Paul Fisher wrote:
>
> Stuart Ballard <[EMAIL PROTECTED]> writes:
>
> > The same thing could maybe even be achieved with javap and some
> > clever grepping and sedding.
>
> I wouldn't do this -- unless you're talking about using javap on
> Classpath files. It would be better to write a test which compares
> Sun's javadoc output to our classes and methods. If I remember
> correctly, downloading the javadocs does not require one to enter into
> a license agreement with Sun (unlike downloading the JDK).
I was thinking more like distributing the database that we have to
match, and then testing against that. For example, say we write a
program (or a script using javap) called dumpsigs, which outputs every
public and protected class and member signature in the java.* hierarchy.
Someone on the classpath team runs dumpsigs on JDK and puts the results
in the source tree. Then the testsuite runs dumpsigs on classpath and
compares the two outputs.
Admittedly, it requires someone to have the JDK, but I wouldn't say that
using Reflection on the public and protected members of the java.*
classes is really violating the clean room. Even more so if you use
javap, but then you get problems if the output of javap changes between
versions of the JDK - a program written by ourselves could be designed
to work on every version from 1.1.0 through 2.0 and beyond. This would
also be a problem with using the javadoc, as I don't see any automatic
script that we could write now being able to handle even the output of
1.2FCS.
Stuart.
