Pale Moon 15.3 beta 3 (public beta)

[Moonchild]

After testing the waters with a few internal builds and some testing by the beta testing team, I'm releasing a beta version of Pale Moon to the public. Mainly, because this involves a change in development environment and compiler, and such a switch is always tricky and can result in unstable programs.
For example: I've already had to implement some changes for the user interface since it broke trying to compile the current source with the new compiler.

Important notes about the beta:

• This build requires Windows Vista or later, due to toolset restrictions in Visual Studio 2012 and lack of documentation. The Final CTP has been used of Update 1, but there is no clear documentation yet to build XP compatible binaries from the command-line.
• Full optimizations have been used when building this, so performance should be (close to) release levels.
• Vectorization and parallelization of the code has been used. Modern processors should absolutely love this.
• This beta uses Incremental Garbage Collection for javascript by default. Swtich off with javascript.options.mem.gc_incremental in case of trouble
• This beta uses Azure for rendering content by default. Switch off with gfx.content.azure.enabled in case of trouble
• This is a 32-bit build, considering that is still the main focus for Pale Moon (use of 64-bit browsers is still fairly limited due to lack of support by plugin developers... as a result, only about 20% of Pale Moon users are using the 64-bit one). In addition, 64-bit system software, especially graphics drivers, tend to give false positives (an issue is found that is not Pale Moon's fault) and/or resulting bug reports that aren't valid for the actual Pale Moon code. Code development is therefore always done on the x86 platform, and 64-bit builds are only made of close to release-ready versions, which this beta 3 definitely is not. The 32-bit build will run equally fine on 32-bit and 64-bit operating systems.

Feedback on this forum thread is greatly appreciated.
Both positive and negative feedback, but please remain constructive

Download link for the beta: http://ge.tt/1mUiepQ?c

[Ryrynz]

Flick us a Change log MC? A brief one will be fine.

[Moonchild]

I just gave you one
Changes:
- Change in build environment/compiler
- Fixups for the change in build environment/compiler
- Changes as outlined above in the OP!

[Ryrynz]

Oh, I was expecting more..

I got a 2% increase in V8 and a 5% increase on Impact HTML 5 benchmark with no problems so far.
Other benchies are about the same except for one but I'll run that again.

[Moonchild]

Oh, I was expecting more..

Make no mistake, changing to a whole new major compiler version is very involved - it's like touching every single line of code.

I don't expect much difference in JS benchmarks, since that is already pretty much maximized in 15.2.1 - everything else in the browser should benefit, though.

[Ryrynz]

Oh, I was expecting more..

Make no mistake, changing to a whole new major compiler version is very involved - it's like touching every single line of code.

I remember you saying, I just thought something from 17 might make it in here as well.

I don't expect much difference in JS benchmarks, since that is already pretty much maximized in 15.2.1 - everything else in the browser should benefit, though.

Yeah I got that feeling early on from the results. Next year though.. I love the performance increase Ionmonkey will bring in 18, the speed improvement is quite tangible in my experience.

Okay, you might want to look at this, I retested that benchmark that was lower and there's a definite performance regression using Webvisbench on 15.3

Tested on a blank profile and my own with similar results, on a blank profile I score 4440 and with 15.2.1 with 15.3 I'm only getting 3630.

[Moonchild]

"Something from 17"? 17 won't have much to write home about in significant changes, so, very unlikely to be adding anything from it.
Ionmonkey will be the first significant change to come.

As for Webvizbench: Try switching off azure for content. Webvisbench is a horrible torture test for the rendering back-end and non-typical use of the rendering engine. Even if azure lowers the score there, it'll probably do better with normal use, but it'd be something to consider.

[Ryrynz]

Yeah Azure was the issue, strange considering a year ago they were touting 3x the performance improvements over Cairo with Webvisbench.
I'll re-enable Azure and suck up the benchmark loss, it's not a real world test anyway, I can live with that.

BTW I ran Firefox 17 with Azure enabled and got the same performance as Pale Moon with Azure disabled.. might be something in 17 after all..

[Moonchild]

Thanks for the feedback - I'll see if I can find the Mozilla bug dealing with this regression.

EDIT: Seems there's no less than 6 bugs involved. A fix of this performance regression breaks other things. It's complicated, as per usual for Azure/Cairo.

[Ryrynz]

Actually Firefox 17b4 has the same performance on Webvisbench with azure enabled as it does disabled so either azure is running better or azure isn't running at all..