RE: [Ikvm-developers] RE: ikvm performance

[Jeroen F. <je...@su...>]

Jonathan Pierce wrote:
> I looked at the generated assembly and answered some of my=20
> own questions.
>=20
> In general, if the compilation generates code that always=20
> fails at runtime, then ikvmc should consider it an error,

It will probably become an option to specify this behavior (or maybe the
other way around). Surprisingly, there are actually quite a few Java
applications that have such incorrect code (that's never called).

> I see how ClassNotFound issues during ikvmc compile time can=20
> be treated as a WARNING since they can recover the missing
> class at runtime using ByteCodeHelper.DynamicInvokevirtual

Right.

> For missing methods, you call:
>=20
> ExceptionHelper.ThrowHack(new=20
> NoSuchMethodError("javax.swing.JDialog.()V"));
>=20
> It would be better if you could generate code that could=20
> recover at runtime for methods that are no longer missing
> in dynamically referenced assemblies or assemblies that were
> changed since ikvmc was run on this assembly.

Interesting idea, but why would you want this? Remember that the is a
steep performance penalty for such dynamic access. Wouldn't it be better
to just recompile the assembly?

Regards,
Jeroen