[Springframework-developer] JDK 1.3 compatibility (again)

[Darren D. <da...@da...>]

After a recent mail about Spring issues in JDK 1.3, I did a little more=20
testing on my own machines, and the SF CF machines.  It's quite important=20
to me as we're one of the laggard companies that still uses it via=20
WebSphere 5.0.x!

I ran the entire test suite (built obviously with a 1.4 JDK) on 1.3 and was=
=20
actually pretty pleased with the results.  There were 6 failures and 24=20
errors.  Here's a brief breakdown for those interested;

Of the 24 errors..

 * 9 are due to references to java.sql.Savepoint in the test cases,

 * 7 are due to running tests explicitly designed for JDK 1.4 (for example=
=20
Jdk14ControlFlowTests and JdkRegexpMethodPointcutTests) which can be=20
excluded with test.excludes=3D...  I renamed JdkRegexpMethodPointcutTests t=
o=20
Jdk14RegexpMethodPointcutTests if that's ok?  Makes it easier to exclude=20
tests by naming patterns.

 * 2 are in a  large test class and reference java.util.prefs - these tests=
=20
have been amended to simply return if JDK 1.3 is detected.

Of the remaining 6, 2 are caused by the new=20
org.springframework.core.CollectionFactory since it doesn't check the=20
initialCapacity parameter before creating a Commons Collection instance on=
=20
JDK 1.3.  Commons Collections cannot have an initial capacity of 0.  I've=20
changed the code in this class and added corresponding tests Juergen.  Just=
=20
committed if you want to check.

The remaining 4 errors probably need looking at on a case by case basis to=
=20
see if they would cause usage issues or if they are just test code issues =
=20
=2D I haven't had time to check these yet.

The 6 failures are all in ResourceBundleMessageSourceTests and are mildly=20
concerning as I think this would cause a usage failure if anyone depended=20
on this functionality.  It seems (without going into it closely) that all=20
the failures occur when a resource bundle is referenced with a Locale that=
=20
specifies a country, language AND a variant in its constructor.  It seems=20
that the variant is ignored and the resource bundle corresponding to the=20
country and language only is used instead.

I'll post more later.  Got to go sort out my son's birthday party now.


=2D-=20
Darren Davison
Public Key: #DD356B0D