julia: init julia_10-bin and julia_16-bin #123188

ghost · 2021-05-16T01:46:32Z

Motivation for this change

Recent attempts to package Julia v1.6.x and unbreak v1.0.x have failed and stalled. As it stands, we know that we have test failures and breakages for the source-based expressions in Nixpkgs. As a response to this, I adapted an older expression that uses autoPatchelfHook on the official binary releases and added a few patches to resolve test failures. This is what you are seeing in this PR.

Note that I am currently only supporting x86_64-linux, but expect me to add i686-linux and aarch64-linux as soon as I have access to hardware that will allow me to develop and test them. My strategy is to get on the NixOS aarch64 build box, but pointers for sensible development and testing for i686-linux would be welcome. If the merging of this PR drags on I will try to add support for them here, otherwise expect swift PRs to follow.

I think there are good reasons to maintain both source and binary packages for Julia given the complexity that comes with packaging it – as seen in the other issue and failures to get v1.6.x into a passable state. Furthermore, I have managed to get system image patching working, so that we could even potentially have a binary release with patches necessary to unbreak the parts of the Julia package ecosystem that is currently failing for both source and binary releases on Nix. Note that this is not in this PR as it is not necessary to get a binary release on par with our existing source-based packages.

Minor note, patches lifted from my Julia fork for the relevant releases. Thankfully these could be recycled from my failed attempts at source builds.

@7c6f434c, @garrison, and @rbvermaa: I am currently standing as sole maintainer for this one, but if you are up for it I am happy to add you to share the “load” as you stand as maintainers for the source-based Julia packages.

@samuela: I am still unable to develop and test on x86_64-darwin, so if you (or anyone else) want to (well “can”, really) maintain it I would be very thankful.

Things done

Edit: Spelling and grammar.

7c6f434c · 2021-05-16T09:26:42Z

pkgs/development/compilers/julia/1.0-bin.nix

+    cp -r . $out
+    # Setting `LD_LIBRARY_PATH` resolves `Libdl` failures. Not sure why this is
+    # only necessary on v1.0.x and a cleaner solution is welcome, but after
+    # staring at `strace` for a few hours this is as clean as I could make it.


Maybe 1.0 adds $out/lib to RPATH but links no libraries (using dlopen instead), so Nixpkgs automatic patchelf removes it from RPATH?

Thank you, that could be it. Will investigate.

Yep, I suspect you are bang on:

> objdump -x lib/libjulia.so | grep RPATH RPATH $ORIGIN/julia:$ORIGIN > objdump -x /nix/store/pxs6qkddkhpvk1hpkahgz4w18ahfd3m1-julia-bin-1.0.5/lib/libjulia.so | grep RPATH >

Digging through the manual I thought that what I should do was to set dontPatchELF to true, but I still end up with the RPATH being removed.

If this is not easily addressable, I am happy to merge this as is and try to clean it up later.

7c6f434c · 2021-05-16T09:27:31Z

pkgs/development/compilers/julia/1.0-bin.nix

+    homepage = "https://julialang.org";
+    # Bundled and linked with various GPL code, although Julia itself is MIT.
+    license = lib.licenses.gpl2Plus;
+    maintainers = with lib.maintainers; [ ninjin ];


Fine with being added. For example, I will be notified when it's time to merge updates…

Happy to have you onboard! Added and pushed.

7c6f434c · 2021-05-16T09:39:32Z

Looks good in general

ghost · 2021-05-16T10:03:10Z

pkgs/development/compilers/julia/1.0-bin.nix

+    description = "High-level, high-performance dynamic language for technical computing";
+    homepage = "https://julialang.org";
+    # Bundled and linked with various GPL code, although Julia itself is MIT.
+    license = lib.licenses.gpl2Plus;


Unrelated to this PR, but does this mean that we should chance the license from MIT to GPL 2+ for our source-based Julia packages as well?

Nobody is sure what exactly our license metadata means in any non-trivial case, so …

Here it is clear: here a GPL2+ artifacy is downloaded.

Relevant patches lifted from the `release-1.6` branch of my Julia fork: https://git.sr.ht/~ninjin/julia-nix/tree/release-1.6

Almost as clean as it could be, but was forced to set `LD_LIBRARY_PATH` to work around a `Libdl` failure that is unique to v1.0.x.

samuela

Overall LGTM. Just a little confused about the comments.

samuela · 2021-05-16T18:59:04Z

pkgs/development/compilers/julia/1.0-bin.nix

+  # Julia’s source files are in different locations for source and binary
+  # releases. Thus we temporarily create symlinks to allow us to share patches
+  # with source releases.


I don't really understand this comment. AFAICT there's no overlap between the patches used for the source and binary derivations. All of the patches seem to be split up: patches/1.0, patches/1.0-bin, etc.

It is not referring to the location in the Nixpkgs source tree, but rather the paths in the patches. If we do not do this trick we would have to alter the patches themselves as the paths are different between the Julia release tarballs.

samuela · 2021-05-16T18:59:33Z

pkgs/development/compilers/julia/1.6-bin.nix

+  # Julia’s source files are in different locations for source and binary
+  # releases. Thus we temporarily create a symlink to allow us to share patches
+  # with source releases.


Same as my other comment.

samuela · 2021-05-16T19:02:11Z

pkgs/top-level/all-packages.nix

@@ -11113,6 +11113,13 @@ in
  julia-stable = julia_15;


I think we should reshuffle some of these as well. In particular, I don't think it makes sense for any of the "flagship" attributes to point to julia_15 since that one's broken.

Also I think an alias julia_16 = julia_16-bin would make sense.

+1 julia_16 = julia_16-bin

I agree that this makes sense, but I would rather address that in a separate PR and/or issue.

samuela · 2021-05-17T08:48:02Z

I took a quick stab at getting this running on x86_64-darwin, but the test suite isn't happy: https://gist.github.com/samuela/257b66127bceffcc11453a6992a92d3d. It's something having to do with spawning the test threads, potentially related to the net_on adventures...

samuela · 2021-05-17T08:49:37Z

But getting it building on x86_64-darwin shouldn't be a requirement for this PR IMHO. We can always follow up later to broaden platform support.

Follow up to NixOS#123188. Now that we have binary builds for julia v1.6.x, we should alias `julia_16` to point to `julia_16-bin`.

ofborg bot added 8.has: package (new) This PR adds a new package 11.by: package-maintainer This PR was created by the maintainer of the package it changes 10.rebuild-darwin: 0 This PR does not cause any packages to rebuild on Darwin 10.rebuild-linux: 1-10 labels May 16, 2021

This was referenced May 16, 2021

Julia 1.6.0 #116697

Closed

julia_15: run test suite and mark as broken #121114

Merged

7c6f434c reviewed May 16, 2021

View reviewed changes

ghost commented May 16, 2021

View reviewed changes

Pontus Stenetorp added 2 commits May 16, 2021 14:12

julia: init julia_16-bin at 1.6.1

351f911

Relevant patches lifted from the `release-1.6` branch of my Julia fork: https://git.sr.ht/~ninjin/julia-nix/tree/release-1.6

julia: init julia_10-bin at 1.0.5

8949ae1

Almost as clean as it could be, but was forced to set `LD_LIBRARY_PATH` to work around a `Libdl` failure that is unique to v1.0.x.

ofborg bot requested a review from 7c6f434c May 16, 2021 14:22

samuela requested changes May 16, 2021

View reviewed changes

7c6f434c merged commit 183350a into NixOS:master May 17, 2021

ghost deleted the julia-bin branch May 17, 2021 12:25

samuela mentioned this pull request May 17, 2021

julia_10-bin, julia_16-bin: add x86_64-darwin support #123394

Closed

samuela added a commit to samuela/nixpkgs that referenced this pull request May 17, 2021

julia_16: alias to julia_16-bin

7e3e790

Follow up to NixOS#123188. Now that we have binary builds for julia v1.6.x, we should alias `julia_16` to point to `julia_16-bin`.

samuela mentioned this pull request May 17, 2021

julia_16: alias to julia_16-bin #123395

Closed

10 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

julia: init julia_10-bin and julia_16-bin #123188

julia: init julia_10-bin and julia_16-bin #123188

ghost commented May 16, 2021 •

edited by ghost

Loading

7c6f434c May 16, 2021

ghost May 16, 2021

ghost May 16, 2021

ghost May 17, 2021

7c6f434c May 16, 2021

ghost May 16, 2021

7c6f434c commented May 16, 2021

ghost May 16, 2021

7c6f434c May 17, 2021

samuela left a comment

samuela May 16, 2021

ghost May 17, 2021

samuela May 16, 2021

ghost May 17, 2021

samuela May 16, 2021

GTrunSec May 16, 2021

ghost May 17, 2021

samuela commented May 17, 2021

samuela commented May 17, 2021

julia: init julia_10-bin and julia_16-bin #123188

julia: init julia_10-bin and julia_16-bin #123188

Conversation

ghost commented May 16, 2021 • edited by ghost Loading

Motivation for this change

Things done

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

7c6f434c commented May 16, 2021

Choose a reason for hiding this comment

Choose a reason for hiding this comment

samuela left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

samuela commented May 17, 2021

samuela commented May 17, 2021

ghost commented May 16, 2021 •

edited by ghost

Loading