms of licensing or my own area of child protection. I guess
it's a question of use cases.
Our Resource Grouping mechanism has to cope with being able to cover
everything from multiple Web sites through to individual resources; we
also support resource grouping by property, not just URI, so there's a
lot there. *IF* there is a sufficient use case within the WAF's purview
to warrant such flexibility then there would be a case for an HTTP
Header/XML PI that pointed to a POWDER-style Resource Set or maybe even
a full Description Resource??
On a more mundane level, one of us needs to write the pattern matching
syntax and the other one needs to point to it! As pointed to before,
we're defining a bunch of XML data types [under development at 2] and
can readily include one for this if desired.
Phil.
[1] http://www.w3.org/TR/urispace
[2] http://www.dicom.uninsubria.it/~andrea.perego/powder/?doc=xsd
Jonas Sicking wrote:
I think an even more urgent question is, how much of the design is
POWDER and AC going to share. I see very little value in sharing just
the URI-matching syntax, so we need to figure out if there are more
things that can be shared to know if we should bother with it at all.
/ Jonas
Arthur Barstow wrote:
Thanks for the update Phil.
Regarding the value of the {include,exclude}UriPatter properties, one
of the open questions for the Read Access spec [1] is essentially
"should *.foo.com match foo.com?". The lastest draft says "no" thus
must use one rule for *.foo.com and another rule for foo.com.
Mozilla's Jonas Sicking summarized the issue, providing a list of both
the Pros and Cons [2].
We are particularly interested in your feedback on this question (e.g.
use cases, evidence of past precedence, etc.).
Art
---
[1] <http://www.w3.org/TR/2007/WD-access-control-20070618/>
[2]
<http://lists.w3.org/Archives/Public/public-appformats/2007Jul/0002.html>
On Jul 13, 2007, at 9:47 AM, ext Phil Archer wrote:
Just a quick further response to this.
The POWDER face to face meeting discussed the issue at our face to
face meeting this week (member-only minutes at [1]) and resolved*
"That we create a new RDF predicate that will support the kind of
wildcard-based URI pattern matching used by the WAF Group in their
Enabling Read Access document."
The plan is to:
1. Rename the existing URI Pattern predicates to include/exclude
UriRegEx
2. Create a new RDF predicate of include/exclude UriPattern and use
the WAF wild-card syntax. We'll extend it a little to enable control
over paths as well, so that *example.com/foo* means that the path
must begin with /foo on any subdomain of example.com if it's to be in
the set.
3. We'll add this to the XSD document we're working on. This will
soon be converted into a working draft but for now can be seen, very
unofficially, at [2].
Does this sound right to you? By all means ping me directly rather
then continue on the public lists if you prefer.
Phil.
[1] http://www.w3.org/2007/07/09-powder-minutes.html (member only)
[2] http://www.dicom.uninsubria.it/~andrea.perego/powder/?doc=xsd
* As most group members are European and the meeting was in
Washington, the number of people present wasn't high. Therefore, all
resolutions were taken at the f2f are subject to giving the remainder
of the group time to review the resolutions.
Phil Archer wrote:
cc. POWDER public list.
Hi all,
I'm glad Art responded to this so quickly as the original mail was
caught in my spam filter from which I've just retrieved it.
Mea culpa - I was unaware of the Enabling Read Access work. I will
make sure that the POWDER WG considers it fully and that we report
back (we have a face to face meeting on Monday so we can do this
quickly).
Meanwhile, we have moved on significantly since the April blog entry
referred to. Yes, we do support Perl regular expressions but we
don't rely on them. Actually, we warn against their use for most
scenarios, precisely because of the same reasons Art cites,
preferring instead to use simple string matches against URI
components. Unless something goes horribly wrong, the member only
document at [1] will be a first public working draft on Monday and
this is all set out there.
It would be easy to include a wild card pattern - indeed, there's an
example of exactly that in the extension mechanism section (which
refers to Google's URL pattern format which also looks similar to
the one in the WAF document).
Clearly, there should be a common approach, or at least a fully
compatible one. I notice that Opera is in both groups. Chaals, Anne
- I hereby nominate you as coordinators!
Phil.
[1] http://www.w3.org/2007/powder/Group/powder-grouping/20070709.html
Arthur Barstow wrote:
[[ ++ public-appformats - the mail list the WAF WG uses for its
technical discussions ]]
Stuart - thanks for raising this issue. I have not discussed this
issue with the WAF WG nor the WAF Public Community but here is *my*
take on the syntax issue.
We discussed using regular expression syntax instead of just star.
I recognize there would be some advantages to using the richer
syntax but we decided to follow the KISS principle here,
particularly given the negative side effects of incorrect syntax
(e.g. accidentally giving access to a domain that should not have
access). [I assume here the probability of incorrect syntax is
higher with regular expressions than just star but I have no real
data to back my assumption.]
It also appears our decision is consistent with the decision made
in the P3P spec [1].
WAF WG & Community - please see the issue the TAG raises below.
Regards,
Art Barstow
---
[1] <http://www.w3.org/TR/P3P/#ref_file_wildcards>
On Jul 6, 2007, at 10:16 AM, ext Williams, Stuart (HP Labs,
Bristol) wrote:
Art, Phil,
In response to a request from the WAF-WG [1] to review "Enabling Read
Access to Web Resources" [2] the TAG is concerned to ensure that
there
is good alignment between your WGs wrt the specification of resource
sets.
We observe that [2] involves the specification of 'allow' and 'deny'
sets of resources (which in this case happen to be the origins of
scripted behaviours executed by user agents). There is some resonance
between [2] and POWDER work on grouping resource sets by address. We
believe that there is or should be some common interest in the
specification of such resource sets between your WGs.
Given that web masters are the likely authors of configuration
information for both script access controls (as in [2]) and for
content-labeling (a POWDER application) and that both involve making
assertions about sets of resources (allow/deny assertions v
assertions
about the nature of web content) we believe that there should be at
least some conceptual coherence and ideally some syntactic
coherence in
the way that both POWDER and WAF-WG approach the description of
sets of
resource that are the subject of such assertions.
For example, consider the scenario in which the author of a resource
identified by http://www.sales.example.com/strategy.html wishes to
allow
cross-domain access from any resource identified by an example.com
URI.
Per [2], this set is specified with a pair of 'access items' as:
http://*.example.com
https://*.example.com
Whereas using the 'PERL regexp' based approach being considered by
POWDER (option 5 at [3]), the same set is specified as:
^https?://[^:/?#]+\.)*example\.com/
We think having two similar-but-different mechanisms to achieve
the same
goal should be avoided if at all possible.
We would be interested to hear from you whether you think there is
any
possibility of seeking considerably more alignment between the
work of
your two groups, so that where their requirements overlap there is at
least cross-reference, and at best sharing of terminology,
operational
semantics and perhaps even syntax.
Best regards
Stuart Williams
for W3C TAG
--
[1] http://lists.w3.org/Archives/Public/www-tag/2007Jun/0114.html
[2] http://www.w3.org/TR/2007/WD-access-control-20070618/
[3]
http://www.w3.org/blog/powder/2007/04/27/meeting_summary_26_27_april_200
7
--
Hewlett-Packard Limited registered Office: Cain Road, Bracknell,
Berks
RG12 1HN
Registered No: 690597 England
