Discuss JaegerMonkey and bugs here

[nemm]

It was just reported to me that today's nightly build has issues/crashes:

today's trunk build crashes on non SSE2 CPUs with Jaegermonkey turned on
http://crash-stats.mozilla.com/report/i ... 4d02100912

I'm having instant crashes with JaegerMonkey build. As soon as I open Firefox and enter some address, Firefox crashes. I reverted to the previous build. Bummer.

http://crash-stats.mozilla.com/report/index/bp-4327fd58-aff9-49df-94bc-13f0f2100911
http://crash-stats.mozilla.com/report/index/bp-ec66bb4a-3e0e-4263-8c54-7769b2100911
http://crash-stats.mozilla.com/report/index/bp-acbeebe5-ac9e-47ff-a124-e0efb2100911
http://crash-stats.mozilla.com/report/index/bp-9695ddae-c634-4e68-b4c7-c3d6c2100911

[insano]

Is this a known JM Bug?
Clean profile, methodjit enabled on content, tracejit disabled on content, logon to GMail, click on any mail. Upon trying to return to Inbox or any other folder, firefox either stops with the "Loading... "message for GMail or it finishes loading but shows no emails.
It works with tracing and method jit enabled.

[bala_gamer]

sys: e7200 @ 3.6 ghz
9600 gt 1gb ddr3
4gb DDR 2 @ 880 mhz


Richards scores best in this build....
Lets hope others do the same in the upcoming builds

[sabret00the]

So just to confirm, to turn JM on, we set all of the prefs in the first post to true?

Yes inquiring mind want to know?

methodjit.chrome is not yet on by default. More so that the others, use at your own risk.

EDIT: tracejit.chrome was not on by default until well after tracejit.content was on by default. JM is mimicking TM in that regard.

Sorry, my brain is like swamp vomit today, I'm sure I'll read this tomorrow and wonder what was wrong with me. But was that a yes? Turn all to 'true'?

[WildcatRay]

@sabret00the,

As I said, methodjit.content--JM for web page content--is now on by default. methodjit.chrome--JM for the browser and its add-ons--is off by default. That means in the eyes of the developers, you turn on methodjit.chrome AT YOUR OWN RISK, that it is not yet ready for full blown testing on the Trunk builds. Thus, just keeping the default settings you will be using/testing JM for web page content.

Also, looks like the Google Maps JM bug has been fixed. I was able to view, get directions and move and scroll maps on there using today's build.

[WonderCsabo]

Yes.

bala_gamer: V8 is a very wrong test, and Fx wont be faster on that than Chrome. They're focusing on real world cases.

[Terepin]

Around 460ms in SS, not so much improvement, plus entire browser reacts a lot worse. Now this is what I'm calling "speed up"...

[sabret00the]

@sabret00the,

As I said, methodjit.content--JM for web page content--is now on by default. methodjit.chrome--JM for the browser and its add-ons--is off by default. That means in the eyes of the developers, you turn on methodjit.chrome AT YOUR OWN RISK, that it is not yet ready for full blown testing on the Trunk builds. Thus, just keeping the default settings you will be using/testing JM for web page content.

Also, looks like the Google Maps JM bug has been fixed. I was able to view, get directions and move and scroll maps on there using today's build.

Thanks and my apologies for having to make you type all of that.

[neo86]

20-30 FPS in chrome but only 1-2 FPS in Firefox JM+TM. I tried enabling/disabling varying combinations of D2D/DW, D3D layers, and trace JIT content but I couldn't find any combination that worked above 2 FPS. Athlon X4 635 + Radeon HD4650.

Did you confirm acceleration was off in about:support? I think the fps drops after a certain point regardless of which browser as that's the nature of the demo, but with D2D/layers on it never goes above 2 fps even at the start. With D2D disabled it seems to at least start off at 7 fps for me before dropping.

[Cru_N_cher]

i pinpointed my problem with joost.com wasn't about the localstorage @ all but the dom storage (which if you take it seriously is the same just another option)
http://kb.mozillazine.org/index.php?tit ... ntable=yes

joost.com doesn't like that to be disabled (and i guess many other Next gen Web apps also not, especially if they use the dojo framework it seems)
and yeah the only bug on firefox sides seems that -safe-mode resetting doesn't reset it properly to true if it's falsed so bug report on it's way (case closed, puhh i was scared that i had todo the whole profile conversion)

so no JM/TM problem @ all but a nasty Next Gen Web app thing, and a small bug in -safe-mode profile resetting

so whenever you see this
Error: uncaught exception: [Exception... "Security error" code: "1000" nsresult: "0x805303e8 (NS_ERROR_DOM_SECURITY_ERR)"

take first a look @ the dom.storage.enabled setting, most probably it gonna fix it if you enable it (which it is by default)

[sciguyryan]

Yes.

bala_gamer: V8 is a very wrong test, and Fx wont be faster on that than Chrome. They're focusing on real world cases.

Quite true. I'm glad it's not only me that realizes that one

[Cru_N_cher]

though this SCS could give a whole new way for user tracking (if they aren't enclosed in their own Application space) yet another one independent of Cookies or Flash Cookies (also all applications that you currently use cant avoid this) not really comfortable to know that. So i will try some tricking for such Web apps and see if i can get the Noscript or Addblock devs to implement them if they are successful

[mattcoz]

Here are my SunSpider results, from slowest to fastest:

IE 9 Preview 3: 637.4ms +/- 0.4%
Firefox 4b6pre: 585.5ms +/- 4.3%
Opera 10.6.1: 456.2ms +/- 1.2%
Chrome 7: 445.9ms +/- 2.0%

Respectable result for the initial merge. This is with a fresh profile, with my default profile I get much higher and much more inconsistent results. I'm sure that has to do with some extensions I have installed.

I'm getting a crash on startup when using methodjit.chrome, but that's not expected to be stable yet. No crashes or bugs otherwise yet.

[WhtDwrf]

It was just reported to me that today's nightly build has issues/crashes:

today's trunk build crashes on non SSE2 CPUs with Jaegermonkey turned on
http://crash-stats.mozilla.com/report/i ... 4d02100912

Insta-crash on Updating to 'Mozilla/5.0 (Windows NT 5.1; rv:2.0b6pre) Gecko/20100912 Firefox/4.0b6pre' (Built from http://hg.mozilla.org/mozilla-central/rev/cd3c926a7413);
Minefield just starts opening then:

http://crash-stats.mozilla.com/report/index/bp-20d59534-6a90-4b4f-9ee6-335bd2100912
http://crash-stats.mozilla.com/report/index/bp-b9757314-8a7b-4147-bc21-26a332100912
http://crash-stats.mozilla.com/report/index/bp-d1a75635-91f5-415d-90ad-d0ca92100912
http://crash-stats.mozilla.com/report/index/bp-07c251d8-7a24-4d63-a575-093902100912
http://crash-stats.mozilla.com/report/index/bp-5f68dcf5-60ab-489b-9693-0c2842100912
...
All except bp-b9757314-8a7b-4147-bc21-26a332100912 are '[@ EnterMethodJIT]'.

And, all are do to "Crash Reason: EXCEPTION_ILLEGAL_INSTRUCTION" (most likely an SSE2, (3, 4...) instruction trying to be ran on my perfectly good Athlon XP 2600+ )...

Bugs open for above crashes:
Bug 595351 - Crash in [@ EnterMethodJIT ]
Bug 595635 - crash in [@ js::Invoke(JSContext*, js::CallArgs const&, unsigned int) ]

Temporary workarounds: (that WFW, hopefully more)

1) It seems running Minefield in safe mode, does not crash, checked about:config and all JIT options were default:
javascript.options.methodjit.chrome;false
javascript.options.methodjit.content;true
javascript.options.tracejit.chrome;true
javascript.options.tracejit.content;true
Why safe mode works? Extensions causing JIT to execute SSE2 instructions? IDK
2) Turn off javascript.options.methodjit.content; javascript.options.tracejit.chrome; javascript.options.tracejit.content then don't need to use safe mode.
Ed: 3) So far it looks like you can leave javascript.options.tracejit.chrome & javascript.options.tracejit.content = true, turning on javascript.options.methodjit.content causes crash (as soon as I change the config option).

[BigWillyStyle42]

1) It seems running Minefield in safe mode, does not crash, checked about:config and all JIT options were default:

javascript.options.methodjit.chrome;false
javascript.options.methodjit.content;true
javascript.options.tracejit.chrome;true
javascript.options.tracejit.content;true
Why safe mode works? Extensions causing JIT to execute SSE2 instructions? IDK

Safe mode disables JIT.