chore: update vendored liblbfgs from https://github.com/chokkan/liblbfgs

[szhorvat]

This attempts to update liblbfgs, and causes a crash in the test suite. Let's check if the crash is due to an issue that is present independently of the liblbfgs update. Funny thing is, it doesn't happen when compiling in Debug mode.

This is what I see on my machine:

$ test/test_hzeta
dyld[29868]: main executable has no entry point
fish: Job 1, 'test/test_hzeta' terminated by signal SIGABRT (Abort)

Occurs only with test_hzeta.

[szhorvat]

This is because tests use internal functions.

[szhorvat]

Why are the tests not failing here when using a shared library build? Some tests use non-public functions that don't get exported from the shard lib.

[szhorvat]

I see there's some hackery in the TEST_CASES_INTERNAL section. But now Windows tests are failing when they shouldn't.

[szhorvat]

Windows tests are failing

This is because the DLL is not in the PATH. My attempted fix fails because the generator expression $<TARGET_FILE_DIR:plfit> is not getting evaluated. I don't know why, I give up.

EDIT: This is fixed now. The reason was that the add_test(name command) signature doesn't evaluate generator expressions. We need the add_test(NAME name COMMAND command) signature. But then we also need to use the -C option of ctest.

I see there's some hackery in the TEST_CASES_INTERNAL section.

The main executable has no entry point error might relate to this, but again I'm stuck and I give up.

EDIT: This is not the reason.

[szhorvat]

The main executable has no entry point error occurs only with -O2 or -O3, and only with Apple Clang. It doesn't happen with GCC or LLVM Clang. Maybe let's just ignore this.

[szhorvat]

I think we should merge this as-is:

• It updates liblbfgs
• Cleans up test headers
• Expands CI runs to shared builds
• Makes sure that shared build tests pass on Windows

Quite a heterogeneous set of changes, I know.

[ntamas]

Thanks!