[SOLVED] Note on FF-79.0

3rensho · 07-27-2020, 07:13 AM

Just built FF-79.0 and found that one needs to apply the patches/workarounds that Ponce researched and found when I tried to build Thunderbird-78.x due to our version of rust. Details here - https://www.linuxquestions.org/quest...-a-4175678965/
Then it builds and runs fine.

ponce · 07-27-2020, 07:20 AM

I haven't tried yet but that shouldn't actually be needed anymore if you use the latest SlackBuild in the current tree, because Pat applied an alternate patch to fix building with the newer rust.

3rensho · 07-27-2020, 07:27 AM

Ach so. Thanks for the heads-up. Some day I'll learn to look at the new stuff before fixing my old file.

Edit: Builds fine with the current SlackBuild. Lesson leaned.

TheRealGrogan · 07-30-2020, 03:32 PM

I only had to do:

export CARGO_PROFILE_RELEASE_LTO=true

in rust.mk to get Firefox 79.0 (gkrust) to build with Rust 1.45.

This is something that made it into the ESR, but not the 79.0 release.

Code:

+# Pass -Clto for older versions of rust, and CARGO_PROFILE_RELEASE_LTO=true
+# for newer ones that support it. Combining the latter with -Clto works, so
+# set both everywhere.
 cargo_rustc_flags += -C lto
+export CARGO_PROFILE_RELEASE_LTO=true

ponce · 07-30-2020, 03:38 PM

here it seems to build fine with the SlackBuild in current without any modification (also 80.0b1 does).

LuckyCyborg · 07-31-2020, 05:28 PM

Quote:

Originally Posted by TheRealGrogan

I only had to do:

export CARGO_PROFILE_RELEASE_LTO=true

in rust.mk to get Firefox 79.0 (gkrust) to build with Rust 1.45.

This is something that made it into the ESR, but not the 79.0 release.

Code:

+# Pass -Clto for older versions of rust, and CARGO_PROFILE_RELEASE_LTO=true
+# for newer ones that support it. Combining the latter with -Clto works, so
+# set both everywhere.
 cargo_rustc_flags += -C lto
+export CARGO_PROFILE_RELEASE_LTO=true

Permit me to disagree.

In fact, looks like that setting that CARGO_PROFILE_RELEASE_LTO introduces at least a nasty but obscure issue, which I observed with some of video streams of local TV channels. It manifests by a high load of CPU and the playing experience is bad, like frames are dropped. True, it is also probably about low power CPUs as what I have in my boxes.

Too bad if that was inserted into ESR releases, and tells a lot about how stable are them.

Apparently, at least according with Mozilla, the right way to fix that build issue is to set "-Cembed-bitcode=yes" when it is used "-Clto" for rustc, like in the patch for there:

https://hg.mozilla.org/releases/mozi...v/5049f067f189

I've backported it manually to firefox-79.0 and I've managed to build it fine. Bonus, my video streams works fine again.

TheRealGrogan · 07-31-2020, 10:01 PM

Oh well, no issues here. At the time I did it, that was the fix for the build problem. I did wonder why they weren't just setting the -Cembed-bitcode flag the error was complaining about along with -Clto, but assumed they did it with the CARGO_PROFILE_RELEASE_LTO variable instead because it didn't cause older versions of rust to barf.

I did note an extra long build time, which makes sense if crates were having to be rebuilt more times than usual.

LuckyCyborg · 07-31-2020, 10:13 PM

Quote:

Originally Posted by TheRealGrogan

Oh well, no issues here. At the time I did it, that was the fix for the build problem. I did wonder why they weren't just setting the -Cembed-bitcode flag the error was complaining about along with -Clto, but assumed they did it with the CARGO_PROFILE_RELEASE_LTO variable instead because it didn't cause older versions of rust to barf.

I did note an extra long build time, which makes sense if crates were having to be rebuilt more times than usual.

It is not about building, but about how it is working. Apparently, that CARGO_PROFILE_RELEASE_LTO introduces obscure issues and slowdowns.

Of course, I understand well that in your All American HTPC with 64 cores / 128 threads living on liquid nitrogen freezing, no matters those issues...

However, I for one, I use 45W CPUs.

And, BTW... I observed those strange slowdowns first on the ESRs shipped now by -current.

TheRealGrogan · 08-01-2020, 01:45 AM

As I said, there are no issues here. I do mean running, on an older quad Corei7 CPU. This browser is working the same as, or better in performance, video performance and browserbench.com benchmarks (which I do with every build). I would notice performance regressions. I build it my own way though, using optimizations that work best for me.

On the next build, I'll use -Cembed-bitcode=yes instead though, now that I know about it.

Code:

. $topsrcdir/browser/config/mozconfig
 export CFLAGS="-mtune=nehalem"
 export CXXFLAGS="-mtune=nehalem"
 export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/opt/pulseaudio/lib"
 export PKG_CONFIG_PATH="$PKG_CONFIG_PATH:/opt/pulseaudio/lib/pkgconfig"
 export CPPFLAGS="-I/opt/pulseaudio/include"
 export LDFLAGS="-L/opt/pulseaudio/lib"
mk_add_options MOZ_MAKE_FLAGS=-j10
ac_add_options MOZ_PGO=1
ac_add_options --enable-lto
ac_add_options --prefix=/opt/firefox
ac_add_options --enable-official-branding
ac_add_options --enable-release
ac_add_options --disable-tests
ac_add_options --enable-rust-simd
ac_add_options --enable-optimize=-O3
ac_add_options --enable-alsa
ac_add_options --enable-pulseaudio
ac_add_options --disable-crashreporter
ac_add_options --disable-updater
ac_add_options --enable-linker=gold
ac_add_options AR=/usr/bin/llvm-ar
ac_add_options NODEJS=/storage2/shit/build/node/bin/node

The pulseaudio shit is just because I have it installed only for linking (so I can use the same build on my Slack, LFS and Manjaro)

bassmadrigal · 08-03-2020, 11:40 PM

Quote:

Originally Posted by LuckyCyborg

Of course, I understand well that in your All American HTPC with 64 cores / 128 threads living on liquid nitrogen freezing, no matters those issues...

However, I for one, I use 45W CPUs.

Do you realize how demeaning this is? I see you post similar things all over.

Many forum members living in the US have low powered or extremely old hardware and some forum members living in other parts of the world have monster machines. Yes, fast hardware exists and some of us run it, but I've ran my fair share of extremely low TDP computers. My current htpc is running a 65W APU (Ryzen 3 2200g), the one previous to that was a 25W AMD Athlon 5350, and before that I had a 12W Intel Atom D525 NUC box that was a size of a small book.

You could simply say that these issues might only be noticeable on low-end/older hardware rather than slam all Americans because "tHEy hAvE FasT STuFf".