I was meaning to write about some of the new openinfocard features this weekend, but instead I spent my time trying to get the openinfocard selector working against the Windows Live Beta site supporting Information Cards. I finally found the problem and really just could not believe it. A few months ago, I received a similar, yet unrelated, bug report against my own libraries. A person was using my wsse/xmlsec libs to communicate against a .NET SOAP service that required messages to be signed and told me that it would not work unless the XMLDSIG elements used a default namespace. This means that ... would work, while .. would not. Anyone who works with XML is fully aware that these two are exactly the same, so I just couldn't accept that. In fact, when mixing namespace (as often is the case when working with XMLDSIG and XMLENC) its easier to read the document when prefixes are used.
Within my libraries, I do prefix namespaces. Unfortunately for that person, I did not believe them and stressed that it had to be a coding error either on their part or from the service provider. Note that I didn't just dismiss their report. I was unable to reproduce the issue, was not given access to test against their service, do successfully interoperate with other .NET systems, and have a large number of users implementing my code against .NET services employing encryption and digital signatures. I hadn't heard anything more and ended up forgetting about it... that is until now.
Over the weekend, I spent a good amount of time comparing tokens from various selectors and trying different parameters. There were only two differences between tokens from Openinfocard and those from CardSpace. The first, which I spent most of my time on, was the timestamps. CardSpace provides a full hour for token validity. Openinfocard, on the other hand, allows the token 10 minutes of validity. I have run into a number of problems in the past due to the clock from either the client or server not being in synch. A fudge factor is usually built into the interactions (the client might set their NotBefore time to a few minutes prior to the current time, and the server might allow an extra ten minutes past the expire time), but I have seen cases, especially due to day light savings and systems not being updated, that the clocks between the client and server are too far off and the token is not considered valid when submitted. Anyways, this didn't end up being the problem.
Come to find out, the Windows Live Beta site has the exact same problem when dealing with the SAML token as the bug reported I told you about before. The issue is unrelated to the client code (so thankfully it wasn't an issue in my library - written in PHP); Openinfocard, which is the selector having a problem working with Windows Live, is in fact written in Java, and also prefixes namespaces. After altering the openinfocard code to use default namespaces, building new jars, installing the new jars and restarting firefox (sounds simple, yet REALLY time consuming), I finally got Windows Live to accept my Infocard. The underlying problem itself still eludes me. All I know is that the issue lies on the server side. I have no idea if this is a problem stemming from a particular version of the .NET libraries or if a third party library is being used.
Either way, I would have expected more from Microsoft. It's forgivable that a developer from a small company might use an outside library to work with digital signatures (that also happens to be buggy), but for a company that pushes the WS-* stack (XML Digital Signatures being a core component of WS-Security) and provides core libraries for working with it, this is a serious issue. It also seems to not be isolated either; as exemplified by the same issue against a .NET SOAP service.
Now hopefully someone can get me some answers to where this problem stems from so in the future I have a little more insight if I personally end up encounter it when dealing with a SOAP service; or at least can provide some help to someone reporting the issue to me again.