OutOfMemory using scalac in ant

On Mon, Jun 14, 2010 at 4:33 PM, Vladimir Kirichenko <vladimir.kirichenko@gmail.com> wrote:

when script with scalac invoked multiple times via ant-call (4 is
enough) it causes:  java.lang.OutOfMemoryError: PermGen space

As the name implies, that space is not GC'd. You can increase it with -XX:PermSize=...
What is an ant-call? Are you using the scalac task, or? Please be more specific.
 

https://lampsvn.epfl.ch/trac/scala/ticket/3565

RC5

BTW is https://lampsvn.epfl.ch/trac/scala/ticket/3486 (missing java
signatures) going to be fixed before 2.8 final? If it's not - it's a
major step back in java integration of scala generics :(

That never worked, so it can't be a major step back. It would be a major step forward if it was fixed (and I agree with you it's important). Unfortunately we're way too far in the RC process to add new (risky) code. It will be fixed in 2.8.1, which isn't that far away.
iulian 

--
Best Regards,
Vladimir Kirichenko

--
« Je déteste la montagne, ça cache le paysage »
Alphonse Allais

On Tue, Jun 15, 2010 at 5:01 PM, iulian dragos wrote:
> On Mon, Jun 14, 2010 at 4:33 PM, Vladimir Kirichenko
> wrote:
>>
>> when script with scalac invoked multiple times via ant-call (4 is
>> enough) it causes:  java.lang.OutOfMemoryError: PermGen space
>
> As the name implies, that space is not GC'd. You can increase it with
> -XX:PermSize=...

While this is far too often true, it is not so by design. The PermGen
space is a part of the JVM's heap which contains stuff aside from
actual Java object instances. These are interned strings and class
metadata, bytecodes etc. If you encounter a PermGen OOM exception it
is most likely that classes that once were loaded are not being
released properly. These memory leaks are quite often since it is
really difficult to get (dynamic) class loaders stuff right. The
underlying problem is that classes loaded by one ClassLoader can only
be freed if there are no references to _any_ of the classes (or their
instances) of the class loader.

But as Iulian said, difficult to say what is happening without exactly
knowing what you are doing. Increasing PermGen space is at least a
short-term solution if you can estimate how much space you will need.

On Tue, Jun 15, 2010 at 05:37:46PM +0200, Johannes Rudolph wrote:
> If you encounter a PermGen OOM exception it is most likely that
> classes that once were loaded are not being released properly. These
> memory leaks are quite often since it is really difficult to get
> (dynamic) class loaders stuff right. The underlying problem is that
> classes loaded by one ClassLoader can only be freed if there are no
> references to _any_ of the classes (or their instances) of the class
> loader.

For a potentially interesting, very scala-specific illustration of the
above, see:

https://lampsvn.epfl.ch/trac/scala/ticket/2365
"structural type cache should not hold strong reference to classes"

...which is my favorite kind of bug to fix (I say that in the normal
human sense and not my frequently used reverse sense) because I had no
real idea going in what I'd have to do to fix it. It's like parachuting
behind enemy lines with nothing but a hunting knife and your wits. Oh,
and google. And maybe a nice cup of coffee.

https://lampsvn.epfl.ch/trac/scala/changeset/20538

iulian dragos wrote:
> What is an ant-call? Are you using the scalac task, or? Please be more
> specific.

I have added example to reproduce.