Woodruff: Weird architectures weren't supported to begin with

Posted Mar 1, 2021 17:45 UTC (Mon) by anholt (guest, #52292)
Parent article: Woodruff: Weird architectures weren't supported to begin with

I went through this recently in the Mesa project, where I am working to remove tens of thousands of lines of unmaintained, architecturally incorrect (as in "no route to passing the conformance tests"), non-CI-tested, unused code. Another developer complained that my alternative (the less-unmaintained, architecturally-plausible, CI-tested software rasterizers) would be slower on non-LLVM architectures for the debian packages which were using the old driver.

So I looked at debian's popcon, and the sum of all of the non-LLVM architectures installing Mesa's drivers was .02% of the install base of amd64 alone. And that's in debian, where the hobby maintainers of those architectures are more likely than the average user (or even more so the users of debian derivatives) to have popcon installed and thus are overrepresented.

In the end, I just talked to the debian package maintainer, who said (paraphrasing) "Oh, I can just flip the non-llvm architectures away from the old driver and have less untested special-case package build system for these dead arches? I'll do that right now" and then I got to go delete stuff now since it was unreachable from any distribution. But I don't think that should have even been necessary.

Hobby architectures are fun. I've got my own spare-time project for supporting a hardware driver in Mesa that's been effectively unmaintained for a decade or more since it has basically no users left. But hobbyists like me need to understand that software maintenance is work and nobody else has to keep our fun time projects going.

And, for the Red Hats of the world taking money for shipping s390x, it's actually their job then to make s390x work, not anyone else. But then, I don't see RHers doing so in my world, it's more in the open source purists where I see the push-back against regressing hobby hardware.

Woodruff: Weird architectures weren't supported to begin with

Posted Mar 1, 2021 17:54 UTC (Mon) by rvolgers (guest, #63218) [Link]

One thing that got kind of buried in the discussion is that Rust actually supports s390x and provides official releases for it: https://doc.rust-lang.org/nightly/rustc/platform-support....

It's just that some distros didn't package it on that architecture, and some people *did* actually need s390 and then others read the discussion as saying s390x was unsupported.

Woodruff: Weird architectures weren't supported to begin with

Posted Mar 1, 2021 18:15 UTC (Mon) by pizza (subscriber, #46) [Link] (2 responses)

> And, for the Red Hats of the world taking money for shipping s390x, it's actually their job then to make s390x work, not anyone else. But then, I don't see RHers doing so in my world, it's more in the open source purists where I see the push-back against regressing hobby hardware.

Let's not forget that it's those same "open source purists" that have received little to no direct compensation for their efforts over the years and are now finding that the social contract has been unilaterally changed on them.

That "non-hobby" stuff cuts both ways; why should they continue supporting non-hobby users when it's clear it's not reciprocated?

Woodruff: Weird architectures weren't supported to begin with

Posted Mar 1, 2021 18:54 UTC (Mon) by anholt (guest, #52292) [Link] (1 responses)

I've made a lucrative career out of being an open source purist[1]. So have the open source developers I work with daily.

I've also been solidly on the hobbyist side, back when I was a college student porting drivers to FreeBSD just because I wanted to (though I did a couple of times get a "free" graphics card from a contractor who would have had to do the porting otherwise, and I thought that was a pretty sweet deal at the time and I still appreciate what they did for me).

I agree there are some weaknesses in corporate interests supporting shared infrastructure of open source, particularly for efforts like reproducible builds or CI farms. But of the open source software that I'm relying on, I see it overwhelmingly written and maintained by caring open source developers who are doing it in their day jobs, not by people like me-in-college doing it for fun.

[1] I've stooped as low as "run closed source simulators as part of otherwise pure open source driver development" and "hold unreleased-hardware driver code behind closed doors for up to a year before releasing under the MIT license", and I once got burned by a vendor who never let us release our few thousand lines of work we did before we caught on that the project was doomed. So, not as pure as the more pure people I know, who incidentally are also employed full time in free software.

Woodruff: Weird architectures weren't supported to begin with

Posted Mar 1, 2021 20:43 UTC (Mon) by Wol (subscriber, #4433) [Link]

> [1] I've stooped as low as "run closed source simulators as part of otherwise pure open source driver development" and "hold unreleased-hardware driver code behind closed doors for up to a year before releasing under the MIT license",

If you're not a Free Software Fanatic, why not? And even then, when you look at the RMS printer story, what you did should be perfectly okay!

> and I once got burned by a vendor who never let us release our few thousand lines of work we did before we caught on that the project was doomed. So, not as pure as the more pure people I know, who incidentally are also employed full time in free software.

Which is when you realise the NDA should have said "while the product is in pre-release development", not "until the product is released".

Cheers,
Wol

Woodruff: Weird architectures weren't supported to begin with

Posted Mar 2, 2021 17:15 UTC (Tue) by daenzer (subscriber, #7050) [Link]

I merely pointed out:

* The old swrast code is usable with some real-world interactive apps, while softpipe isn't.

* At the time there were still Debian architectures where swrast was shipped and llvmpipe wasn't available.

I never made any claims about how many users this would actually affect. Even if it's > 0, one can certainly make an argument that the number is too small to continue offering swrast as a choice. All I asked was for this to be a conscious and honest decision, so that any affected users might have felt treated with a modicum of respect.