It is hard to be consequent with CMake for such a large and complex project
as DOLFIN. That said I think the rationale is that we only copy stuff to
the build directory if it is needed by a certain target. It is obvious that
the tests are needed for all run_foo_tests so therefor is the line:
DEPENDS copy_data_test_demo
added to the target declaration. It should be equivalent to using
add_dependencies. There are no direct requirements to have updated tests in
the build directory when dolfin is built and they are therefore not copied.
I can see that it could be convenient to have them though. I do not have
any strong opinions of copying or not copying.
Johan
On Thu, Dec 19, 2013 at 8:24 AM, Anders Logg <[email protected]> wrote:
> Yes, this is the correct list.
>
> I have experienced similar problems with not being able to edit unit
> tests and then rerun them from the build directory.
>
> I don't know enough about CMake to say whether the fix is correct.
> Johannes can comment.
>
> --
> Anders
>
>
> On Wed, Dec 18, 2013 at 10:37:46PM +0000, Myles English wrote:
> > Bah, it only 'works' now because of the 'fix' I made so my first email
> > is still correct. (By the way is this the right list?)
> >
> > On 18 December 2013 22:05, Myles English <[email protected]> wrote:
> > > Nevermind, I caused a strange error whereby there was a file called
> > > dolfin/dolfin/la/.#Matrix.h which stopped CMake working properly.
> > >
> > > Myles
> > >
> > > On 18 December 2013 15:07, Myles English <[email protected]>
> wrote:
> > >> Hello,
> > >>
> > >> If I build with cmake.local, then change the file
> > >> test/unit/la/python/Matrix.py in the source directory, then do
> > >> 'make' again in the build directory, I would have expected the
> > >> changes to be copied to the build directory, but they are not.
> > >>
> > >> If I do 'make run_quicktest' the changes are copied.
> > >>
> > >> I think this is because the initial copy was done using the
> > >> script copy_data_test_demo called from the top level
> > >> CMakeLists.txt and so CMake cannot keep track of the files involved.
> > >>
> > >> I believe this could be changed by adding an add_dependencies()
> > >> line, for example:
> > >>
> > >> if (DOLFIN_ENABLE_TESTING)
> > >>
> > >> # Add target "run_memorytests" [sic] for running memory tests
> > >> add_custom_target(copy_data_test_demo
> > >> COMMAND ${PYTHON_EXECUTABLE}
> > >> ${DOLFIN_SOURCE_DIR}/cmake/scripts/copy-test-demo-data
> > >> ${CMAKE_CURRENT_BINARY_DIR}
> > >> WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR})
> > >>
> > >> add_dependencies(dolfin copy_data_test_demo) <--- THIS LINE ADDED
> > >>
> > >> [snip]
> > >>
> > >> # Add target "run_quicktest" for running only Python unit tests
> > >> add_custom_target(run_quicktest
> > >> COMMAND ${PYTHON_EXECUTABLE} test.py --only-python
> > >> DEPENDS copy_data_test_demo
> > >> WORKING_DIRECTORY "${CMAKE_CURRENT_BINARY_DIR}/test/unit")
> > >>
> > >> endif()
> > >>
> > >> However, I notice that there are no other occurences of
> > >> add_dependencies() so I am wondering:
> > >>
> > >> 1) if the current behaviour is intentional, and also
> > >>
> > >> 2) if the change could have unintended consequences, and
> > >>
> > >> 3) what about all the other places where files are copied to the
> > >> build tree with a COMMAND?
> > >>
> > >> Thanks,
> > >> Myles
> > _______________________________________________
> > fenics mailing list
> > [email protected]
> > http://fenicsproject.org/mailman/listinfo/fenics
> _______________________________________________
> fenics mailing list
> [email protected]
> http://fenicsproject.org/mailman/listinfo/fenics
>
_______________________________________________
fenics mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics
