Official firefox-pgo thread

toxygen · 2009-09-29 21:50:25

ok, it is what i thought. you have to edit PKGBUILD on your build directory, comment out the last md5sum as such:
================

source=(
#ftp://ftp.mozilla.org/pub/mozilla.org/firefox/releases/namoroka/alpha1/source/firefox-"$pkgver".source.tar.bz2
ftp://ftp.mozilla.org/pub/mozilla.org/firefox/nightly/"$pkgver"-candidates/build2/source/firefox-"$pkgver".source.tar.bz2
mozconfig
100-system-hunspell-corrections.patch
firefox.desktop
firefox-safe.desktop
ldflags.patch
ldflags2.patch
# x64config.patch (this was commented out in the PKGBUILD originally)
)
md5sums=('3bda192831af077dac080959c08e26e6'
'f12db64cedf53e5b105381a8588bfaae'
'5efd6772ed0ecf8eddec5d5650191d3c'
'4b766d6d21e1168dd365f467e4275e13'
'1034e29d85c2d22faeb5899732321559'
'9e461e5fc73b241d68d08be908405a5e'
'9587e0e6da42c883e34a46ee35cc3b7d')
# '96615d45f14540a9d9f3c0bcb7c286dd') (this has to be added as well as paranthesis on above line)

==========
notice the paranthesis has to be added to the previous line. or you can remove both lines on the PKGBUILD since the x64config.patch file is not needed (i think that was blasse's patch to add TM)

Last edited by toxygen (2009-09-29 21:52:47)

blasse · 2009-09-30 05:30:05

Ok, I see it now. I will correct pkgbuild. x64config.patch was ment to enable TM on x86_64, however it does not work, because in 3.6a2 there is no source for it

ammon · 2009-09-30 11:14:10

@toxygen

Yep, that fixes it. Thanks.

@blasse

Why to firefox-pgo-beta to conflict with firefox-pgo?

Last edited by ammon (2009-09-30 11:16:27)

blasse · 2009-09-30 14:01:37

Because of firefox.desktop files and bin files. The can be manually renamed and pkgs will no longer conflict

toxygen · 2009-10-09 22:43:52

Hey blasse (or anyone) have you been able to build 3.7a1 with pgo? I havent had any success since you first put the PKGBUILD in aur. I can build it successfully by commenting out the pgo build portion, and it works really well, but pgo build always fails with "no rule to make clobber-all", have you run into problems?
and does it make a difference to have gcc-snapshot?

just rebuilt 3.7a1 today (after the gcc-snapshot upgrade), and the sunspider test seems to have improved slightly (~80ms improvement on the test i ran)

bangkok_manouel · 2009-10-10 06:00:18

toxygen wrote:

Hey blasse (or anyone) have you been able to build 3.7a1 with pgo? I havent had any success since you first put the PKGBUILD in aur. I can build it successfully by commenting out the pgo build portion, and it works really well, but pgo build always fails with "no rule to make clobber-all", have you run into problems?

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

blasse · 2009-10-18 15:53:26

Bug for enabling TM on x86_64 is marked as resolved fixed. Hovewer code has not been merged into upcoming 3.6b1. So TM on x86_64 still only in FF 3.7

blasse · 2009-10-29 17:56:30

Links on first page updated.

jleach · 2009-10-30 12:18:50

As noted on the AUR page, build3 is now available.

To get it to build I just removed 'clobber_all' from the line 229 of mozilla-1.9.2/Makefile.in

-maybe_clobber_profiledbuild: clobber_all
+maybe_clobber_profiledbuild:

a line of

sed -i 's/clobber_all//' Makefile.in || return 1

inserted near the top of build() of PKGBUILD accomplishes this.

blasse · 2009-10-30 17:27:22

Added to pkgbuild, seems to work fine now - thx!

blasse · 2009-10-30 18:34:44

Ted Mielcarek wrote, that removing clobber_all isn't proper way for this. I don't know if I understand him right, but he suggested, that this may in fact make pgo doesn't work...

toxygen · 2009-10-30 20:44:41

verified the new line works on 3.7a1, pgo build finaly worked!

Last edited by toxygen (2009-10-30 20:50:21)

blasse · 2009-10-30 21:35:52

toxygen wrote:

verified the new line works on 3.7a1, pgo build finaly worked!

Be warned, Ted is one of mozillas developers responsible for PGO:

Ted Mielczarek wrote:

That's not a good fix. I don't see how a PGO build could possibly work with
that. With GCC, you need to recompile on each pass, which is why the clobber is
there.

flamelab · 2009-10-30 23:42:38

Ι managed to build firefox-pgo-beta (3.6x) without !makeflags and make -j1 !

harryNID · 2009-10-31 02:10:24

Hi all,

Not much help here but I thought you might like the input.

I manage to finally build the new version as well but after trying it I went back to firefox-pgo-beta 3.6a2-1.

Here's why,

First, I noticed everything was sluggish. I checked and 3.6b1-2 was using almost 90M of memory. Next, after double-checking to see if jit was enabled (it was) I ran a Sunspider test where it would repeatedly crash before completion. I also noticed browsing was considerably slower so I think blasse's warning is valid. Something is not right there.

Has anybody else had these issues? If not, I will recompile and try again.

P.S.
After reinstalling 3.6a2-1 I was again able to do Sunspider and scored a respectable 1833 on a P4 machine.

toxygen · 2009-10-31 03:02:04

blasse wrote:

Ted Mielczarek wrote:
That's not a good fix. I don't see how a PGO build could possibly work with
that. With GCC, you need to recompile on each pass, which is why the clobber is
there.

That's very interesting. I did let the compile go without checking, and when i came back it was done, so I dont know if the pgo testing failed or anything. i'm recompiling again to see if any errors come out, but so far i havent had issues. did the sunspider test, still getting similar results as before, around 950ms now (did the test before compiling but didnt bookmark it). havent done the html 5 tests yet, will wait for re-compiling. i'll keep an eye out on the bugzilla report though, see if any updated info comes down the line.

[edit] after recompiling, it looks like all went well with the compilation, however i dont think (as ted pointed out) that a second recompilation actually was made. the browser test did come up, after the first compile, but afterwards i only saw a few python scripts run, and the final "make install", so i think this "fix" didnt do anything other than skip the "make clobber-all" step (same as commenting out the pgo-build part of the PKGBUILD.

Last edited by toxygen (2009-10-31 04:38:06)

pogeymanz · 2009-11-03 00:33:57

Hey guys,

I'm trying to build Firefox-3.5.4 w/ PGO, but not from AUR. I'd like to try to do it myself. But I get this error:

/usr/bin/make export_tier_testharness
make[4]: Entering directory `/var/yaourt/yaourtbuild/firefox/src/mozilla-1.9.1'
export_tier_testharness
make[4]: Leaving directory `/var/yaourt/yaourtbuild/firefox/src/mozilla-1.9.1'
/usr/bin/make libs_tier_testharness
make[4]: Entering directory `/var/yaourt/yaourtbuild/firefox/src/mozilla-1.9.1'
libs_tier_testharness
make[4]: Leaving directory `/var/yaourt/yaourtbuild/firefox/src/mozilla-1.9.1'
make[3]: Leaving directory `/var/yaourt/yaourtbuild/firefox/src/mozilla-1.9.1'
make[2]: Leaving directory `/var/yaourt/yaourtbuild/firefox/src/mozilla-1.9.1'
make[1]: Leaving directory `/var/yaourt/yaourtbuild/firefox/src/mozilla-1.9.1'
OBJDIR=/var/yaourt/yaourtbuild/firefox/src/mozilla-1.9.1 python /_profile/pgo/profileserver.py
python: can't open file '/_profile/pgo/profileserver.py': [Errno 2] No such file or directory
make: *** [profiledbuild] Error 2
==> ERROR: Build Failed.
    Aborting...

Here is my PKGBUILD:

# $Id: PKGBUILD 56839 2009-10-28 10:25:01Z jgc $
# Contributor: Jakub Schmidtke <sjakub@gmail.com>

pkgname=firefox
pkgver=3.5.4
pkgrel=1
_xulver=1.9.1.4
pkgdesc="Standalone web browser from mozilla.org"
arch=(i686 x86_64)
license=('MPL' 'GPL' 'LGPL')
depends=("xulrunner=${_xulver}" 'desktop-file-utils')
makedepends=('zip' 'pkgconfig' 'diffutils' 'libgnomeui>=2.24.1' 'python')
replaces=('firefox3')
install=firefox.install
url="http://www.mozilla.org/projects/firefox"
source=(http://releases.mozilla.org/pub/mozilla.org/firefox/releases/${pkgver}/source/firefox-${pkgver}.source.tar.bz2
        mozconfig
        firefox.desktop
        firefox-safe.desktop
        mozilla-firefox-1.0-lang.patch
        browser-defaulturls.patch
        firefox-version.patch)

build() {
  cd "${srcdir}/mozilla-1.9.1"
  patch -Np1 -i "${srcdir}/mozilla-firefox-1.0-lang.patch" || return 1
  patch -Np0 -i "${srcdir}/browser-defaulturls.patch" || return 1
  patch -Np1 -i "${srcdir}/firefox-version.patch" || return 1

  cp "${srcdir}/mozconfig" .mozconfig
  unset CFLAGS
  unset CXXFLAGS

  export LDFLAGS="-Wl,-rpath,/usr/lib/firefox-3.5"

  #export CFLAGS="-march=native"
  #export CXXFLAGS="-march=native"
  #export LDFLAGS="-Wl,-rpath,/usr/lib/firefox-${pkgver} -Wl,-O1,--sort-common,--hash-style=gnu,--as-needed"


  make -j1 -f client.mk profiledbuild MOZ_MAKE_FLAGS="${MAKEFLAGS}" || return 1
  make -j1 DESTDIR="${pkgdir}" install || return 1

  rm -f ${pkgdir}/usr/lib/firefox-3.5/libjemalloc.so

  install -m755 -d ${pkgdir}/usr/share/applications
  install -m755 -d ${pkgdir}/usr/share/pixmaps
  install -m644 ${srcdir}/mozilla-1.9.1/browser/branding/unofficial/default48.png ${pkgdir}/usr/share/pixmaps/firefox.png || return 1
  install -m644 ${srcdir}/firefox.desktop ${pkgdir}/usr/share/applications/ || return 1
  install -m644 ${srcdir}/firefox-safe.desktop ${pkgdir}/usr/share/applications/ || return 1
}
md5sums=('0b6ccb1e50d96b7127a18a69399fcf05'
         'a73265d2de5c3f7ade37396e81e71ce8'
         '68cf02788491c6e846729b2f2913bf79'
         '5e68cabfcf3c021806b326f664ac505e'
         'bd5db57c23c72a02a489592644f18995'
         '346d74ec560e7bbf453c02ff21f4b868'
         '1dd9a10df0b9e4cf332eadc326d78e07')

And my mozconfig:

. $topsrcdir/browser/config/mozconfig

ac_add_options --prefix=/usr
ac_add_options --libdir=/usr/lib
ac_add_options --with-system-nspr
ac_add_options --with-system-nss
ac_add_options --with-system-jpeg
ac_add_options --with-system-zlib
ac_add_options --with-system-png
ac_add_options --with-pthreads
ac_add_options --disable-tests
ac_add_options --disable-debug
ac_add_options --enable-optimize
ac_add_options --disable-installer
ac_add_options --enable-xinerama
ac_add_options --enable-default-toolkit=cairo-gtk2
ac_add_options --disable-xprint
ac_add_options --enable-strip
ac_add_options --disable-pango
ac_add_options --enable-system-cairo
ac_add_options --enable-svg
ac_add_options --enable-canvas
ac_add_options --enable-startup-notification
ac_add_options --enable-libxul
ac_add_options --disable-crashreporter
ac_add_options --with-libxul-sdk=/usr/lib/xulrunner-devel-1.9.1
ac_add_options --disable-safe-browsing
ac_add_options --with-branding=browser/branding/unofficial

mk_add_options PROFILE_GEN_SCRIPT='$(PYTHON) $(MOZ_OBJDIR)/_profile/pgo/profileserver.py'

export BUILD_OFFICIAL=1
export MOZILLA_OFFICIAL=1
export USE_SHORT_LIBNAME=1
mk_add_options BUILD_OFFICIAL=1
mk_add_options MOZILLA_OFFICIAL=1
mk_add_options USE_SHORT_LIBNAME=1

Last edited by pogeymanz (2009-11-03 00:46:10)

toxygen · 2009-11-03 03:58:01

comparing your PKGBUILD with the one on the aur package, you only seem to have one make command;

  make -j1 -f client.mk profiledbuild MOZ_MAKE_FLAGS="${MAKEFLAGS}" || return 1
  make -j1 DESTDIR="${pkgdir}" install || return 1

the aur pgo packages shows 3 make commands (if you use x86-64, for 32bit i think it uses only the two builds).

  make -f client.mk MOZ_MAKE_FLAGS="${MAKEFLAGS}" || return 1
  make -f client.mk profiledbuild MOZ_MAKE_FLAGS="${MAKEFLAGS}" || return 1
  make -j1 DESTDIR=${pkgdir} -C ff-pgo install || return 1

the error indicates that /_profile/pgo/profileserver.py is not found, can you find it afterwards in the mozillia-1.9.1 folder/

other than that, i'll wait for more experienced people to answer

blasse · 2009-11-03 06:00:47

One make is just enough. My is just for compatibility with x86_64
Your pkgbuild seems fine (it's just a copy of official one, isn't it?) - double check if the profilserver.py is in source.
However, you're trying to build pgoed GUI, which doesn't make sens... When you are using xulrunner pkg, xulrunner is used as engine, while firefox pkg plays only GUI role. So to gain some speed from pgo, you should compile xulrunner-pgo, which isn't possible, because xulrunner has no GUI on his own. And that is way my pkgbuild does not use xulrunner

pogeymanz · 2009-11-03 13:41:02

blasse wrote:

One make is just enough. My is just for compatibility with x86_64
Your pkgbuild seems fine (it's just a copy of official one, isn't it?) - double check if the profilserver.py is in source.
However, you're trying to build pgoed GUI, which doesn't make sens... When you are using xulrunner pkg, xulrunner is used as engine, while firefox pkg plays only GUI role. So to gain some speed from pgo, you should compile xulrunner-pgo, which isn't possible, because xulrunner has no GUI on his own. And that is way my pkgbuild does not use xulrunner

I see! So why are we able to not use xulrunner? You would think that would be an important piece...

It's weird, because the profileserver.py is in there and it appears to be where it's supposed to be... Oh well, I gave up and used the AUR package, which worked very well.

blasse · 2009-11-03 17:16:02

Xulrunner is the "engine" for all mozilla apps. That's why most linux distribution uses it as standalone pkg. When using it, we can build on top of it firefox, thunderbird, songbird or every other mozilla program. Because building on top of xulrunner is close to building gui-only, firefox from extra repo is just 1mb package. It should be possible to compile firefox with pgo on top of xulrunner, but it optimize only the gui part. Applying pgo to xulrunner is not possible at this moment, because pgo need to start gui, and xulrunner itself doesn't have one. However, ubuntu devs are working to resolve this, so there is light in tunnel
My firefox-pgo is just old fashioned build, which integrates xul into firefox (just as it was in firefox2, before standalone xulrunner release), and so it makes possible to optimize firefox as whole, engine and gui.

pogeymanz · 2009-11-04 01:24:03

Ah, thank you very much for the explanation. That clears up quite a bit.

Pank · 2009-11-04 12:39:12

Is 3.6b1 unstable as hell for you guys too? It crashes constantly on my system! I have tried with a new profile so it must be something else. Normally Mozilla-betas are pretty solid but it doesn't seem to be the case here. . .

Thanks,
Rasmus

PS: Blasse, the build for 3.6b1 seems pretty good. It compiled the first time (unlike 3.5 which I have had quite some trouble with)

blasse · 2009-11-04 14:13:51

It is rock-stable for me. Maybe try recompiling it?

toxygen · 2009-11-04 15:55:26

i havent used 3.6 after 3.7a1pre came out, but i didn't have any crashes for the week or two i used it. so far 3.7 has been stable for me as well

Arch Linux

#376 2009-09-29 21:50:25

Re: Official firefox-pgo thread

#377 2009-09-30 05:30:05

Re: Official firefox-pgo thread

#378 2009-09-30 11:14:10

Re: Official firefox-pgo thread

#379 2009-09-30 14:01:37

Re: Official firefox-pgo thread

#380 2009-10-09 22:43:52

Re: Official firefox-pgo thread

#381 2009-10-10 06:00:18

Re: Official firefox-pgo thread

#382 2009-10-18 15:53:26

Re: Official firefox-pgo thread

#383 2009-10-29 17:56:30

Re: Official firefox-pgo thread

#384 2009-10-30 12:18:50

Re: Official firefox-pgo thread

#385 2009-10-30 17:27:22

Re: Official firefox-pgo thread

#386 2009-10-30 18:34:44

Re: Official firefox-pgo thread

#387 2009-10-30 20:44:41

Re: Official firefox-pgo thread

#388 2009-10-30 21:35:52

Re: Official firefox-pgo thread

#389 2009-10-30 23:42:38

Re: Official firefox-pgo thread

#390 2009-10-31 02:10:24

Re: Official firefox-pgo thread

#391 2009-10-31 03:02:04

Re: Official firefox-pgo thread

#392 2009-11-03 00:33:57

Re: Official firefox-pgo thread

#393 2009-11-03 03:58:01

Re: Official firefox-pgo thread

#394 2009-11-03 06:00:47

Re: Official firefox-pgo thread

#395 2009-11-03 13:41:02

Re: Official firefox-pgo thread

#396 2009-11-03 17:16:02

Re: Official firefox-pgo thread

#397 2009-11-04 01:24:03

Re: Official firefox-pgo thread

#398 2009-11-04 12:39:12

Re: Official firefox-pgo thread

#399 2009-11-04 14:13:51

Re: Official firefox-pgo thread

#400 2009-11-04 15:55:26

Re: Official firefox-pgo thread

Board footer