Thanks to Andrew, Craig, and David for your responses.
I agree with Andrew that writing custom doclets sounds like a lot of
tricky work.
I like Craig's suggestion that we ship two sets of public javadoc, a
JDBC3 javadoc for users who run on jdk1.3-1.5 and a JDBC4 javadoc for
users who run on jdk1.6. We could prune irrelevant classes from each
set. Besides solving the immediate problem, this approach would be less
confuing. Right now the composite javadoc, which contains both the JDBC3
and JDBC4 apis mixed together, is puzzling. Craig's solution would look
like this:
i) We would generate the JDBC3 api using the 1.4 javadoc tool. The
missing subclass references shouldn't confuse customers because the
subclasses wouldn't appear in the JDBC3 api.
ii) We would generate the JDBC4 api using the 1.6 javadoc tool. This api
would not contain the lying classes from the JDBC3 api.
iii) We would bolt a webpage on top of the top apis, explaining which
api to consult, based on the client VM.
Does this sound acceptable?
Thanks,
-Rick
Craig L Russell wrote:
Hi Rick,
I'm intentionally cross-posting to derby-user just because lies in
javadoc are supposed to affect users, not only developers.
How about:
3. Build two sets of javadoc, one using jdk 1.4 and another using 1.6.
Distribute both sets of javadoc. Require the user to choose which
javadoc to use based on which jvm she chooses to run. Tell no lies.
I'm sure I'm missing something; hopefully it's not obvious. :-)
Craig
Thanks to everyone who responded to this thread. It doesn't seem that
anyone has a solution to this problem. Does anyone have a preference
for which lie we tell: (1b) or (2d)? Barring a preference here, the
default would be (1b), our current behavior.
Thanks,
-Rick
Rick Hillegas wrote:
Right now the javadoc generated for jdk1.6 is telling a shocking
lie. I can fix this but only by inducing javadoc to tell a different
lie. I would like advice on how to get javadoc to tell the truth,
the whole truth, and nothing but the truth. If that's not possible,
I'd like to know which lie the community prefers.
1) How it is today.
Right now, if you point your ant.properties at a 1.6 installation,
we build javadoc with the 1.6 javadoc tool. The tool assumes that
you built your whole classtree against 1.6 and that the compiler
therefore caught certain kinds of errors. In particular, if a class
successfully compiles under jdk1.4 against the jdk1.4 version of an
interface, then the 1.6 javadoc tool assumes that the class
implements additional methods added to that interface in jdk1.6.
Here's an example of the problems this causes:
a) EmbeddedDataSource, compiled under jdk1.4, implements the 1.4
version of javax.sql.DataSource
b) The 1.6 javadoc falsely says that EmbeddedDataSource implements
the Wrapper methods added to javax.sql.DataSource by jdk1.6
2) A possible fix and its countervailing lie
It would be possible to use the 1.4 javadoc tool to build javadoc
for all the classes compiled under 1.4. Then we could use the 1.6
compiler to build the whole classtree again. With a little
jiggery-pokery, we might be able to copy the additional javadoc html
into the 1.4 javadoc tree and use the 1.6 index.html to zipper the
two trees together. Mind you, I haven't built this yet, I'm just
waving my hands. For the example case above, we would end up with
something like the following:
c) EmbeddedDataSource would NOT assert that it implements the jdk1.6
Wrapper methods
d) However, now EmbeddedDataSource would fail to say that it has an
important subclass, EmbeddedDataSource40
3) Other solutions?
Does anyone have a better solution? Better means easier and/or more
truthful.
4) Lies and the lying liars who like them
If not, which lie do you prefer: (1b) or (2d).
Thanks,
-Rick
Craig Russell
Architect, Sun Java Enterprise System http://java.sun.com/products/jdo
408 276-5638 mailto:[EMAIL PROTECTED]
P.S. A good JDO? O, Gasp!
