[Jikesrvm-core] Compatibility with Classpath release vs cvs head

[Michael H. <hi...@us...>]

My 2 cents

At a minimum our goal should be
 1. our next release (and current head) works with a Classpath release
0.05.
 2. After the release, our CVS head works with Classpath's head

A more ambitious goal is
 3. our current head always works with classpath head and 0.05 (both before
and after the release)

One downside of not achieving goal #3, is that folks living off the Jikes
RVM head after the release will also need to live off the Classpath head.
I'm not sure how onerous this is.

Opinions?

Mike

- - - - - - - - - - - - - - - -
Michael Hind, Manager, Dynamic Optimization Group, Jalapeño Project

IBM Watson Research Center
http://www.research.ibm.com/people/h/hind
hi...@wa..., 914 784-7589, tie: 863-7589
Jikes RVM open source release:
http://www.ibm.com/developerworks/oss/jikesrvm


|---------+------------------------------------------------->
|         |           David P Grove/Watson/IBM@IBMUS        |
|         |           Sent by:                              |
|         |           jik...@ww...uthbury.|
|         |           usf.ibm.com                           |
|         |                                                 |
|         |                                                 |
|         |           03/05/2003 10:00 AM                   |
|         |           Please respond to jikesrvm-core       |
|---------+------------------------------------------------->
  >----------------------------------------------------------------------------------------------|
  |                                                                                              |
  |       To:       jik...@ww...                                  |
  |       cc:                                                                                    |
  |       Subject:  [Jikesrvm-core] Re: [Jikesrvm-regression] mauve failures:        0 ;         |
  |        classpath CVS head regression FAILED       32  tests                                  |
  >----------------------------------------------------------------------------------------------|








      Changes in classpath cvs to java.io.FileDescriptor broke the build of
Jikes RVM vs. the classpath CVS head.

      This puts us in a slightly awkward situation because at first glance
if we want Jikes RVM to build against both the classpath CVS head and
version 0.05 of classpath we're going to have to add more preprocessing to
our code to use a different java.io.FileDescriptor API depending on the
classpath version.  This may be unavoidable

There are two changes that are an issue:
      (1) getNativeFD was renamed to getNativeFd
      (2) the constructor that takes a unix fd and the setNativeFD method
were made private instead of package so we can't access them from
java.io.JikesRVMSupport anymore.

      Julian, could you take a look.  This is a little harrier than I want
to deal with myself.  In the long run I suspect the right answer for
JikesRVM is to push for a VMFileDescritptor. We may need to provide a patch
for this to make it happen.  Especially since in the long run we're going
to want to restore the use of non-blocking file io that we had in the
hybrid libraries.

      I'd like to try to get some resolution to this fairly soon because if
we can't build against classpath CVS head in our nightly regression tests
we risk more surprises like this in the future....

--dave

_______________________________________________
Jikesrvm-core mailing list
Jik...@ww...
http://www-124.ibm.com/developerworks/oss/mailman/listinfo/jikesrvm-core