Firefox 4 released [LWN.net]

Firefox 4 released

Posted Mar 22, 2011 15:26 UTC (Tue) by erwbgy (subscriber, #4104) [Link]

It definitely feels faster, much more like using Chrome or Konqueror than previous versions.

Firefox 4 released

Posted Mar 22, 2011 15:28 UTC (Tue) by xav (guest, #18536) [Link] (7 responses)

Just tried the update on a CentOS 5.3, and now it fails because of a missing 'GLIBCXX_3_4_9' version in libstdc++ ...
How do I get back ?

Firefox 4 released

Posted Mar 22, 2011 15:35 UTC (Tue) by bjacob (guest, #58566) [Link]

CentOS is still based on GCC 4.1:
http://distrowatch.com/table.php?distribution=centos

Mozilla-built Firefox 4 binaries do no longer support GCC 4.1-based distros.

However, if you used a package provided by CentOS itself, then you should definitely report a bug to them (not to Mozilla).

Firefox 4 released

Posted Mar 22, 2011 15:43 UTC (Tue) by dowdle (subscriber, #659) [Link] (3 responses)

Like with so many thing (take LibreOffice for example)... most distros won't have packages until their next major release. Yes you can download the tar.gz from the official sites but they probably aren't going to work well on every distro.

Firefox 4 released

Posted Mar 22, 2011 19:35 UTC (Tue) by armijn (subscriber, #3653) [Link] (2 responses)

Or worse. I tried FF 4 on Fedora 14 using the Mozilla binary build, with my default profile. Now my FF 3.6.15 toolbar has been rearranged and searching Google the way I usually do it (open a tab, go to Google, search) now gives me a blank page after I hit 'search'. Annoying.

Firefox 4 released

Posted Mar 23, 2011 4:41 UTC (Wed) by yhager (guest, #50165) [Link] (1 responses)

It is usually advised to create a new profile for experimenting with new versions of Firefox. Only when you are certain you are not planning to go back, use your normal profile.

Firefox 4 released

Posted Mar 23, 2011 10:50 UTC (Wed) by ebirdie (guest, #512) [Link]

Or keep backups of profiles before jumping to new versions. I had no problems with profiles during FF4 development phase. I started using FF4b2 and since then used 3.6 and 3.5 in parallel. Main usage of FF4 has been on OS X 10.6.x, however, but copied my profile to Debian Lenny and later Squeeze desktop occationally. Can't live long without my up-to-date profile. Havent experienced nothing like the above with FF.

My take on the cause of the problem are: plugins (FF4 invaliated many plugins, which was reason for my jumping between FF3 and FF4), graphics toolbox.

Firefox 4 released

Posted Mar 22, 2011 15:44 UTC (Tue) by lkundrak (subscriber, #43452) [Link] (1 responses)

A quick fix (I'm using with binary Mozilla builds on our CentOS5-based selenium browser testing farm) is just to get the file from later libstdc++ package and place it in directory with firefox-bin. Get it e.g. from here: http://kojipkgs.fedoraproject.org/packages/gcc/4.3.0/0.2/...

Firefox 4 released

Posted Mar 22, 2011 15:54 UTC (Tue) by xav (guest, #18536) [Link]

Great ! I rpm -Uvh'ed your link, and here I am with a brand new FF4.
Thanks, that greatly helped.

Xav

64-bit

Posted Mar 22, 2011 16:12 UTC (Tue) by marduk (subscriber, #3831) [Link] (18 responses)

Are there any 64-bit builds?

64-bit

Posted Mar 22, 2011 16:24 UTC (Tue) by angdraug (subscriber, #7487) [Link] (10 responses)

I'm using a 64-bit Debian package since 4.0beta12 (rebranded as Iceweasel, of course), no complaints here. Couldn't enable hardware acceleration, but even without that it's a lot faster than 3.6.

64-bit

Posted Mar 22, 2011 16:27 UTC (Tue) by lkundrak (subscriber, #43452) [Link] (9 responses)

Is there anything beyond WebGL where is Firefox capable of using any sort of hardware acceleration?

Hardware acceleration on Linux

Posted Mar 22, 2011 16:38 UTC (Tue) by bjacob (guest, #58566) [Link] (8 responses)

Firefox can use your GPU for 3 things: WebGL, content acceleration, compositing acceleration.

WebGL is on by default everywhere, but it obviously depends on your OpenGL driver, see below (*).

Compositing acceleration using OpenGL is implemented, but not yet enabled by default on X11, mostly because a bit of code was still missing: using texture-from-pixmap to avoid a round-trip to the X client and avoid losing the benefit of XRender (see below). This is now being worked on, see https://bugzilla.mozilla.org/show_bug.cgi?id=640082 . Should be in Firefox 5 in a few months. There also is a crash bug when using the Flash plugin. If you still want to use compositing acceleration today, just go to about:config and set layers.acceleration.force-enabled. This of course also depends on your OpenGL driver (*).

Content acceleration is currently using XRender, like in Firefox 3.6. There are plans to move to something based on OpenGL, but that hasn't happened yet (very big project). The Windows version is already using Direct2D to great effect.

(*) OpenGL drivers on X11: In Firefox 4, unfortunately, only the NVIDIA driver is whitelisted. This is mostly because of difficulty in safely getting driver version info. See https://bugzilla.mozilla.org/show_bug.cgi?id=639842 : that will be in Firefox 5 (just a few months away) and will allow to whitelist many recent drivers (for example, recent Intel i965, probably some recent radeon and fglrx, etc). Meanwhile, if you want to try, you can un-block your driver by running with the MOZ_GLX_IGNORE_BLACKLIST environment variable.

Hardware acceleration on Linux

Posted Mar 22, 2011 17:10 UTC (Tue) by lkundrak (subscriber, #43452) [Link] (7 responses)

Thanks for the extensive answer. Does content acceleration with OpenGL/Direct2D actually make anything faster? The other day I've read a blog entry that concluded that simply putting cairo on top of opengl is actually slower for most configurations [1]; is Firefox's content acceleration comparable to that?

[1] http://blogs.gnome.org/otte/2010/06/26/fun-with-benchmarks/

Hardware acceleration on Linux

Posted Mar 22, 2011 17:36 UTC (Tue) by bjacob (guest, #58566) [Link] (6 responses)

First of all, there is no such thing as 'just putting Cairo on top of OpenGL': implementing a 2D API such as Cairo on top of a low-level API like OpenGL is a very, very complex task. There is a 'Cairo OpenGL back-end' but I have no idea how good it is. Firefox is not using it.

What Firefox 4 does on Windows, is that it implements a Direct2D back-end for Cairo. Direct2D is a relatively efficient 2D graphics library that uses Direct3D 10 to talk to the GPU. Since it's already a 2D library, layering Cairo on top of it is much simpler.

To answer your main question: yes, given good hardware and drivers, that is already a very large performance boost over what can be done in software. See all the canvas/2D benchmarks currently floating around the web.

However there is still room for improvement: the Cairo API is inherently not very well suited to hardware acceleration. The (long term) plan is to eventually get rid of it by implementing our own 2D graphics library, that would use OpenGL or Direct3D 10/9 depending on your platform.

Hardware acceleration on Linux

Posted Mar 22, 2011 19:01 UTC (Tue) by amaranth (subscriber, #57456) [Link] (2 responses)

But if you're using XRender right now you're already getting hardware acceleration. The open source drivers, at least, accelerate this fairly well.

Hardware acceleration on Linux

Posted Mar 23, 2011 1:08 UTC (Wed) by joedrew (guest, #828) [Link]

(Disclaimer: I work with bjacob at Mozilla)

XRender is not a very good API to hardware accelerate. Our new content acceleration layer will be much more like Direct2D, which is a good 2D API, and more amenable to hardware acceleration.

Hardware acceleration on Linux

Posted Mar 23, 2011 6:59 UTC (Wed) by elanthis (guest, #6227) [Link]

That is much like saying that if you're using the core X protocol, you're already getting accelerated line rendering.

RENDER was designed as the "new rendering API" about 11 years ago. 11 years. Over a decade ago. The first programmable GPU didn't arrive on the consumer market until 2001, a year later.

RENDER is no longer a particularly relevant API, and is just as in need of retiring as the old X core protocol. It can do only a teeny little fraction of what Cairo (and its client applications) need out of a low-level rendering API.

Hence all the interest in Wayland lately, which basically does away with all rendering APIs in the windowing system so that we stop putting ourselves in this endless cycle of deprecation. We have the low-level rendering API support and the client-level rendering API support, and it's dumb that we keep having to invent entirely different middle-layers like X, RENDER, GLX, and so on to connect applications to the hardware, especially when those middle layers keep becoming obsolete every six months.

Hardware acceleration on Linux

Posted Mar 22, 2011 21:40 UTC (Tue) by endecotp (guest, #36428) [Link]

> the Cairo API is inherently not very well suited to hardware
> acceleration. The (long term) plan is to eventually get rid
> of it by implementing our own 2D graphics library, that would
> use OpenGL or Direct3D 10/9 depending on your platform.

Presumably OpenVG could be applied here, when it's available.

Hardware acceleration on Linux

Posted Mar 23, 2011 6:44 UTC (Wed) by elanthis (guest, #6227) [Link] (1 responses)

Why is this Direct2D backend not in Cairo itself instead of in Firefox?

(Apologies if it is or soon will be; I just haven't seen anything in Cairo's git repos the last I looked, which admittedly was not recently.)

Hardware acceleration on Linux

Posted Mar 23, 2011 11:27 UTC (Wed) by bjacob (guest, #58566) [Link]

Actually this is in Cairo 'itself', except that it's in Mozilla's copy of Cairo instead of being in the official upstream Cairo. They are synced from time to time to minimize the diff, so if upstream Cairo is interested, I guess the Direct2D back-end will find its way into it. Anyways, this has always been developed in the mozilla-central repository:

http://hg.mozilla.org/mozilla-central/file/tip/gfx/cairo/...

(see e.g. cairo-win32.h)

64-bit

Posted Mar 22, 2011 16:28 UTC (Tue) by bjacob (guest, #58566) [Link] (6 responses)

On Linux, 64bit and 32bit builds are supported on an equal footing.

64-bit

Posted Mar 22, 2011 16:42 UTC (Tue) by marduk (subscriber, #3831) [Link] (3 responses)

I meant the where could I actually download the 64-bit builds. When I go to firefox.com to download 4.0, I only see options for the 32-bit versions. But nm it looks like my distro has already caught up and provides a 64-bit build.

64-bit

Posted Mar 22, 2011 16:55 UTC (Tue) by bjacob (guest, #58566) [Link] (2 responses)

It's available on FTP:
ftp://ftp.mozilla.org/pub/mozilla.org/firefox/releases/4.0/

Not sure why the official download link points to the 32bit version. For sure, all the continuous testing is enabled on Linux 64bit:
http://tbpl.mozilla.org/

64-bit

Posted Mar 22, 2011 19:24 UTC (Tue) by xxiao (guest, #9631) [Link] (1 responses)

ftp://ftp.mozilla.org/pub/mozilla.org/firefox/releases/4....

550 permission denied?

64-bit

Posted Mar 22, 2011 20:28 UTC (Tue) by Guhvanoh (subscriber, #4449) [Link]

Try the same link using http instead...

64-bit

Posted Mar 23, 2011 1:09 UTC (Wed) by joedrew (guest, #828) [Link] (1 responses)

Actually we (apparently) don't officially support Linux 64. (This was news to me too. :) )

64-bit

Posted Mar 23, 2011 11:31 UTC (Wed) by bjacob (guest, #58566) [Link]

I would argue that's a bug that we need to fix. 64bit linux distros are shipping 64bit firefox, so we really should support it 'officially'. Thankfully, 64bit firefox/linux seems to be just as well tested as 32bit is.

Firefox 4 released

Posted Mar 22, 2011 16:12 UTC (Tue) by xxiao (guest, #9631) [Link] (3 responses)

on windows, FF4 crashed and could never started even once, had to use chrome to download another FF3.6, will stay with FF3 for now.

many FF4 features seemed similar to what chrome already had though.

Firefox 4 released

Posted Mar 22, 2011 16:44 UTC (Tue) by bjacob (guest, #58566) [Link] (1 responses)

Did you try Firefox 4 final, or some old beta?

If final, then please at least go to about:crashes and paste here your crash report links, so we can have a look. https://crash-stats.mozilla.com/products/Firefox is suggesting that Firefox 4 is a rather non-crashy release, but YMMV.

Firefox 4 released

Posted Mar 22, 2011 18:02 UTC (Tue) by xxiao (guest, #9631) [Link]

i believe it's the final version, md5sum is: 417e9bc84...
yes i submitted a crash report already.

Firefox 4 released

Posted Mar 23, 2011 18:04 UTC (Wed) by MortenSickel (subscriber, #3238) [Link]

Runs great under windows here, so you obiously were unlucky with something.