[Opensim-dev] Uris: Errors/Inconsistencies/TBD

[Alan M Webb]

Stefan

OK. I didn't pick up on the transition from your original note which 
referred to URI's in general that end with a "/" and a URI with an empty 
path, which is a special case I think, in being considered to be 
equivalent to an absolute empty path. But is this case really the one you 
were concerned about? A relative path is not required to start with a "/" 
so inserting one would certainly not be desirable. If this is the only 
case then I'm not sure I have understood your concerns properly.

If your issue is solely with the empty-path case then I have no problem 
with a "/" being required, or being introduced if omitted. However, I do 
still think that we would be best served by an unconditional decomposition 
of all inbound URI's into the elements that are described in the RFC, with 
the "path" (sic) being decomposed into a sequence of segments. Even in 
this canonical form the 1st "/" is clearly special and the fact of its 
presence needs to be preserved.

Also, are you concerned with incoming URI's or outoing (OpenSim composed) 
URI's. I thought all inbound URI's were coming through the HttpListener 
interface which I would have (naiively) expected to have always used the 
underlying Uri mechanism.

Sorry if I'm being dense here.

Best regards
Alan
-------------------
T.J. Watson Research Center, Hawthorne, NY
1-914-784-7286
alan_webb at us.ibm.com



Stefan Andersson <stefan at tribalmedia.se> 
Sent by: opensim-dev-bounces at lists.berlios.de
09/10/2008 09:32 AM
Please respond to
opensim-dev at lists.berlios.de


To
<opensim-dev at lists.berlios.de>
cc

Subject
Re: [Opensim-dev] Uris: Errors/Inconsistencies/TBD






Alan,
 
I believe the Uri behaviour I'm talking about just concerns the host node. 
What happens if you change that to

Uri absURI = new Uri(http://localhost:9000/); 
vs
Uri simURI = new Uri(http://localhost:9000); 

? I believe you should get the same result on mono.
 
of course, if you specify segments, the uri should be reported back as it 
was specified. Otherwise, Uri would change the semantics of the last node. 
Which would be bad, indeed.
 
I concur with your suggestion to use a uniform Uri strategy thru OpenSim. 
I do believe that it should be built upon Uri and UriBuilder over 
home-grown code. Given that I haven't found segment building support in 
neither, we might want to combine Uri, UriBuilder and the uri building 
parts of RestClient into our own OSUri class that encapsulates building 
and handling uris.
 
The goal is, as I guess you state, to make the system as resilient as 
possible. And, to some extent, to introduce a bit more fine-grained a 
typeset than 'string' for passing Uris.

Best regards,
Stefan Andersson
Tribal Media AB
 
Join the 3d web revolution : http://tribalnet.se/
 




To: opensim-dev at lists.berlios.de
From: alan_webb at us.ibm.com
Date: Wed, 10 Sep 2008 08:07:47 -0400
Subject: Re: [Opensim-dev] Uris: Errors/Inconsistencies/TBD


Hi Stefan 

The .NET behavior is bemusing. Running mono on Linux, the behavior is NOT 
the same - which may be cause for even bigger concern. On Linux, given: 

Uri absURI = new Uri("http://localhost:9000/abc"); 
Uri relURI = new Uri("abc", UriKind.Relative); 
Uri simURI = new Uri("http://localhost:9000/abc); 

then Console.WriteLine(xxxURI) and Console.WriteLine(xxxURI.ToString()) 
both produce exactly what went into the constuctor. If I change the 
initial string to have a trailing slash, only then do I see a trailing 
slash on output. On the face of it I agree with your assertion that using 
the existing behavior of Uri should be preferred, and yet if our goal is 
to obtain consistent behaviour for OpenSim on multiple platforms, the 
tests above would suggest that Uri is not going to do that for us, and 
that the RestClient code, in doing things explicitly, is more likely to 
succeed. It would be nice to get the same behavior from mono and .net, but 
that's probably much harder to achieve. 

I probably take issue with your criticism of the RestClient's (who wrote 
this code - they should be defending this!) removal of a single trailing 
slash. A single trailing slash is an accepted, if not entirely proper, 
redundancy. Two trailing slashes would be an invalid URI in my opinion and 
should always be rejected. As an externally supplied value we can not 
adopt a pre-condition stance as to its validity, so testing and rejection 
should be preferred I think. In the subsequent code that you used in your 
example, given the preceding, then this IS based upon a slashless URL 
representation. 

If I were writing the RestClient I think I would prefer to see the 
decomposition of the URI into its component pieces, and reconstruction 
from that abstract form immediately prior to use. 

So my vote would be for a single class in OpenSim that provides a 
guaranteed URI behaviour for everybody who deals in this area. No URI 
would be admitted into OpenSim except via this class, and none generated 
from any other source. All modifications to the URI would also be made 
using the class. 

By the way, just so the original issue doesn't get lost, our objective 
here is for OpenSim to have an HTTP interface (REST, XMLRPC, whatever) 
that does not behave differently when a supplied URI has, or does not 
have, a trailing "/", right? 

Best regards
Alan
-------------------
T.J. Watson Research Center, Hawthorne, NY
1-914-784-7286
alan_webb at us.ibm.com_______________________________________________
Opensim-dev mailing list
Opensim-dev at lists.berlios.de
https://lists.berlios.de/mailman/listinfo/opensim-dev

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://opensimulator.org/pipermail/opensim-dev/attachments/20080911/484bd2aa/attachment-0001.html>