Hello,

I am attempting to package the latest upstream version of "trash-cli"¹,
which differs quite a lot from the current package on the build side of
things.
https://github.com/andreafrancia/trash-cli

The problem occurs during the post-build test phase. The tests cannot
find the scripts they are trying to test. The build appears to be a
standard python setuptools/distutils thing and the packaging is a
standard dh --with python3 --buildsystem=pybuild. The Build runs

/usr/bin/python3 setup.py build

Which outputs the following (elided)

running build_py
creating /<<PKGBUILDDIR>>/.pybuild/cpython3_3.9/build/trashcli
copying trashcli/fstab.py -> /<<PKGBUILDDIR>>/.pybuild/cpython3_3.9/build/trashcli

...

So for the build_py phase, the destination for copied files is rooted
under <<PKGBUILDDIR>>. The next build phase handles scripts:

running build_scripts
creating build
creating build/scripts-3.9
copying and adjusting trash -> build/scripts-3.9

...

Here, the scripts are copied to a path *not* rooted in <<PKGBUILDDIR>>.
So far as I can tell, both phases are normal parts of the
setuptools/distutils build lifecycle, and I haven't spotted anything
obviously non-standard about the way that is specified.

Where is it that the build_py phase's path are specified or overridden
such that they are correctly rooted under <<PKGBUILDDIR>>, and why isn't
the same true of the build_scripts phase?

I've pushed my WIP packaging to a temporary branch. https://salsa.debian.org/debian/trash-cli/-/tree/jmtd-update-0.21.7.24
A BTS bug requesting the new version with some back-and-forth trying to
figure this out is
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=988597

Thank you in advance for any tips you can offer!


Best wishes,

--

👱🏻 Jonathan Dowland
✎ jmtd@debian.org
🔗 https://jmtd.net

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

Hi Jonathan,

* Jonathan Dowland <jmtd@debian.org> [2021-10-18 11:04]:

Here, the scripts are copied to a path *not* rooted in <<PKGBUILDDIR>>.
So far as I can tell, both phases are normal parts of the >setuptools/distutils build lifecycle, and I haven't spotted anything >obviously non-standard about the way that is specified.
[...]
A BTS bug requesting the new version with some back-and-forth trying to >figure this out is
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=988597

The main issue is the implicit upstream assumption that the tests
are run from the source root dir; I've had a similar problem with
tweepy and my solution was to make sure that the tests find
everything they need in the build folder, i.e. add the following
snippet to d/rules:

export PYBUILD_BEFORE_TEST=cp {dir}/trash-* {dir}/script/bump.py {build_dir} export PYBUILD_AFTER_TEST=rm {build_dir}/trash-* {build_dir}/bump.py

The second line is needed so the extraneous stuff is not installed
with the built package. It's not very pretty, and hopefully someone
else has a better solution, but it gets the job done.

You'll also have to patch tests/run_command.py to run python3 and
add python3-psutils to your Build-Depends.


diff --git a/tests/run_command.py b/tests/run_command.py
index 2e31e11..746cdad 100644
--- a/tests/run_command.py
+++ b/tests/run_command.py
@@ -17,7 +17,7 @@ def run_command(cwd, command, args=None, input='', env=None):
if args == None:
args = []
command_full_path = os.path.join(base_dir, command)
- process = subprocess.Popen(["python", command_full_path] + args,
+ process = subprocess.Popen(["python3", command_full_path] + args,
stdin=subprocess.PIPE,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE,


Cheers
Timo

PS. I worked around the missing fork
PPS. I thought about patching run_command.py to use the scripts from
the source directory, but that will use the trashcli module from the
source tree, not from the build tree, and contaminate the source tree
with __pycache__ folders.

--
⢀⣴⠾⠻⢶⣦⠀ ╭───────────────────────────────────────────�