I think I found a solution that covers all Java version specific
Pack200 problems that we've encountered so far. The jarprocessor needs
to follow two strict rules.
1. Only pack (or condition) jar files that contain .class files or
nested jars (i.e. typically never pack source bundles and features).
2. A jar that contain nested jars but no .class files must be packed
using option -E0 (effort level 0).
These two rules will ensure that no java version dependent Pack200
magic is used unless a .class file is present in the jar. -E0 means
that Pack200 switches to "Copy-mode" in which case no version dependent
magic is used. That's ideal for jars that contain nested jars but no
.class files.
See also https://bugs.eclipse.org/bugs/show_bug.cgi?id=275533
- thomas
John Arthorne wrote:
Yes, this is a potential problem. I
think many of our packed artifacts currently require the Java 6 version
of pack200 for unpacking. Another example I noticed this morning is
source
bundles, which don't contain any class files, are not packed in a
backwards
compatible way. I entered bug 275295 for this. I can't remember if
there
was a strong reason for using the java 6 edition of pack200 for the
packing.
Obviously if we packed with the java 5 edition of pack200, we wouldn't
have this compatibility problem.
Pack200 has one fairly serious quirk that might
have
an effect on how we
want to set up rules in P2. Here's an excerpt from the docs:
"In order to maintain backward compatibility, if the input JAR-files
are
solely comprised of 1.5 (or lesser) classfiles, a 1.5 compatible
pack
file is produced. Otherwise a 1.6 compatible pack200 file is
produced."
This perhaps looks simple and clear, but there is one implication that
isn't mentioned and that is what happens when there are zero classfiles
in the jar. Zero classfiles means that Pack200 see no reason to be
backward compatible with 1.5. No big deal right? Just refrain from
packing such files. But no, we need to look deeper. Most jars that are
nested, contains zero class files in the top jar. And that's where it
has implications on P2.
Consider the following:
A.jar contains B.jar and C.jar. Both B.jar and C.jar contain class
files
but A.jar does not. With our current approach, this leads to an A.jar
that, if packed by a Java 6 JRE, only can be unpacked by a Java 6 JRE.
The fact that all .class files are Java 1.5 compatible doesn't help. So
what can we do?
Let's assume that we only pack jars that contain .class files. This
would mean that we get A.jar that now contains B.jar.pack.gz and
C.jar.pack.gz. This doesn't work very well, since P2 now only sees the
A.jar. It doesn't know that it contains nested jars that needs to be
unpacked. We cannot call the jar A.jar.pack.gz either since P2 then
assumes that it's packed. What we end up with is a container with
packed
material that cannot itself be packed. How do we solve that?
A workaround for the problem would be to add a dummy 1.5 compliant
.class file when this situation occurs. The unpacker could remove it so
it would never become visible.
Thoughts?
- thomas
_______________________________________________
p2-dev mailing list
p2-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/p2-dev
_______________________________________________
p2-dev mailing list
p2-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/p2-dev
|