Hi Sahoo,
You make a good point about measuring the effect of the optimization.
We do not yet have any testing of that kind, but it would definitely be
a good idea. We might want to enter a TASK for this.
I also agree that avoiding creating the TempClassLoader when WEAVING =
false would be a good idea. An REF sounds good.
Thanks,
Tom
Sanjeeb Kumar Sahoo wrote:
>Hi Tom,
>
>I agree, your change affects only Java SE deployment.
>
>Do you have any kind of test that shows the result of optimization done
>in TempClassLoader? An additional reading of XML file might negate the
>class loading optimization that is put in place in TempClassLoader to a
>large extent. If you have some numbers showing improvement in
>performance, then that will help when I try to address
>https://glassfish.dev.java.net/issues/show_bug.cgi?id=1147.
>
>Also, there are other optimizations that can be done. We can set up
>application class loader as TempClassLoader when WEAVING = false and in
>such a case, we will avoid creating TempClassLoader and there by
>executing the method you have added. Do you agree? Should we file an RFE
>for this?
>
>Thanks,
>Sahoo
>Tom Ware wrote:
>
>
>>Hi Sahoo,
>>
>> I believe this change will only affect an SE deployment. We only
>>call this method from our JavaSE initialization code. It is used to
>>build a set of classes used to optimize the Temporary Class Loader we
>>use in SE.
>>
>> So in answer to your question, yes, we will process files more than
>>once, but it will only occur in SE deployments, and I am not sure it
>>is really avoidable as long as we are optimizing the Temp class loader
>>the way we are.
>>
>> As a result, I do not think 1115 needs reopening.
>>
>>What do you think?
>>
>>-Tom
>>
>>Sanjeeb Kumar Sahoo wrote:
>>
>>
>>
>>>Hi Tom,
>>>
>>>I have not run any test to validate what I am writing here, but I
>>>thought I should let you know any way. I now see a new method called
>>>*private static Set<String>
>>>buildPersistentClassSetFromXMLDocuments(PersistenceUnitInfo,
>>>ClassLoader)* in PersistenceUnitProcessor class. This method creates
>>>a MetadataProcessor and reads all the mapping XML files in order to
>>>build a list of classes and throws away the MetadataProcessor. As a
>>>result of this change, are we not reading XML mapping files more than
>>>once? Shall I reopen
>>>https://glassfish.dev.java.net/issues/show_bug.cgi?id=1115 ?
>>>
>>>Thanks,
>>>Sahoo
>>>
>>>
>>>
>>>