This is the mail archive of the
java@gcc.gnu.org
mailing list for the Java project.
Re: over-eager verification
Tom Tromey wrote:
>>>>>>"Per" == Per Bothner <per@bothner.com> writes:
>
> Per> We should verify a method the first time it is invoked, not when
> Per> its class is prepared, since since most methods will never get
> Per> called.
>
> I don't think we can do this. See JVM Spec Section 4.9.1, Pass 3:
>
> Still during linking, the verifier checks the code array of the
> Code attribute for each method of the class file by performing
> data-flow analysis on each method.
>
> So it explicitly requires checking during preparation and not at
> invocation.
Who cares? It's just a book. I know the guy who wrote it ...
We have to look at the big picture: Making sure that invalid code
does not violate the Virtual Machine constraints is the big picture.
> Per> (Actually, ideally verfication should be done in conjunction with
> Per> JIT-compilation, since a JIT will need the type information that
> Per> the verifier has recoved.)
>
> This seems unlikely given our desire for modularity, and also that the
> first JIT will most likely be ORP, which as far as I could see has no
> way of feeding it this info.
I would think any decent JIT already does 90% of the job of
verification, at least if it does any optimization. So the most
efficient implementation would be to (if necessary) fix the JIT to do
the remaining 10%.
In any case, this is something to think about later.
--
--Per Bothner
per@bothner.com http://www.bothner.com/per/