Official firefox-pgo thread

Pank · 2009-11-04 17:53:31

I tried to recompile the problem still seems persistent.
I get the following message when firefox crashes:

/usr/lib/firefox-3.6b1/run-mozilla.sh: line 131: 28326 Segmentation fault      "$prog" ${1+"$@"}

Procyon · 2009-11-08 12:19:25

@Pank: Same issue here.

Ranguvar · 2009-11-09 20:37:11

Procyon, Pank, I usually get that problem when I don't get rid of clobber_all with sed, or when I compile PGO without doing a normal compile first.

There is a new firefox-pgo-beta written by me that enables TraceMonkey on x86_64, among other things. blasse modified it for a new firefox-pgo, but that one has a few bugs as of right now. (Fixed)

Last edited by Ranguvar (2009-11-11 06:40:00)

toxygen · 2009-11-11 06:37:19

just built the firefox-pgo-beta from Ranguvar's updated pkgbuild (in the aur page). worked well, but doesnt compare to 3.7a1 (in terms of the sunspider test):

TEST                   COMPARISON            FROM                 TO             DETAILS

=============================================================================

** TOTAL **:           1.12x as fast     1007.4ms +/- 6.7%   898.6ms +/- 11.9%     significant

3.6b2 on the left, 3.7a1 on the right (both built within the last hour). also the html5 video on http://www.spreadfirefox.com/5years/en-US/ plays for both, but audio only works on 3.7a1 for some reason (both have oss enabled).
Nice job with the PKGBUILD though. I'll see if i can change my current PKGBUILD for 3.7 with some of the things you did on the 3.6 one. mine's a mess

Ranguvar · 2009-11-11 06:43:03

Daaamn. I was planning on playing with 3.7a1 soon, now I'll definitely have to

And yeah, I have some issues with HTML5 audio on Firefox (that audio works okay with 3.6b2), glad to hear that will may go away with 3.7.

Anything I should know about 3.7 from your experience with it? Is clobber_all fixed, or the Hunspell patch applied, or does --enable-canvas3d work, or..? I already know that I won't need to patch to get 64-bit TraceMonkey
I'll test myself soon of course

Last edited by Ranguvar (2009-11-11 06:43:33)

toxygen · 2009-11-11 07:27:01

there are some bugs in 3.7, like the html 5 video buttons (on the gui for the video player) not showing anythign other than a gray box, and occassional crashes with html5 videos (dailymotion page mainly). i dont have canvas3d enabled, but i might play with that when i re-do my pkgbuild. TM is definitely enabled
i still cant get pgo to build due to the clobber-all line (using the sed workaround seems to work, but I'm not convinced it's actually building a pgo build, instead it seems to be doing a single build (see my previous post on this issue)).
other than that, it's been stable, i've been using 3.7 as my main browser for a while now, havent encountered any issues other than the previously mentioned (and xmarks extension crashes the browser).
definitely looking forward to what you can do with 3.7a1 though.

hanbin · 2009-11-11 10:59:23

I have a seg fault again~ I failing every time since I succeded my 6th try;;;

If firefox dosn't work, if I just the application blah options value ( originally browser ) to xulrunner, I can build a xulrunner then?

blasse · 2009-11-11 11:11:03

Better way would be bumping original xulruner pkg version using abs, then adding --enable-libxul-sdk=path_to_xulrunner to mozconfig. You will loose most of pgo goodies, but TM should be fine that way.

pyther · 2009-11-11 19:39:16

I submitted a bug report for the strange square characters/symbols that are being displayed. (3.6b2)
https://bugzilla.mozilla.org/show_bug.cgi?id=527980

Comment on the issue if it affects you!

yellowfox · 2009-11-14 18:53:03

blasse Why libgnomeui is in makedepends ?

blasse · 2009-11-14 19:01:20

Because it theoretically enables some functions in firefox... I don't really know, if it's true. It is only makdepend, so you can delete it just after compiling

Last edited by blasse (2009-11-14 19:06:10)

Ranguvar · 2009-11-14 22:25:28

Yeah, that's supposedly the case. Also, I like to keep modified packages as close to main Arch as possible, and Arch's xulrunner/firefox has a libgnomeui makedepend. So, there would be where to complain first about it if you think it makes no difference

Ranguvar · 2009-11-15 03:46:21

Another bug I noted where the toolbar buttons do not visually depress, please comment if you also have the issue:

https://bugzilla.mozilla.org/show_bug.cgi?id=528769

Solid1986Snake · 2009-11-15 10:43:43

is it possible to integrate the opensuse-kde patches? I had a quick look on them and could not manage it on my own....

blasse · 2009-11-15 11:33:03

Could you post a link to this patches?

mechmg93 · 2009-11-15 11:45:40

blasse wrote:

Could you post a link to this patches?

http://gitorious.org/firefox-kde-opensuse

harryNID · 2009-11-15 12:48:44

I'm at my wit's end here.

I'm building for i686.

I've tried as "Ranguvar" suggested and built this every which way but loose but I'm still getting the seg fault (as seen previously) at very erratic places with every build. I've tried normal, non-PGO and some other things. I though it might be the addons I have been forcing with Nightly Tester Tools so I created a new profile. Alas that wasn't the answer either. TraceMonkey is enabled as I'm able to do a partial Sunspider test before it crashes. It always CTD's during the Sunspider test and what's funny is it always does it almost right at the end with all the builds I've made. I also agree with "toxygen" that the sed fix for the clobber-all line is only allowing the build without creating an actual PGO optimized output.

I've been following the thread here and know this is somewhere upstream. I know clobber-all has been reported but has there been a bug report filed for this? If so could somebody post a link so I can follow it.
Right now I'm using firefox-pgo-beta 3.6a2-1 as that is the last build that was stable for me.

Any knowledge about this mysterious seg fault would be appreciated. I'd really like to figure this one out as the new build was absolutely flying for me in-between CTD's.

blasse · 2009-11-15 12:52:19

https://bugzilla.mozilla.org/show_bug.cgi?id=520704

Links for bug followed in this thread are in first post, on first site

Ranguvar · 2009-11-15 17:15:25

@mechmg93: Interesting, but as I am not a KDE user (nor is blasse IIRC), it would be difficult for me to test, so it's not of very high priority for me. However, if the patches do not affect normal non-KDE operation, and either a package pops up with the patches applied that I can borrow off or someone helps me by modifying the PKGBUILD to use those patches, then I may add that to the package

@harryNID: I'm going to post generic i686 and x86_64 builds soon, can you see if they (i686 for you, then) work? Obviously negates some of PGO's benefit, etc., but if it's the only way you can get something working...
I'm going to try to address the clobber_all problem with a proper fix sometime soon, hopefully. I have no Makefile experience, so I'll just have to learn

toxygen · 2009-11-15 17:34:22

is it me or is your avatar's beard getting longer?

Ranguvar · 2009-11-15 17:45:42

toxygen wrote:

is it me or is your avatar's beard getting longer?

Check the links in my sig

Ranguvar · 2009-11-15 22:04:57

I have a new package I'll upload soon, firefox-pgo-minefield. It uses the latest mozilla-central source (3.7a1pre right now), but it's not a normal Mercurial package. I will bump the pkgver manually every now and again when a revision passes Mozilla's Tinderbox unit tests.

EDIT: Never mind, clobber_all still a problem...
Button depression bug still exists. Haven't witnessed the strange symbols problem yet, but it occurred randomly in 3.6b2, so it might just not have reared its ugly head.

Here it is, uploaded now: http://aur.archlinux.org/packages.php?ID=32030

Last edited by Ranguvar (2009-11-16 06:37:24)

harryNID · 2009-11-16 00:51:00

Ranguvar, I just rebuilt the latest version of firefox-pgo-beta you posted today. I am writing this post from it as we speak. Unfortunately Sunspider is still seg faulting with this build for me. I haven't tested it long enough to see if it randomly does it again or not. I will of course try all options laid out on the table and keep you informed on how they work.

blasse, Thank you for posting the link. I had seen that before but I guess that is the heart of the matter heh? I hope the devs look deeper into this soon.

Thanks for the work guys it's much appreciated!:)

Ranguvar · 2009-11-16 01:23:57

For those who can't get a compile working (please let me know if these then work or don't):

3.6b2-2 x86_64 generic (w/ TraceMonkey patch): http://omploader.org/iMnMzYw
sha256sum: cc87c1e3dcd84c36c853558b79d622fd10478501dbc354d2394bc9d8da235be7

3.6b2-2 i686 generic (w/ TraceMonkey, as always): http://omploader.org/iMnMzdg
sha256sum: 657f27a4788e6b1a9b1c69148596a0c95b63928dc056c3c14ddffaa30447020b

Last edited by Ranguvar (2009-11-16 01:24:14)

toxygen · 2009-11-16 05:10:59

nice job on firefox-pgo-minefield! just built it, unsure if pgo is being build as before, but my sunspider improved from my last score:

TEST                   COMPARISON            FROM                 TO             DETAILS

=============================================================================

** TOTAL **:           -                 898.6ms +/- 11.9%   878.0ms +/- 3.5%

will test during the next few days. thanks!

[edit] btw, where do i find the _workingrev in mozilla-central? is it the number next to the diff in this page?
http://hg.mozilla.org/mozilla-central/

Last edited by toxygen (2009-11-16 05:12:57)

Arch Linux

#401 2009-11-04 17:53:31

Re: Official firefox-pgo thread

#402 2009-11-08 12:19:25

Re: Official firefox-pgo thread

#403 2009-11-09 20:37:11

Re: Official firefox-pgo thread

#404 2009-11-11 06:37:19

Re: Official firefox-pgo thread

#405 2009-11-11 06:43:03

Re: Official firefox-pgo thread

#406 2009-11-11 07:27:01

Re: Official firefox-pgo thread

#407 2009-11-11 10:59:23

Re: Official firefox-pgo thread

#408 2009-11-11 11:11:03

Re: Official firefox-pgo thread

#409 2009-11-11 19:39:16

Re: Official firefox-pgo thread

#410 2009-11-14 18:53:03

Re: Official firefox-pgo thread

#411 2009-11-14 19:01:20

Re: Official firefox-pgo thread

#412 2009-11-14 22:25:28

Re: Official firefox-pgo thread

#413 2009-11-15 03:46:21

Re: Official firefox-pgo thread

#414 2009-11-15 10:43:43

Re: Official firefox-pgo thread

#415 2009-11-15 11:33:03

Re: Official firefox-pgo thread

#416 2009-11-15 11:45:40

Re: Official firefox-pgo thread

#417 2009-11-15 12:48:44

Re: Official firefox-pgo thread

#418 2009-11-15 12:52:19

Re: Official firefox-pgo thread

#419 2009-11-15 17:15:25

Re: Official firefox-pgo thread

#420 2009-11-15 17:34:22

Re: Official firefox-pgo thread

#421 2009-11-15 17:45:42

Re: Official firefox-pgo thread

#422 2009-11-15 22:04:57

Re: Official firefox-pgo thread

#423 2009-11-16 00:51:00

Re: Official firefox-pgo thread

#424 2009-11-16 01:23:57

Re: Official firefox-pgo thread

#425 2009-11-16 05:10:59

Re: Official firefox-pgo thread

Board footer