DNS JNDI provider

[Andrew Haley <aph at cambridge dot redhat dot com>]

Pete Chown writes:
 > I've made a fair amount of progress on this and I should have
 > something that you can have a look at soon.  I have a couple of
 > questions, though, about code which is going into gcc:
 > 
 > 1.  Sun's implementation of this provider uses a class which is under
 >     the com.sun hierarchy.  For example Sun's documentation talks
 >     about the java.naming.factory.initial property as follows:
 > 
 >     > This property is used to select the DNS service provider as the
 >     > initial context. It is not used by the provider itself. It
 >     > specifies the class name of the initial context factory for the
 >     > provider, and may be set as in the following example:
 >     >
 >     > env.put(Context.INITIAL_CONTEXT_FACTORY,
 >     >         "com.sun.jndi.dns.DnsContextFactory");
 > 
 >     I find this rather odd because I thought the idea was that com.sun
 >     packages were "private" rather than being part of the published
 >     Java API.  At present there are no com.sun packages in gcj CVS
 >     (unless I'm looking in the wrong place).  This would seem to fit
 >     with the idea that com.sun packages are private.
 > 
 >     The problem this creates is that to make my JNDI provider a
 >     drop-in replacement for Sun's I would need to put it under
 >     com.sun.  This seems like a bad idea for trademark reasons if
 >     nothing else.

Interesting.  It's very hard to know what to suggest in this case.

Perhaps all we should do is produce a special error message from the
compiler.

 > 2.  I've had a look at some of the documentation about submitting
 >     patches to gcc, and they indicate that I should provide test
 >     cases.  Of course this is a good thing.  However, what test cases
 >     would you like to see for a DNS component?  At the moment I have a
 >     test zone which I install.  It is not published on the Internet;
 >     rather it is just installed as "test." in the local named.conf.
 > 
 >     People are going to get annoyed, I imagine, if they run the gcc
 >     tests and get failures because they don't have this zone
 >     installed.  I could publish the zone on the Internet but then
 >     people might get failures because they are building gcc on a
 >     machine that isn't Internet connected.

Then they get test failures because they haven't provided all of the
test infrastructure.  I find it hard to see why that's a problem.

Is there any reason why you can't use the root name servers for
testing?  Perhaps they don't allow all of the requests that you need?

Andrew.