Slooooow debugging in linux

The difference could be in concrete JPDA implementation. Could you please take a cpu snapshot on Linux?

--
Best regards,
Eugene Zhuravlev
Software Developer
JetBrains Inc.
http://www.jetbrains.com
"Develop with pleasure!"

"Alexander Babaev" <no_mail@jetbrains.com> wrote in message news:31382502.1139144460759.JavaMail.itn@is.intellij.net...
>I debug web application in windows and linux (Resin/JDBC). In windows debug stepping is fast, in linux it is much, much slower. In
>windows I can press F8 several times without delay, it is normal, but in linux I have to wait several seconds after every press.
>

I use same EAP (4148) version everywhere.

>

Linux - Gentoo, kernel 2.6.15, Pentium-M 1.7GHz with 1Gb memory...

I've got a very similar problem. Take a look at this thread: http://intellij.net/forums/thread.jsp?forum=27&thread=195506&tstart=0&trange=15

I set up a breakpoint, reached it and stepped some times. Here is a cpu snapshot.

Thanks Alexander, a few questions:
1) Have you always had performance troubles on Linux, or previous IDEA builds behaved better?
2) How many threads are shown in the ThreadsView when the program hits the breakpoint?
3) In Debugger toolwindow, try to switch to Console view and perform several stepOver actions. Does it work faster?
4) Are there any method or exception breakpoints enabled or breakpoints with complex conditions? By "complex" conditions I mean any expressions that include method invocation
5) Try to turn off "Alternate views for Collection classes" and disable "toString()" object view. Does it work faster?

Thanks in advance for any information about this.

Answers:

1. Yes, it was definitely better. I worked in Linux for about half a year (during spring/summer of 2005) - EAP builds of Idea 5 were better. In autumn I worked in Windows, and now switched to Linux again... Late 5.1 builds were bad (final 5.1 is little faster).
2. About 20 threads, it's a usual Resin web application, about 4 threads are mine, others are Resin's.
3. Yes, if I switch to Console/Watch views - stepping goes faster, I use this workaround from time to time.
4. No, only usual breakpoints.
5. No, it doesn't help. Now I've all these settings turned off, it do not affect speed.

Also hotswap is much slower, than it was in EAP 5 builds. I liked HotSwap speed that time (it was very fast), after that I used Idea 5 in windows - HotSwap was slow there, but now it is slow in Linux too! May be it is same issue?

Thanks.

Can I hope, that slow linux debugging problem will be resolved?

Yep, We'll try to reproduce this and fix if possible. The problem is that on our linux boxes the speed is quite acceptable.

--
Best regards,
Eugene Zhuravlev
Software Developer
JetBrains Inc.
http://www.jetbrains.com
"Develop with pleasure!"

"Alexander Babaev" <no_mail@jetbrains.com> wrote in message news:29612499.1140013727027.JavaMail.javamailuser@localhost...

Can I hope, that slow linux debugging problem will be resolved?

May be, you need more cpu snapshots? In community forum there is similar topic. There are many of us, so, I think that it would be not difficult to collect statistics.

Please see the discussion:
http://www.jetbrains.net/jira/browse/IDEA-6540?page=comments#action_45886

I've noticed you are also using kernel 2.6.15. Is it possible to check IDEA behaviour on kernel 2.6.14?

Yes, I'll try it right now.

It seems, that 2.6.14 really solves the problem. In some days I tell you more information, but first feeling is that debugging became as fast as in windows again.

Thank you very much.

It would be interesting at this point to see if anyone has discovered a bug in the 2.6.15 kernel that might be related to this sort of thing. Maybe it forces the nagle algorithm in some cases? Maybe it has a bug in interprocess socket communication? Maybe thread scheduling has changed slightly?

I'm using ubuntu 6 (kernel 2.6.15.24) and it is definitly slow. The command "sudo sysctl -w net.ipv4.tcp_abc=0" has made things much faster though, and debugging is almost as fast as in windows.