Eclipse Community Forums: Eclipse 4 » InjectorImpl Performance

Help

Home

Home » Eclipse Projects » Eclipse 4 » InjectorImpl Performance

Show: Today's Messages :: Show Polls :: Message Navigator

InjectorImpl Performance [message #894827]

Tue, 10 July 2012 12:49

Joseph Carroll
Messages: 156
Registered: May 2012
Location: Milwaukee, WI

Senior Member

I am still working to fully understand the DI engine, but as I was reviewing the code I noticed that the class (InjectorImpl) is not declared as final. Is there a reason that it is not declared as final? Wouldn't there be a performance increase if it were because the compiler could "in-line" the code?

[Note to self: read "Java Performance"]

[Updated on: Tue, 10 July 2012 12:55]

Report message to a moderator

Re: InjectorImpl Performance [message #894851 is a reply to message #894827]

Tue, 10 July 2012 14:52

Brian de Alwis
Messages: 242
Registered: July 2009

Senior Member

Give it a try and report back Smile

To be honest, I think the bigger win would be to use ASM or something similar and generate the injection code. This is what Guice does and is a big part of its speed.

Brian.

Report message to a moderator

Re: InjectorImpl Performance [message #894853 is a reply to message #894851]

Tue, 10 July 2012 15:17

Joseph Carroll
Messages: 156
Registered: May 2012
Location: Milwaukee, WI

Senior Member

Brian-

I'll give it a try, hopefully sometime this weekend I can have results. (Tonight I need to work on the PerspectiveSwitcher!)

Are there any existing performance tests for the framework?
I see the standard unit-tests, but there aren't any performance tests.

The other thing I would be interested in knowing is if the "HotSpot" JVM is doing its job and actually compiling down the bytecode. If methods are too long, the JVM will keep the method as bytecode even if it is a "hot" method. The issue is that longer methods are more difficult for the JVM to understand and optimize (to put it simply). Unfortunately, there is no single definition of a "long" method.

I might need some help in thinking up the logic for any performance tests, if none are presently available. But then I can look at both finalization and compilation of the different methods and see if anything makes a difference.

Report message to a moderator

Re: InjectorImpl Performance [message #894987 is a reply to message #894853]

Wed, 11 July 2012 06:54

Sopot Cela
Messages: 448
Registered: December 2010

Senior Member

Hooking up JVisualVM to a running instance of Eclipse might help.

Report message to a moderator

Re: InjectorImpl Performance [message #895038 is a reply to message #894987]

Wed, 11 July 2012 09:35

Joseph Carroll
Messages: 156
Registered: May 2012
Location: Milwaukee, WI

Senior Member

My thoughts exactly, I was planning on using the profiling tool that is provided. And then I was also going to use the OpenJDK as it offers greater insight into the JIT Compiler.

That really leaves just the test case. Is there any documentation specifying performance benchmarks? See the event broker performance benchmarks as a reference:
http://wiki.eclipse.org/E4/Event_Processing#Eclipse-specific_design_parameters

Report message to a moderator

Previous Topic:	Migration from eclipse 3.x
Next Topic:	Dragging source files between Windows

Goto Forum:

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

]

Current Time: Fri May 24 13:54:10 EDT 2013