Re: Mozilla Software on Sparc64/Linux
Hello!
On 11/14/21 17:58, Riccardo Mottola wrote:
>> i would be very interested in getting Firefox and Thunderbird (and possibly
>> Seamonkey, but this isn't available at all with debian it seam) running, if
>> possible at all for the newer versions.
>
> yes, SeaMonkey is for me a missed package for both Debian/Devuan and FreeBSD.
> I just love it. I like the classic interface and can't stand the chrome-like
> of Firefox.
There is currently no one willing to step up being the maintainer of Seamonkey
in Debian which is why it's currently not included.
> SeaMonkey up to 2.49 was FF 52 based, so very portable and no rust. It should
> be a decent candidate to get it running on SPARC, since a specific Mac PPC G5
> version was maintained for a long time. Current SM is FF 60 based and so has rust
Rust is available on SPARC and fully supported.
>> Rust seems not to be an issue any more for sparc64/linux, but the general upstream
>> neglect of sparc64 seems to be the major problem.
>
> Rust is evil... however as you write it does not to be the biggest problem here.
Calling Rust "evil" is not a technical argument but an emotional one.
> The major issue is endianness - current FireFox runs on PPC64 but is totally unusable.
> Everything in the UI has mangled colors, endianness is broken at about every level.
> E.g. FF relies now a lot on skia and officially it will never support BE. And so a
> lot of other issues, including Mesa.
As I mentioned before, Oracle officially support Firefox on Solaris on SPARC and thus
on a big-endian system. Debian ships Firefox on s390x as a supported package.
Claiming that Firefox is a mess on big-endian is definitely not accurate although
Firefox upstream does not officially support big-endian targets, but they don't
prevent it either.
> Add to endianness issues that SPARCs are sensitive to memory alignment and issues and you get the crashes.
The more Rust code Firefox has, the less likely this is going to happen as Rust is much
stricter with alignment than C/C++.
> I am working since a long time on ArcticFox and intend to keep it as much as possible
> cross-platform, close to Firefox, but incorporating as many fixes done for TenFourFox
> as I can. Currently, it is quite usable on PPC, although there are endianness issues
> with image composition operations. I can browser Wikipedia and even watch a youtube
> video with audio. That is already impossible with standard Firefox.
Well, it's based on an ancient Firefox code base which is why it still has many of the
old bugs that have been fixed in the mean time.
>> I would like to hear opinions on how to proceed or if you think this is a lost cause?
>
>
> It is not a lost cause, but it is a hard cause, not well supported by upstream. I intend
> to generalize a lot of TenFourFox patches in ArcticFox, but it requires work.
>
> NetBSD should have a working FF (I think FF 52) on SPARC.
>
> I was finally able to compile ArcticFox on NetBSD/SPARC64 and Linux/SPARC64. It will start,
> show a window but crash very soon. That's already an improvement compared to crashing immediately
> as it did one year ago!
We have already had a fully working Rust-based Firefox on SPARC. I don't think it's a good
idea to use these ancient versions, especially since there is no official security support.
> For this cause, I need help - I am working alone and I have no real Gecko knowledge. I need
> to port patches from Gecko to improve certain parts, so I need somebody how knows Firefox
> code more than me to help me understand why certain give issues. Not specific SPARC, but I
> work mostly on amd64 and arm64 and then "test" on PPC and SPARC (my netra T1 takes 20 hours
> to compile ArcticFox....) So if you know somebody who can help me with some specific issues
> please ping me. I have some blocking issues I need to solve so that ArcticFox doesn't die
> out and can evolve, specifically some JavaScript and JIT issues, currently verified on Intel.
It's probably a better idea to start working on separating the Javascript transpilation process
in the Firefox build system from the build process for the native code as this will remove
the NodeJS build dependency for Firefox.
Firefox itself does not require NodeJS. It's just that the Mozilla developers decided to pull
in NodeJS as a hard dependency for the build so that the Javascript files are always freshly
transpiled from the source.
However, the transpilation process does not necessarily have to happen on the same architecture
and one can transpile the Javascript files on x86_64 and copy them into the Firefox build
tree later. This is what Oracle does on Solaris and we could do the same on Debian by moving
the transpiled Javascript packages into a firefox-common package which will be used by the
firefox package on the various architectures.
This way we could use current versions of Firefox and wouldn't be stuck to old and buggy
versions. I have already a concept for the common package, but I didn't have the time for
turning the concept into code yet.
Adrian
--
.''`. John Paul Adrian Glaubitz
: :' : Debian Developer - glaubitz@debian.org
`. `' Freie Universitaet Berlin - glaubitz@physik.fu-berlin.de
`- GPG: 62FF 8A75 84E0 2956 9546 0006 7426 3B37 F5B5 F913
Reply to: