Bug 189456

Summary:

Formatter is slow on big files

Product:

[Eclipse Project] JDT

Reporter:

David Audel <david_audel>

Component:

Core

Assignee:

Olivier Thomann <Olivier_Thomann>

Status:

VERIFIED FIXED

QA Contact:

Severity:

normal

Priority:

CC:

frederic_fusier, jerome_lanneluc, maxime_daniel, philippe_mulet

Version:

3.3

Keywords:

performance

Target Milestone:

3.3 RC3

Flags:

frederic_fusier: review+
jerome_lanneluc: review+
philippe_mulet: review+

Hardware:

OS:

Windows XP

Whiteboard:

Attachments:

Description	Flags
Proposed fix	none
Performance tests for formatter	none

Description David Audel

2007-05-28 11:22:08 EDT

build 3.3RC2

It take around 20sec to format GenericTypeTest on my computer.

I profile the execution of the format operation.
The format operation last 17sec and 12 sec are spend inside CharOperation#indexOf(char[], char[], boolean, int, int).

The problem could be in Scribe#hasNLSTag() and Scribe#printLineComment() which call indexOf() to search an nls tag from the current position to the the end of the file.
The fact that the file is big and that there is no real nls tag inside could be responsible of the slowness.

Comment 1 Olivier Thomann

2007-05-28 22:34:18 EDT

Created attachment 68989 [details]
Proposed fix

David, could you please let me know if it is faster with this patch?
If yes, I'll request reviews for RC3.

Comment 2 Olivier Thomann

2007-05-29 09:56:41 EDT

Maxime, Jérôme and Frédéric, please review.
If ok, I'll tag it as 3.3RC3.

Comment 3 Frederic Fusier

2007-05-29 10:11:45 EDT

Adding a new performance test suite with one test formatting a big file
(GenericTypeTest.java), here are the results I got:

3.2     15,776.5ms
RC2:    15,779.7ms
patch:   3,586.0ms

So, sounds like a good candidate for global fingerprints :-)

Comment 4 Frederic Fusier

2007-05-29 10:12:19 EDT

Obviously +1 for 3.3 RC3 with previous performance results...

Comment 5 Jerome Lanneluc

2007-05-29 10:50:32 EDT

+1 for 3.3 RC3

Comment 6 Frederic Fusier

2007-05-29 11:00:22 EDT

Here are the results I got when formatting is done on a medium file
(Parser.java: 225176 chars):

3.2     1568.7ms
RC2:    1567.1ms
patch:   684.4ms

Comment 7 Philipe Mulet

2007-05-29 11:02:47 EDT

+1 for rc3

Comment 8 Olivier Thomann

2007-05-29 11:04:59 EDT

Philippe, please review

Comment 9 Philipe Mulet

2007-05-29 11:06:40 EDT

Patch looks good.

Comment 10 Olivier Thomann

2007-05-29 11:12:19 EDT

Released for 3.3RC3

Comment 11 Frederic Fusier

2007-05-29 11:15:08 EDT

Created attachment 69071 [details]
Performance tests for formatter

Add new performance test suite: FullSourceWorkspaceFormatterTests.

This suite has 2 tests:
- testFormatDefault:
    format Parser.java file (225176 chars) ten times
- testFormatDefaultBigFile:
    format big file GenericTypeTest.java (1297242 chars) once

10 measures (ie. start/stop) are done for each test with a warmup of 5 iterations...

Comment 12 Frederic Fusier

2007-05-29 11:46:25 EDT

(In reply to comment #11)
> Created an attachment (id=69071) [details]
> Performance tests for formatter

Released for 3.3 RC3

Comment 13 David Audel

2007-06-01 11:30:20 EDT

Verified for 3.3RC3 using I20070601-0010.

When folding is enabled the code is formatted in the editor after 20 sec.
When folding is disabled the code is formatted after 5 sec.