Floris Bruynooghe writes: > > It is possible to run this without installing, but it does need a build > step since cffi (in the mode used - which is the recommended mode) needs > to build an extension module. I did something like this, using my > debian testing system-installed python > > $ export PYTHONPATH=$(pwd)/bindings/python-cffi > $ pushd bindings/python-cffi > $ python3 notdb/_build.py # creates notdb/_capi.cpython-37m-x86_64-linux-gnu.so > $ popd > $ pushd test > $ ./T391-pytest.sh Yes, I think I arrived at a similar place, except 1) using "python3 setup.py --build-lib build/stage" to build. I'm not sure which is better, I think it will depend a bit on when we try to get out of tree builds working. It is a bit nicer to have the build output out of tree, but then I have to copy the tests. 2) instead of changing PYTHONPATH, use "python3 -m pytest", which picks up the module in the current directory. > Does that more or less work? One problem with this is that it will pick > up the system-wide installed notmuch though. I guess the way to change > this is by tweaking CFLAGS=-I... LDFLAGS=-L... or so when building? But > than you also have the whole RPATH/LD_LIBRARY_PATH stuff going on as > well. Does notmuch abstract any of this away already for it's test > suite? The LD_LIBRARY_PATH is already set by the test harness, as is PATH (to find notmuch). It looks like your function notmuch is not respecting PATH (see attached log). if I hack something like diff --git a/bindings/python-cffi/tests/conftest.py b/bindings/python-cffi/tests/conftest.py index 1b7bbc35..ac17397c 100644 --- a/bindings/python-cffi/tests/conftest.py +++ b/bindings/python-cffi/tests/conftest.py @@ -31,7 +31,7 @@ def notmuch(maildir): accidentally do this in the unittests. """ cfg_fname = maildir.path / 'notmuch-config' - cmd = ['notmuch'] + list(args) + cmd = ['../../notmuch'] + list(args) print('Invoking: {}'.format(' '.join(cmd))) proc = subprocess.run(cmd, then the tests pass, but this is obviously not a good solution.