roottest should not pollute source directory with build artifacts

[guitargeek]

Description

Migrating the following Jira ticket to give it more visibility:
https://its.cern.ch/jira/browse/ROOT-9972

Since roottest was merged into the main ROOT repo, it's not so painful to work on these kind of tickets.

Also, I got bitten by this problem 2 days ago when compiling and testing the same ROOT source twice, to compare different Python versions.

Reproducer

• build ROOT with testing=ON and roottest=ON
• run ctest -j<N> . in the root_build/roottest directory

You will see that while running the tests, artifacts will come and go in your source directory, like this lock and ACLiC stuff:

        roottest/root/io/event/bigeventTest.write.lock/
	roottest/cling/reflex/iotype_rflxdriver_ACLiC_dict.cxx
	roottest/cling/reflex/iotype_rflxdriver_ACLiC_linkdef.h
	roottest/cling/reflex/iotype_rflxdriver_ACLiC_mainC
	roottest/cling/reflex/iotype_rflxdriver_ACLiC_map.in
	roottest/cling/reflex/iotype_rflxdriver_ACLiC_map.out
	roottest/cling/template/separateDictNamespace/libslave2_ACLiC_dict.cxx
	roottest/cling/template/separateDictNamespace/libslave2_ACLiC_linkdef.h
	roottest/cling/template/separateDictNamespace/libslave2_ACLiC_mainC
	roottest/cling/template/separateDictNamespace/libslave2_ACLiC_map.in
	roottest/cling/template/separateDictNamespace/libslave2_ACLiC_map.out

It's hard to get the full list, because these files get created and deletes as the tests run. I think it's only locks and ACLiC though. But this should all happen in the build directory, such that multiple builds from the same source can be concurrently tested.
dsaf

[linev]

All Makefile-based tests running is source directories.
Therefore first all tests must be migrated to cmake.
And even then - some ROOTTEST_<SOMETHING> cmake macros may create extra files in source directories.

Like problem resolved here: #19054

It is very long way to go

[dpiparo]

@linev according to your analysis, is makefile-driven testing the only source of artifacts? If yes, it's good news. Of course I agree with you on the amount of work the migration require. Let me underline that migrating tests allows us to re-discover them, update them and ultimately improve our coverage and ROOT overall stability. The effort in migrating is very, very well spent.

[linev]

Another source of polluting source directory is usage of:

PRECMD ${ROOT_root_CMD} -b -q -l ${CMAKE_CURRENT_SOURCE_DIR}/def.C+

Trying to run macro with + sign will create library in source directory.
Need to be changed with normal macro invocation and fixtures dependency

[linev]

One more problem is ROOT_ADD_PYUNITTEST. Running python script in source directory creates __pycache__ subdir.
One possible solution - copy script to build directory first.

[linev]

One more reason for creating files in source directory is running of ${ROOT_genreflex_CMD} like in root/io/uniquePointer tests. This create *.cpp and *rdict.pcm files in source directory

One have to provide '-o ${CMAKE_CURRENT_BINARY_DIR}/filename.cpp` argument to be sure that files created in build directory