Request for review (vs) - 7199923

[Jon Masamitsu]

On 9/20/2012 2:39 PM, Srinivas Ramakrishna wrote:
> Looks good. Was it mainly for the system class loader that this turned out
> to be a problem because of the
> very large number of classes it loads, although of course the linear walk
It was probably the system classes but I could not rightly tell from the
profiling.  I saw it with something as simple as GCOld so that it was
the system classes is a good bet.

We don't have a reason for a more efficient lookup of a class
other than this assertion checking so it doesn't seem worth
the code.  We considered a debug flag to control it but even
that seemed too much.

Thanks for the review.

Jon
> can result in quadratic lookup
> complexity which makes it bad in general. (The other thing would be to have
> the set be a tree (or a skip list),
> splitting by some key such as full class name. Anyway, OK to throw away the
> frequent assertion checking for now.)
>
> -- ramki
>
> On Thu, Sep 20, 2012 at 11:36 AM, Jon Masamitsu<jon.masamitsu at oracle.com>wrote:
>
>> 7199923: NPG: tools/javac/T7093325.java timeout
>>
>> The method has_defined() checks that a class is in
>> the classloader's list of classes. has_defined() was
>> used in an assertion when doing marking and
>> adjusting of pointers in a class. Doing that check
>> for every class spends too much time looking
>> down the list of a classloader's classes.
>>
>> http://cr.openjdk.java.net/~**jmasa/7199923/webrev.00/<http://cr.openjdk.java.net/%7Ejmasa/7199923/webrev.00/>
>>
>> I'd like to get this back for testing tonight so my
>> apologies if I cut short the review period.  Please
>> send your comments in any event and I'll fix
>> any problems in a second CR.
>>