jsoncpp_readerwriter unit tests failed on Windows

[xkszltl]

Describe the bug

On windows, we always have 2 tests failed.

PS ********\jsoncpp\build> ctest
Test project ********/jsoncpp/build
    Start 1: jsoncpp_readerwriter
1/3 Test #1: jsoncpp_readerwriter ................***Failed    0.95 sec
    Start 2: jsoncpp_readerwriter_json_checker
2/3 Test #2: jsoncpp_readerwriter_json_checker ...***Failed    1.43 sec
    Start 3: jsoncpp_test
3/3 Test #3: jsoncpp_test ........................   Passed    0.01 sec

33% tests passed, 2 tests failed out of 3

Total Test time (real) =   2.41 sec

The following tests FAILED:
          1 - jsoncpp_readerwriter (Failed)
          2 - jsoncpp_readerwriter_json_checker (Failed)
Errors while running CTest

On Linux (CentOS 7/Ubuntu 18.04) they works fine.

Desktop (please complete the following information):

• OS: Windows Server 2019
• Ninja version 1.9.0
• CMake 3.17.2
• VS2019
• Python 3.8.2

[xkszltl]

Test log:

[cdunn2001]

You named the versions of everything except *jsoncpp itself. Why skip that one?

If this were a failure of a specific testcase, we would look into it right away. But the whole suite is broken? To debug that, we would need your help.

Also, we do not support cmake. Users do. I suggest trying meson. But if you have a fix for cmake, please submit a PR. We usually accept those pretty quickly In fact, I just merged one a few minutes ago, so perhaps the tip of master will help you. Maybe rebase your change at 58d552e onto that.

I was telling someone last week that it seems like half the Issues here relate to cmake. Drives me crazy! Anyway, please re-open if you can point to a more specific failure. Or submit a new PR, based on the current tip of `master.

[xkszltl]

@cdunn2001
jsoncpp version is the latest release tag, 1.9.2 currently.
I don't think this is cmake-related, because there's a post-build test at the end of build, which only runs jsoncpp_test and passed.
The 2 bad tests are triggered by a call to ctest afterward, so I'd like to know if they are deprecated.

[cdunn2001]

Oh, only legacy_test_array_0? are failing? Ok. That's unsettling. We always run via AppVeyor before merging.

There is currently a difference between the tests run via cmake and via meson. Fixed a race condition in meson. But I'll bet this is something else. We might need your help to debug it.

[xkszltl]

No just that. A lot of tests with legacy in name under jsoncpp_readerwriter_json_checker and jsoncpp_readerwriter, but not jsoncpp_test.
Let me know how I can help.

[xkszltl]

Ah never mind....I think I probably figure that out.
The dll isn't in the right place in build dir (probably on Linux we have rpath so everything's fine), and the test won't even start.
Once I symlink it to system32 everything's fine.

[xkszltl]

I can probably find some timeslot to help fixing the path in cmake.

[xkszltl]

[xkszltl]

jsoncpp/src/test_lib_json/CMakeLists.txt

Line 22 in 08ddeed

# First, copy the shared lib, for Microsoft.

This is why the other test can run.
No a good solution at all because on Windows usually cmake should write dll and exe to the same dir, but yes it works.