[roottest] Fix roottest trying to invoke /usr/bin/root.exe in tests.

[hageboeck]

When ROOT is installed in the system, roottest would try invoking /usr/bin/root.exe, leading to failures when trying to combine .pcms from two different ROOT builds.

This was caused by a find_package(ROOT) and find_program(root-config) at the start of the roottest config, which find the system ROOT before the ROOT being configured. With roottest being part of the ROOT build, any find_package(ROOT) and discovery of root-config and ROOTSYS could be removed, since these variables are anyway known already in the running CMake configuration.

This entailed a move from if(ROOT_<component>_FOUND) to if(<component>), since the former is only set by find_package(ROOT) whereas the latter always available. As a side effect of this move, 4 tests enabled themselves, because the ROOT_xxx_FOUND apparently didn't propagate correctly to all locations in roottest. These are:

• roottest-python-JupyROOT-cppcompleter_doctest

• roottest-python-JupyROOT-handlers_doctest

• roottest-python-JupyROOT-utils_doctest

• gtest-roofit-roofit-vectorisedPDFs-testLegendre

• Wait for [cmake] Simplify logic disabling runPyClassTest.C on Windows #20830 to be merged

[github-actions]

Test Results

    22 files      22 suites   3d 16h 23m 1s ⏱️
 3 794 tests  3 792 ✅ 0 💤 2 ❌
80 380 runs  80 378 ✅ 0 💤 2 ❌

For more details on these failures, see this check.

Results for commit 58a965d.

♻️ This comment has been updated with latest results.

[guitargeek]

Thank you very much, great initiative!

There is however something that needs to happen at the same time as a corollary. The find_package(ROOT) sets these ROOT feature variables like ROOT_pyroot_FOUND, which are extensively used in the tests (for example here).

Now that these variables are not set, the features are not detected and many tests are skipped, like the linked roottest-python-pythonizations-pythonizations. You can compare your CI runs with the nightlies and count the number of tests, and then the extent of this gets clear. For example, in this PR, 3706 test ran on alma10, and in the last nightlies there were 3742.

What you need to do is the same PR (or maybe another one that we merge before) is to replace all these feature variables with the CMake variables that ROOT sets at build time, like pyroot.

[hageboeck]

@guitargeek

Now that these variables are not set, the features are not detected and many tests are skipped, like the linked roottest-python-pythonizations-pythonizations. You can compare your CI runs with the nightlies and count the number of tests, and then the extent of this gets clear. For example, in this PR, 3706 test ran on alma10, and in the last nightlies there were 3742.

I raise to 3746. I noticed this also locally, since there were subfolders where the (previously plain and local) CMake variables didn't get propagated correctly. Now, since the ifs in rootest rely on global CACHE variables, they are visible in all subdirectories.

[guitargeek]

Nice, thanks for this! Looks good now

[hageboeck]

Nice, thanks for this! Looks good now

OK, needs one more iteration, though. The following ifs don't do what we expected on Windows:

root/roottest/python/cling/CMakeLists.txt

Lines 11 to 18 in 6ecf2d7

	if(NOT MSVC OR CMAKE_SIZEOF_VOID_P EQUAL 8 OR llvm13_broken_tests)
	if(NOT MSVC OR "${CMAKE_GENERATOR_PLATFORM}" MATCHES "Win32" OR win_broken_tests)
	ROOTTEST_ADD_TEST(class MACRO
	runPyClassTest.C
	WORKING_DIR ${CMAKE_CURRENT_SOURCE_DIR}
	OUTREF PyClassTest.ref)
	endif()
	endif()

It's understood now. The logic will be updated in #20830