Only save original working directory if death tests are enabled

[knutpett]

On a diskless system you cannot get the current directory. So if
death tests are disabled anyway, there is no point trying to
get current directory.

Without this fix, running tests on diskless systems will fail,
even when death tests are disabled.

Solves #3089

[knutpett]

Seems like the appveyor build timed out on windows. How can I rerun?

[kuzkry]

Hi @knutpett, the easiest way is to pretend you touched your commit by executing:

1. git checkout working_dir_on_diskless
2. git commit --amend and immediately close an editor
3. git push --force ... onto the same branch.

[knutpett]

I keep getting "24 - googletest-param-test-test (Timeout)
ctest.exe : Errors while running CTest" from appveyor. Is introduced by me, or is this a known problem with this test timing out?

I've touched the PR twice, but I keeps on failing....

[CJ-Johnson]

My apologies for the failing tests. It's a known issue and we've had trouble resolving it. For now, all that is required is the CLA and Travis.

Would you like us to import this PR? We can do additional internal testing, but importing it implies that the PR cannot be updated further.

[knutpett]

@CJ-Johnson Yes, I'm happy with you importing this PR :)

[CJ-Johnson]

Thank you for this PR! It has now been imported for internal testing. Please do not make additional changes as they will be overwritten. (341387352)

[knutpett]

Thanks! What happens now? Does this mean that the change likely will be merged? How do I know when/if it is done? Should I close the PR?

[CJ-Johnson]

Yes, if the internal testing passes, it will be merged and closed automatically. Please do not close it. :)

[puetzk]

This caused a regression for me in finally updating our patches for winegcc (not submitted yet, trying to upstream the winegcc compiler improvements that enable it into winehq first).

UnitTest::GetInstance()->original_working_dir() is documented public API:
https://github.com/google/googletest/blob/main/docs/reference/testing.md#original_working_dir-unittestoriginal_working_dir and is used for things like UnitTestOptions::GetAbsolutePathToOutputFile. So stubbing it to just return an empty string breaks many other unit tests unnecessarily if you have a filesystem but not death tests

If there's a need to do something special for diskless systems, seems like they should have a distinct check.

[CJ-Johnson]

Thank you for the comment! Do you have a proposed path for addressing the problem?

[gaspardpetit]

I ran in to the same regression, I will propose a fix in the upcoming days. The challenge here is that it is not clear what the original issue was, and I cannot tell if I will reverse the problem back to what it originally was....

[gaspardpetit]

I undid this change in #3862
The original problem seems to be that on a system without a filesystem, FilePath::GetCurrentDir() would get called and cause getcwd to be called.

#3862 moves this logic to gtest-port.h where a custom GetCwd function can be implemented. On a diskless system, it would be possible to simply return empty string or ./.

Thus, I think #3862 solves the original problem raised here without causing the side effects observed when GTEST_HAS_DEATH_TEST is set to 0