lib/systems/parse.nix: mkSkeletonFromList: improve readability

[ghost]

Motivation

The main purpose of this PR is to make the basis for mkSkeletonFromList's decision between cpu-kernel-libcabi vs cpu-vendor-os clear, without changing its behavior. The existing code obscures this decision behind a sequence of prioritized matches (i.e. if-then) which jump around between different coordinates.

Two side benefits of this PR:

1. It makes the root cause of Embedded target triplets (system, config) are used and interpreted incorrectly #165836 obvious: we are missing a case for cpu-vendor-libcabi. This is why nixpkgs stumbles over *-none-*.

2. It illuminates some very weird corner cases in the existing logic, like *-${vendor}-ghcjs overriding the vendor field, and mingw32 being transformed into windows in some cases.

Instructions for verifying no behavior change

curl -L https://raw.githubusercontent.com/amjoseph-nixpkgs/nixpkgs/a9595374806477c40eab88acbc06ca7d641364ea/lib/systems/test.nix > lib/systems/test.nix
nix eval --raw -f lib/systems/test.nix | egrep -ve '-apple-(mingw32|eabi|elf|ghcjs)' > after
git switch --detach origin/master
nix eval --raw -f lib/systems/test.nix | egrep -ve '-apple-(mingw32|eabi|elf|ghcjs)' > before
diff -u before after

The diff should produce no output. You should verify that the *-apple-* tuples removed by the egrep -v filters are nonsensical (changing the behavior in these nonsensical cases allowed simplification -- see below for details). Here is how lib/systems/test.nix was created:

1. Copy-and-paste of the old (unsimplified) version of mkSkeletonFromList
2. Comment it out
3. Form a list consisting of one or more attrsets for each possible coordinate value which could participate in triggering each of the if clauses in the original code
4. Append "default values" for each of the three coordinates.
5. Run mkSkeletonFromList on the cartesianProductOfSets of the attrsets.

Note that if vendor is missing from the return value of mkSkeletonFromList, lib/systems/test.nix will set vendor="unknown"

Things done

• Built on platform(s)
  • x86_64-linux
• Tested basic functionality of all binary files (usually in ./result/bin/)
• Fits CONTRIBUTING.md.

[ghost]

Good to go. For ease of reviewing, see the copypasta at the top of the first post.

Please squash before merging, or ask me to do so.

[ghost]

@sternenseemann should be added as a reviewer here.

[ghost]

@sternenseemann and @Ericson2314, have you had a chance to look at this yet?

There are a few PRs open (like #186484) which, if merged ahead of this one, will invalidate the proof that the behavior of mkSystemFromString is unchanged.

[ghost]

Ping @sternenseemann @Ericson2314 @alyssais @nbp

I want to submit other changes which will create a merge conflict with this PR. What should I do?

We're starting to run into a lot more problems with calling Musl an ABI. Nixpkgs needs understand that "big-endian 64-bit powerpc on Musl always uses the ELFv2 ABI (see #182807)", but since nixpkgs thinks that Musl is an ABI there is no way to express this other than special-purpose hackery like #191436. The fact that {build,host,target}Platform don't have separate fields for libc and abi which can be matched independently is the main obstacle. Fixing that will create a merge conflict with this PR.

I've radically shortened the description at the top of this PR in the hope of attracting some kind of review.

[Ericson2314]

@amjoseph-nixpkgs this looks good! Sorry I didn't see it for so long, Please harass me on Matrix in #cross-compiling:nixos.org to make sure this stuff gets reviewed, this is not the first time you made a good thing that went neglected by us!

[sternenseemann]

Also did some half-elaborate sanity checking of my own which seems to check out.

One thing is that some things in parse.nix have changed since July, but I'm not sure if we need to adjust any code in here necessarily.

[ghost]

Merge conflict due to #82131.

[ghost]

Squashed, because it wasn't worth rebasing past 25 separate commits.

I've added *-*-freebsd* to the cpu-vendor-os fork of the *-*-* case.

Note that freebsd can be used in the other (cpu-kernel-environment) fork as well; x86_64-freebsd-gnu is still valid although I don't think glibc still builds on freebsd. But there's no need to parse it unless somebody is serious about adding that capability.

[Ericson2314]

Sorry we didn't merge before! :(

[github-actions]

Successfully created backport PR #202405 for release-22.11.

[github-actions]

Git push to origin failed for release-22.11 with exitcode 1

[ghost]

Git push to origin failed for release-22.11 with exitcode 1

I think sterni fixed this in #202405. If not, or if I need to do anything more, ping me.