sage: 6.8 -> 8.0 #31298
sage: 6.8 -> 8.0 #31298
Conversation
This "un-breaks" sage while also updating it to 8.0. It compiles sage with its dependencies as one big pile, which is not the best approach but definately better than nothing for now. To be able to shrink the huge output pile a little, it also splits docs from the rest of the output.
| license = stdenv.lib.licenses.gpl2Plus; | ||
| broken = true; | ||
| platforms = stdenv.lib.platforms.all; |
There was a problem hiding this comment.
Do you actually expect this to build on Darwin? And on Cygwin? I would put linux for now…
There was a problem hiding this comment.
I have no Idea, I thought it would be best to take an optimistic approach and restrict it when its not working. If I'd restrict the platform to linux now, probably nobody would test it for other platforms.
I don't know what the policy here is though, and I'll of course update that if the policy is to only add the platforms that were actually tested.
|
Oh, thanks. Re: dependencies — well, there are tests, and there is a mechanism of replacing a dependency with a system-wide one (when I tried to package 7.x that was complicated, and eventually I got stuck when I couldn't understand which test failures were real). Adding proper versions to Nixpkgs is the relatively easy part even without automation. |
|
Yeah that's another reason why I stopped going down that path for now and decided on just shipping something working. I think automation would be needed though, otherwise we would need to basically duplicate all the packaging work sage is doing for every update and the package would just stay outdated. |
|
I have no Idea, I thought it would be best to take an optimistic approach and restrict it when its not working. If I'd restrict the platform to linux now, probably nobody would test it for other platforms.
I think we try to follow a reasonable estimation of success chances. I don't think there is a policy that is actually followed. But this is a huge heavy build (hopefully someone will find a way to make it smaller), and it is known to be brittle, so I would recommend restricting testing it on Darwin to people who know how to override `meta.broken`
|
|
Yeah that's another reason why I stopped going down that path for now and decided on just shipping something working. I think automation would be needed though, otherwise we would need to basically duplicate all the packaging work sage is doing for every update and the package would just stay outdated.
Well, if we had tests, it would probably be enough to just make sure that version is roughly the same for most packages, then just build the correct version (changing only `src`) of the rest.
|
|
They have tests, but they also patch a lot of their dependencies. |
|
Some of their patches seem to be just related to how they organize the builds, though. |
|
Maybe, I didn't look into the details. I think the best approach would be to start with getting it working with all the exact same dependencies, just build by nix. From there its trivial to switch out different dependencies and test what is working. |
My concern was that nobody would actually know that it was available and so nobody would test it. But I'm fine with changing it, maybe someone with a darwin system will see the commit and test. |
|
Well, in reality it does sometimes happen that people using macOS try to build packages declared Linux-only and then ask to change the platform set (or submit a PR with fixes, if some changes to the build are needed). I think the Darwin part of Hydra is weaker than the Linux part, too; and without tests we won't even know if the builds are useful until someone with a Mac investigates. |
|
Okay, you're probably right that would be an unnecessary high load on hydra. |
This "un-breaks" sage while also updating it to 8.0.
It compiles sage with its dependencies as one big pile, which is not
the best approach but definately better than nothing for now.
To be able to shrink the huge output pile a little, it also splits
docs from the rest of the output.
A fair warning for anybody who would like to test this: The final test build
took 2h 45min on my laptop, with the main package taking up 2.1G and the
docs 1.8GB.
As mentioned in the comments there are some things that could be solved
better, but as it already took more time than expected and I don't know
when or if I will find that time I thought I'd push something working now.
I actually started writing a script to convert the spkg declarations to
nix packages. Thanks to nix we could pin down all those dependencies to the
exact versions used upstream and those versions could co-exist with the newest
versions. It would allow gradual picking of dependencies and overwriting single
dependencies. I'm not sure if I will go through with this though.
Fixes #27335. Also we should close #15732.
Motivation for this change
Sage has been broken and outdated for a while.
Things done
build-use-sandboxinnix.confon non-NixOS)nix-shell -p nox --run "nox-review wip"./result/bin/)I tested the execution of bin/sage, but not the other executables as most are irrelevant and/or I don't know what they do.