[Request] zen-browser-bin

Ast3risk-ops commented 2 months ago

Package:

https://aur.archlinux.org/packages/zen-browser-bin

Purpose:

Zen Browser is basically Arc but Firefox, makes Firefox more modern, more beautiful, more private (includes some BetterFox patches), ~~and faster~~.

It has workspaces, themes, and vertical tabs for instance.

Benefits:

~~It's way faster than Firefox and more beautiful and customizable than Firefox.~~ Also more private through said BetterFox patches.

Building:

Run makepkg -si, which will download the latest prebuilt zen.linux-specific.tar.bz2 and extract the files to the right folders. Disables the builtin updater through a policies.json.

Copyright:

MPL-2.0

Expected Interest:

Many

Already available?

No

Unique request?

Yes

Banned package?

No

More information:

Website: https://zen-browser.app Github: https://github.com/zen-browser/desktop Stars: 10.1k

~~This version is optimized for newer AVX2 processors.~~

There is also a generic version for older machines, zen-browser-bin though my request centers on the AVX2 version.

xiota commented 2 months ago

I prefer avoiding -bin browser packages, but suppose it would be fine if different enough from other firefox forks.

Have you benchmarked the two versions to confirm that there is actually performance benefit from avx2?

Ast3risk-ops commented 2 months ago

I prefer avoiding -bin browser packages, but suppose it would be fine if different enough from other firefox forks.

Have you benchmarked the two versions to confirm that there is actually performance benefit from avx2?

Zen has done their own benchmarks.

https://docs.zen-browser.app/benchmarks

Basemark scores are 1912.77 for ff nightly and 1957 for the latest version of Zen.

Speedometer is 21.7 for Zen and 27.0 for Firefox nightly.

Speed graph between versions:

xiota commented 2 months ago

I mean avx2 vs standard build. Have to run benchmarks multiple times to get low, same error margins.

In my past testing for other forks, optimizing for instruction set doesn't make any significant difference for firefox and chromium based browsers. Most likely they already include multiple optimization targets in critical paths.

Ast3risk-ops commented 2 months ago

I mean avx2 vs standard build. Have to run benchmarks multiple times to get low, same error margins.

In my past testing for other forks, optimizing for instruction set doesn't make any significant difference for firefox and chromium based browsers. Most likely they already include multiple optimization targets in critical paths.

No I haven't checked.

Ast3risk-ops commented 2 months ago

I mean avx2 vs standard build. Have to run benchmarks multiple times to get low, same error margins.

In my past testing for other forks, optimizing for instruction set doesn't make any significant difference for firefox and chromium based browsers. Most likely they already include multiple optimization targets in critical paths.

On my flawed test environment:

Clean tar.bz2 generic: 10.5

Maybe not-so-clean outdated AUR package Optimized: 10

Expected score on official benchmarks for Optimized (different machine): 21.7

I did remove all extensions but I did have 3 small themes.

You'll probably want to test this yourself.

xiota commented 2 months ago

The ± errors are important. Suppose generic is 10.5 ± 0.25 and optimized is 10 ± 0.25. That would mean generic performs significantly better than optimized because the ranges do not overlap. But if the results were 10.5 ± 0.65 and 10 ± 0.65, there is overlap, so there would be "no significant difference".

When testing Firefox/Chromium forks with everything the same except microarchitecture optimization, the generic build often had same or slightly "better" results, but still not significantly different. If you check the instruction sets contained by the generic build (easiest is with intelxed), you'll find it already contains AVX/AVX2 instructions ⇒ they already include instruction set optimization in the build process.

So if the only reason you want this fork is for performance optimization, adding it wouldn't be worth the effort. But if it has different features, it could be worth adding. However, it's a -bin package, so users could easily make it themselves.

Ast3risk-ops commented 2 months ago

@xiota Yeah, but I personally notice a decent speed difference compared to normal firefox, and the extraction-then-compression process during install can take a while depending on the computer.

BinaryHarbinger commented 2 months ago

Hi there I'm just commenting to support the request flatpak version basically sucks and who wants to compile whole browser in their system (except that Gentoo guys)

Basically zen browser has lot of things to talk like vertical tabs and faster web engines. That is why we want it in Chaotic Aur

xiota commented 2 months ago

I'll run benchmarks between zen-browser-avx2-bin and zen-browser-bin. If I don't see consistent significant difference, will add zen-browser-bin.

xiota commented 2 months ago

Methodology:

Build and install package
Run browser, create default profile
Run Speedometer 2.1 at least twice. Repeat until error is less than 2.

Results: No significant difference between standard and avx2 versions of zen browser. Meanwhile, firefox from Arch repo and floorp 11.18.0 from Chaotic AUR repo are significantly faster.

firefox 121.0 — 86.6±1.8 (old)
firefox 122.0 — 87.9±1.9 (old)
firefox 130.0 — 91.3±1.6 (new)
floorp 11.8.1 — 70.4±1.1 (old)
floorp 11.18.0 — 87.3±1.7 (new)
zen-browser-bin 1.0.0.a.39 — 79.3±1.2 ⇒ 78.1-80.5
zen-browser-avx2-bin 1.0.0.a.39 — 80.8±1.4 ⇒ 79.4-82.2

Notes:

Using Speedometer 2.1 instead of 3.0 so results can be compared with previous runs on different browsers.
Using error cutoff of 2 because that is what I used in the past. Targeting smaller error required too many runs and usually did not change results. (For zen-browser results to be significant, error would have to be ±0.75.)
When ranges overlap, the "true" values could be within the overlapping region, so there is considered to be "no significant difference".

code-IM-perfect commented 1 month ago

Meanwhile, firefox from Arch repo and floorp 11.18.0 from Chaotic AUR repo are significantly faster.

So according to benchmarks, zen is slower than the stock firefox? Cuz it did feel snappier to use...may have been a placebo

(Sorry for replying on a closed issue but I really needed to know)

Ast3risk-ops commented 1 month ago

Meanwhile, firefox from Arch repo and floorp 11.18.0 from Chaotic AUR repo are significantly faster.

So according to benchmarks, zen is slower than the stock firefox? Cuz it did feel snappier to use...may have been a placebo

(Sorry for replying on a closed issue but I really needed to know)

The current website benchmarks are severely out of date (as are xiota's benchmarks), new versions have come out.

Zen also has a different (better) UI as well as BetterFox patches (and it's not based on Firefox ESR like Floorp which is stale as shit)

xiota commented 1 month ago

The request was approved, with adjustment, based on the alternate interface, not potential performance difference. Although rendering and script execution may not be faster, interface differences can make the overall experience feel better.

I'm not rerunning benchmarks. While newer AVX2 versions could be faster than previous versions, so would newer non-AVX versions and newer Firefox versions. Even if future benchmarks do show significant differences, I would not expect practical difference. (The differences that matter are between different builds of the same browser.)

chaotic-aur / packages