Ensure all jobset attributes evaluate, and check that in CT

[ghost]

Description

aka "eval checks go brrrr"

Please review the commits individually. Produced as part of the non-Hydra tooling used to validate

• default-gcc-version: 12 -> 13 if !isDarwin && !isAarch64 #268097

Description of changes

This PR adds release-attrpaths-superset.nix, which walks the entire nixpkgs tree and emits a superset of all release attrnames in only 44 seconds on a 3ghz CPU, using less than 5 gbytes of memory. By comparison, on the same CPU the nix-env -qaP hack used by ofborg on every PR submission requires waiting 25 minutes and peaks at 58 gbytes, even with checkMeta turned off. Full details below. This is:

• 33x faster (or 2.9% of the elapsed time)
• 12.5x less memory (or 8.0% of the peak memory usage)

If all you need is the attrnames -- for example to launch a mass rebuild -- then you're done. If you want to replace the ofborg outpaths.nix step, see the follow-up PR:

• pkgs/top-level/release-outpaths-parallel.nix 5x faster eval checks #269403

Validation

To confirm that the release-attrpaths-superset.nix really are a superset of the slower method, run

nix-build pkgs/test/release/release-attrpaths-superset.nix -A failures

Additional CT (Continuous Testing) check

This PR adds recurseForRelease=false where necessary within nixpkgs in order to make this possible -- it screens out infinite recursions and several non-tryEval-catchable failures (for example, passing a mistyped argument to a builtin, which is an uncatchable error) and infinite recursions.

Finding all the places where it was necessary to add this was a lot of tedious work, so to make sure that work is not undone, lib/tests/release.nix now verifies that release-attrpaths-superset.nix evaluates successfully.

Although it doesn't really fit in lib/tests, there isn't anywhere else it can be placed and still be run by ofborg on every commit. If some other such place materializes in the future I would be happy to relocate it, but since ofborg is out-of-tree this PR should not block on that happening.

Fixes

• enumerating release jobset names is single-threaded and slow #269357

[ghost]

Measurements

According to GNU Time release-attrpaths-superset.nix consumes:,

Command being timed: "nix-instantiate --eval --strict --json pkgs/top-level/release-attrpaths-superset.nix -A names"
User time (seconds): 44.88
System time (seconds): 8.09
Percent of CPU this job got: 99%
Elapsed (wall clock) time (h:mm:ss or m:ss): 0:53.20
Average shared text size (kbytes): 0
Average unshared data size (kbytes): 0
Average stack size (kbytes): 0
Average total size (kbytes): 0
Maximum resident set size (kbytes): 4823028
Average resident set size (kbytes): 0

Compared to release-outpaths.nix:

Command being timed: "nix-env -qaP --no-name --out-path --arg checkMeta false --argstr path /git/work/pr/release-outpaths -f pkgs/top-level/release-outpaths.nix"
User time (seconds): 1294.48
System time (seconds): 224.37
Percent of CPU this job got: 99%
Elapsed (wall clock) time (h:mm:ss or m:ss): 25:25.01
Average shared text size (kbytes): 0
Average unshared data size (kbytes): 0
Average stack size (kbytes): 0
Average total size (kbytes): 0
Maximum resident set size (kbytes): 58336672
Average resident set size (kbytes): 0

[ghost]

Rebased.

[infinisil]

Looking good to me, tested locally. To prevent further conflicts I'll merge now! 🚀

Edit: Hold on, I'll also quickly test this on Darwin