CI: show logs

[the-horo]

No description provided.

[the-horo]

Let's see if there's anything interesting happening with druntime-test-exceptions-release

[kinke]

Hmm, looks like the non-musl tests are wrongly excluded?

ldc/runtime/druntime/test/exceptions/Makefile

Lines 40 to 42 in 9db6d95

	ifneq ($(IS_MUSL),1)
	TESTS+=memoryerror_null_read memoryerror_null_write memoryerror_null_call memoryerror_stackoverflow
	endif

Maybe some weird Make behavior if the variable isn't defined?

[kinke]

Or the left-over auto-detection in the Makefile now breaks it because the GHA Linux images might come with an unexpected apk tool now?

ldc/runtime/druntime/test/exceptions/Makefile

Lines 6 to 11 in 9db6d95

	ifeq ($(OS),linux)
	# FIXME: detect musl libc robustly; just checking Alpine Linux' apk tool for now
	ifeq (1,$(shell which apk &>/dev/null && echo 1))
	IS_MUSL:=1
	endif
	endif

[kinke]

AFAICT from the CI logs, it's got to be the latter - unknown_gc (excluded with IS_MUSL=1) is included on macOS, but not glibc Linux:

ldc/runtime/druntime/test/exceptions/Makefile

Lines 18 to 20 in 9db6d95

	ifneq ($(IS_MUSL),1)
	TESTS += unknown_gc
	endif

[the-horo]

Or the left-over auto-detection in the Makefile now breaks it because the GHA Linux images might come with an unexpected apk tool now?

It doesn't seem apk exists by default. Let's see if it somehow pops up by the time the tests are run

[the-horo]

Or the left-over auto-detection in the Makefile now breaks it because the GHA Linux images might come with an unexpected apk tool now?

It doesn't seem apk exists by default. Let's see if it somehow pops up by the time the tests are run

Nope, it's still not there. But the tests "detect" it anyways. Is the make shell function doing something we don't expect?

[the-horo]

Alright, I genuinely don't have any idea how this is possible:

ldc/runtime/druntime/test/exceptions/Makefile

Lines 6 to 19 in 2116615

	$(info $(shell which apk ; echo $$?))
	$(info $(shell which apk && echo 1 ; echo $$?))
	$(info $(shell command -v apk ; echo $$?))
	$(info $(shell command -v apk && echo 1 ; echo $$?))
	ifeq ($(OS),linux)
	# FIXME: detect musl libc robustly; just checking Alpine Linux' apk tool for now
	ifeq (1,$(shell command -v apk &>/dev/null && echo 1))
	$(warning ~~~ musl detected)
	IS_MUSL:=1
	else
	$(warning ~~~ musl NOT detected)
	endif
	endif

• linux-aarch64

1817: 1
1817: 1
1817: 127
1817: 127
1817: Makefile:14: ~~~ musl detected

• alpine

1822: /sbin/apk 0
1822: /sbin/apk 1 0
1822: /sbin/apk 0
1822: /sbin/apk 1 0
1822: Makefile:14: ~~~ musl detected

[the-horo]

• linux aarch64

ldc/runtime/druntime/test/exceptions/Makefile

Lines 6 to 9 in 6171ff1

	$(info $(shell which apk &>/dev/null; echo $$?))
	$(info $(shell which apk &>/dev/null && echo 1 ; echo $$?))
	$(info $(shell command -v apk &>/dev/null ; echo $$?))
	$(info $(shell command -v apk &>/dev/null && echo 1 ; echo $$?))

1817: 0
1817: 1 0
1817: 0
1817: 1 0
1817: Makefile:15: ~~~ musl 1

Why is &> changing the exit status???

[the-horo]

/bin/sh is dash in the runner, which doesn't understand &>.

---

There's already

ldc/runtime/DRuntimeIntegrationTests.cmake

Lines 60 to 63 in 9db6d95

	set(musl "")
	if(TARGET_SYSTEM MATCHES "musl")
	set(musl "IS_MUSL=1")
	endif()

we can just use that, instead of looking for apk. Does that sound good @kinke?

[kinke]

Oh god; thx for digging!

we can just use that, instead of looking for apk.

Yeah, we here for LDC can, but I think I've added this apk-logic upstream when adding the Alpine CI job (for DMD). So we could skip the logic for IN_LDC, but DMD upstream needs a solution.