-
1. Re: *Info classes internal array structures
juha Dec 10, 2001 2:21 PM (in response to squirest)I don't there's a problem in making a copy in the MBeanInfo constructor. It is built once and lives through the whole lifecycle of an MBean, so the cost in constructor is not that critical.
The accessor methods are similar in this regard. Since the management interface of an MBean is constant, I'd expect most client programs to retrieve the management interface description only once, or rarely.
However, the Model MBean metadata classes implement the DescriptorAccess interface, which allows the descriptors to be changed, in other words, the Model MBean info is not completely immutable. We need to force the setDescriptor() methods to go through the interceptor stack in order to implement ACLs for different descriptors. Hmm.. need to think about that some more. -
2. Re: *Info classes internal array structures
squirest Dec 10, 2001 3:20 PM (in response to squirest)I agree MMB and DescriptorAccess important - though I'm not sure it's too bad 'cause the *Infos are still immutable.
I think the spec says somewhere about that while the ModelMBean *can* change its capabilities (*Info stuff) during its lifetime, doing so isn't guaranteed to work with all mgmt interfaces (or something like that). Think of a servlet HTML console which caches the *Infos. Heck *you* even mention it in your mail when you say "I'd expect most client programs to retrieve the management interface description only once, or rarely".
As far as the ModelMBean and DescriptorAccess guff is concerned, the MMB should just do what it's told whenever (imho).
The thing that gives off the greatest smell is the damn ModelMBeanInfoSupport class - it extends MBeanInfo *and* it has a copy constructor. Ewwww.
It was that hierarchy that got me thinking about being ultra-paraniod and cloning everything in and everything out.
Trev -
3. Re: *Info classes internal array structures
juha Dec 10, 2001 3:38 PM (in response to squirest)yes go ahead and clone
-
4. Re: *Info classes internal array structures
squirest Dec 11, 2001 12:09 PM (in response to squirest)*Info classes with array members are done, will be checking in soon.
Moving on to ObjectName, got an idea to keep it quick while still immutable. Will add support for patterns at the same time.
That ok?
Trev -
5. Re: *Info classes internal array structures
juha Dec 11, 2001 1:31 PM (in response to squirest)yes, go ahead
ObjectName.equals() seems to have the most effect performance wise in a HashMap based registry, so if you can figure out a way to speed that up, you will likely see the result directly in invocation times. -
6. Re: *Info classes internal array structures
squirest Dec 11, 2001 2:57 PM (in response to squirest)I was going to attack it on two levels.
First, precompute (during construction) the return value for ObjectName.hashCode() based on the hashCode() of the canonical name string. This is safe 'cause it's immutable and it should make equality searches faster (right?)
Second, I was going to create a read-only subclass of java.util.HashTable so that getKeyPropertyList() didn't have to keep returning copy-ctor'd HashTables. This should make thirt-party (i.e. MBeanServer) pattern comparisons less costly.
If you spot flaws in my logic pls let me know.
Trev