Bug 188656

Summary:

[perfs] 2% regression on some Batch Compiler tests since v_756

Product:

[Eclipse Project] JDT

Reporter:

Frederic Fusier <frederic_fusier>

Component:

Core

Assignee:

Olivier Thomann <Olivier_Thomann>

Status:

VERIFIED FIXED

QA Contact:

Severity:

normal

Priority:

CC:

jerome_lanneluc, Olivier_Thomann

Version:

3.3

Flags:

jerome_lanneluc: review+
frederic_fusier: review+

Target Milestone:

3.3 RC2

Hardware:

OS:

Windows XP

Whiteboard:

Attachments:

Description	Flags
Proposed patch	none
New patch	none

Description Frederic Fusier

2007-05-23 10:50:18 EDT

Since JDT/Core v_756, we observed a regression around 2% for Batch Compiler performance tests. Not all tests are concerned, testCompileJDTCoreProjectJavadoc and testCompileJDTCoreProjectAllWarnings seem to be flat.

Comment 1 Frederic Fusier

2007-05-23 10:54:39 EDT

Seems to be an impact of fix for bug 185768. Doc comment support is now activated when annotations are processed but as this option is set by default, the doc comment support is now always activated for the batch compiler hence the performance regression in our tests...

Comment 2 Frederic Fusier

2007-05-23 11:47:03 EDT

Created attachment 68369 [details]
Proposed patch

I post estimated perfs numbers when running local tests will be finished

Comment 3 Olivier Thomann

2007-05-23 12:17:30 EDT

+1. Process annotation should never be set for compliance < 1.6.

Comment 4 Olivier Thomann

2007-05-23 12:23:03 EDT

Sorry. I withdraw my +1. I checked too quickly.
If the option -proc:none has been set, we should not reenable the annotation processing.
So we should add a check to see if the option is already disabled. I'll provide a new patch.

Comment 5 Olivier Thomann

2007-05-23 12:40:25 EDT

Created attachment 68381 [details]
New patch

Comment 6 Olivier Thomann

2007-05-23 12:41:00 EDT

Frédéric, please review.

Comment 7 Frederic Fusier

2007-05-23 12:51:24 EDT

(In reply to comment #6)
> Frédéric, please review.
> 
+1

Comment 8 Jerome Lanneluc

2007-05-23 13:07:23 EDT

The comment should reference both bug 185768 and this bug. Otherwise +1.

Comment 9 Frederic Fusier

2007-05-23 13:14:19 EDT

I got following numbers while running local tests:

Proj	| jdt.core | jdt.core | jdt.core | jdt.core  | swt
sett.	| No_Warn  | Default  | Javadoc  | All_Warns | Default
--------|----------|----------|----------|-----------|--------
v_755	| 53315	   | 51834    | 54310    | 63771     | 23933
patch	| 53412	   | 51840    | 54246    | 63850     | 23978

These numbers show that we are back to M7 numbers with this patch.

Comment 10 Frederic Fusier

2007-05-23 13:16:39 EDT

I got a mid-air collision which seemed to have flushed review flags!
I reset them...

Comment 11 Frederic Fusier

2007-05-23 13:17:22 EDT

Olivier, Jerome, can you set again the review flag, please?
Thanks

Comment 12 Olivier Thomann

2007-05-23 13:48:12 EDT

I'll release for tonight's build.

Comment 13 Olivier Thomann

2007-05-23 14:03:52 EDT

Released for 3.3RC2.

Comment 14 Frederic Fusier

2007-05-25 04:26:38 EDT

Verified for 3.3 RC2 with performance tests results:
http://fullmoon.ottawa.ibm.com/downloads/drops/I20070524-0010/performance/org.eclipse.jdt.core.php?