run-program fails on clisp linux x64 2c exit code matches? FAIL

[Faré]

2 wait-process
 a exit code after exit: synchronous
 b exit code after exit: asynchronous (repeated)
   ......................................................... NOT IMPLEMENTED
 c exit code after signal: synchronous
   exit code matches?....................................... FAIL
...
 2 sub-tests failed.
13 sub-tests could not be run on this platform (not implemented).
Exiting with non-zero code because of unexpected test failures.
Using clisp, test-run-program-unix.script failed
you can retry compilation with:
./test/run-tests.sh clisp test-run-program-unix.script

NB: The summary says 2 sub-tests failed, but I see only one occurrence of FAIL in the logs.

Using /nix/store/g8kyk8d4pfsaq185a9hb280d902kygx5-clisp-2.49/bin/clisp from NixOS with channel /nix/store/6cqdzx8h6ir2mnphjh8gkhjh48ka82ld-user-environment (nixpkgs at 210b3b31 on linux x64).

[pipping (pipping)]

The test suite assumes the FHS and thus hardcodes /bin/sh. Since I see the word NixOS in your bug report, I would assume its complete disregard for the FHS to be the reason for this test failure (and I would expect this to be independent of clisp).

[pipping (pipping)]

This does not just affect the test suite, however. This part of uiop/run-program:

354- (etypecase command
355: #+os-unix (string `("/bin/sh" "-c" ,command))

should affect any CL platform, while this

1015- #+clisp
1016- (let ((raw-exit-code
1017- (or
1018- #.`(#+os-windows ,@'(ext:run-shell-command %command)
1019: #+os-unix ,@'(ext:run-program "/bin/sh" :arguments `("-c" ,%command))

should affect only clisp.

[pipping (pipping)]

Hm. Then again, http://nixos.org/nixos/about.html says...

  [..] The only exception is a symlink /bin/sh to Bash in the Nix store [..]

so that there is actually a /bin/sh, yes? Then the issue must be something else...

[pipping (pipping)]

What happens when you do

  ./asdf/test/killer.sh; echo $?

or

  sh asdf/test/killer.sh; echo $?

directly on nixos?

[Faré (fahree)]

NixOS has /bin/sh and /usr/bin/env, because they are mandated by POSIX, and that's all.

The tests pass on other implementation s, so whatever breaks on CLISP, it's more subtle than that.

NB: I needed a trivial fix to run-tests.sh to get it to run since pwd isn't in /bin. It's in the fixes branch.

[pipping (pipping)]

Have you tried running

This is what the test does (from within the test dir, where the killer.sh script lives): It checks if

  (nth-value 2 (uiop:run-program "./killer.sh" :ignore-error-status t))

equals 143 and fails otherwise. What does that return from within clisp on nixos for you?

[Faré (fahree)]

CLISP returns the exit code 0, which I'll blame on NixOS using an ancient release tarball rather than a recent checkout.

[Faré (fahree)]

Closing. This is a bug due to NixOS using an ancient release tarball, and CLISP not having released their bugfixes in years.

[pipping (pipping)]

I actually regularly test with (a manually built) clisp 2.49 on fedora and there I'm not seeing any issues of this type. Hm...