[RFC][icedtea6]: RH706066

[Denis Lila]

> Think very carefully before back-porting this. I have not seen any
> compiler fixes for inference failures that did not break something
> else.
> Unless this is a real-world problem that affects some high-value
> application we probably should not do it.
> 
> Andrew.

I see.
Perhaps it would be better to suggest changes in the source of the
failing program then? I looked at the failures, and they all
seem pretty simple to fix: just changing "return find(...)"
to "return ThisClass.<T,S>find(...)" should be enough.

Regards,
Denis.

----- Original Message -----
> On 06/01/2011 05:40 PM, Denis Lila wrote:
> 
> > This bug will require a couple of backports:
> > http://hg.openjdk.java.net/jdk7/jdk7/langtools/rev/36c4ec4525b4
> > http://hg.openjdk.java.net/jdk7/jdk7/langtools/rev/dc550520ed6f
> >
> > The attached patch does this. It doesn't contain the regression
> > tests T6369605a.java and T6369605b.java because even after the
> > patch was applied these tests were failing.
> >
> > I think this is ok though because they fail with ecj and
> > proprietary javac too. What's probably happening is that
> > they're assuming other changes in the compiler in addition
> > to the fix for 6369605.
> >
> > I've attached three tests I wrote that were derived from
> > commons-discovery (whose compilation fails, as detailed
> > in the bug report). All three compile with ecj and proprietary
> > javac, but they fail with openjdk javac before this backport.
> > With the backport they compile.
> >
> > I haven't tried compiling commons-discovery yet. I will
> > do that soon and give an update.
> >
> > I ran the langtools regression tests, and there were no changes
> > in the results.
>