build-gnu.sh: Don't symlink missing bins to false and force enable tests

[oech3]

GNU test suite auto-disables tests for misssing binaries. Symlinking them to false cause unwanted failures (e.g. runcon error with SELINUX_ENABLED=0).

[github-actions]

GNU testsuite comparison:

Congrats! The gnu test tests/runcon/runcon-no-reorder is no longer failing!
Note: The gnu test tests/runcon/runcon-compute is now being skipped but was previously passing.

[oech3]

It is also false-negative.

…

Congrats! The gnu test tests/runcon/runcon-no-reorder is no longer failing!

[oech3]

@pixelb Would you add _selinux_required to tests relying on runcon and chcon at GNU?
This is uutils's fault, but I have no idea better than it...

[pixelb]

@pixelb Would you add _selinux_required to tests relying on runcon and chcon at GNU? This is uutils's fault, but I have no idea better than it...

This would be confusing, as then folks may think selinux is required for basic runcon and chcon operation.
uutils has patches for gnu tests, so perhaps this would be an appropriate adjustment for there?

symlinking the binaries to false seems unusual BTW. Can you not just not have them present if they're no supported?

[oech3]

Can you not just not have them present if they're no supported?

It causes another side effect at #9314 (comment)
I want to count runcon-no-reorder and runcon-compute as failed tests with SELINUX_ENABLED=0 since uutils does not have *con buildable without libselinux headers. @sylvestre Thoughts?

[oech3]

Is this is breaking LD_PRELOAD test too?:
'libstdbuf.so' was not built with uutils, using the 'false' program.

[github-actions]

GNU testsuite comparison:

Skipping an intermittent issue tests/misc/tee (passes in this run but fails in the 'main' branch)
Skipping an intermittent issue tests/tail/overlay-headers (passes in this run but fails in the 'main' branch)
Congrats! The gnu test tests/runcon/runcon-no-reorder is no longer failing!
Note: The gnu test tests/runcon/runcon-compute is now being skipped but was previously passing.

[github-actions]

GNU testsuite comparison:

Congrats! The gnu test tests/runcon/runcon-no-reorder is no longer failing!
Note: The gnu test tests/runcon/runcon-compute is now being skipped but was previously passing.

[oech3]

coreutils/util/build-gnu.sh

Lines 238 to 239 in b37da0a

	# Remove the check whether a util was built. Otherwise tests against utils like "arch" are not run.
	sed -i "s|require_built_ |# require_built_ |g" init.cfg

?!

[sylvestre]

GNU testsuite comparison:

Congrats! The gnu test tests/runcon/runcon-no-reorder is no longer failing!
Note: The gnu test tests/runcon/runcon-compute is now being skipped but was previously passing.

What is going on here?

[oech3]

GNU no longer builds runcon and chcon in the non-SELinux system and it affects to GNU test suite. I guess those results were produced since uutils test suite wrapper was matched to the GNU's behavior by this MR. I think ther are making sense by that reason.

[oech3]

I guess runcon-no-reorder is no longer failing! means it was passing only with SELINUX_ENABLED=1 previously and failed with =0 with runcon=false and I just removed the wrong test.

I guess runcon-compute is now being skipped but was previously passing means it was passing on =0 with runcon=false (we should not consider it as passing).

[github-actions]

GNU testsuite comparison:

Skip an intermittent issue tests/tail/overlay-headers (fails in this run but passes in the 'main' branch)
Note: The gnu test tests/chcon/chcon-fail is now being skipped but was previously passing.
Note: The gnu test tests/runcon/runcon-compute is now being skipped but was previously passing.

[oech3]

I merged #9455 to this PR. #9314 (comment) is natural result (for SELINUX_ENABLED=0) since GNU coreutils does not build *con in such situation.

Would you merge this?

[oech3]

@pixelb You said

upstream
removed: arch,coreutils,hostname

So is test suite incomplete (invalid-opt) if uutils does not manually enable hostname too?

[pixelb]

@pixelb You said

upstream
removed: arch,coreutils,hostname

So is test suite incomplete (invalid-opt) if uutils does not manually enable hostname too?

Right. One needs to now manually enable arch,hostname,kill,uptime

[oech3]

#9314 (comment) is useless for some case?

[pixelb]

#9314 (comment) is useless for some case?

Oh right, that's commented out. With that you wouldn't need to explicitly enable arch,kill,... in the build

[oech3]

Thankyou. On the other hand, disabling every binaries with the sed broke test suite. It is unclear behaviro for me...

[oech3]

I'll remake this after fixed getlimits things. While symlink missing bins to false is hacky, it should not be serious problem...

[oech3]

@pixelb Would you add _selinux_required to tests relying on runcon and chcon at GNU? This is uutils's fault, but I have no idea better than it...

This would be confusing, as then folks may think selinux is required for basic runcon and chcon operation.

How about making(detatching) a basic operation or stub specific GNU test and marking _selinux_required other tests?
(I'm temporally opening a PR to force-add _selinux_required by sed )

[oech3]

@pixelb Is it able to redesign runcon-compute.sh to block 'false' -> 'runcon' accidentary passing the test? It seems we cannot stop

coreutils/util/build-gnu.sh

Lines 108 to 116 in ef496b6

	# Any binaries that aren't built become `false` to make tests failure
	# Note that some test (e.g. runcon/runcon-compute.sh) incorrectly passes by this
	for binary in $(./build-aux/gen-lists-of-programs.sh --list-progs); do
	bin_path="${UU_BUILD_DIR}/${binary}"
	test -f "${bin_path}" || {
	echo "'${binary}' was not built with uutils, using the 'false' program"
	cp "${UU_BUILD_DIR}/false" "${bin_path}"
	}
	done

since we inject uutils/coreutils to test suite's framework itself...

[pixelb]

@pixelb Is it able to redesign runcon-compute.sh to block 'false' -> 'runcon' accidentary passing the test?

Done in coreutils/coreutils@8f4a0095a

[oech3]

Thankyou. I'm not sure why uutils was incorrectly passing it. Maybe, the test has 2 different meaning for real runcon and stub runcon, (and previously passing later in a proper sense?)