Hi, On Tue, Dec 21, 2021 at 07:12:26PM +0100, Axel Beckert wrote: > > That seems like a false-positive: The dataclasses backport is only > > required for Python 3.6, but Debian sets "Depends: python3 (>= 3.7)" > > already. > […] > > That smells like a bug in dh-python. > > I tried downgrading dh-python until it no more FTBFS. It doesn't FTBFS > with dh-python 5.20211114 and the first version it FTBFS is 5.20211213. > > The changelog entry of 5.20211213 contains these potential causes, at > least from my point of view: > > * Handle dist-info extra sections the same way we do for requirements.txt. > (closes: 998374)
That seems to be related to the "flit" integration, a build system which qutebrowser does not use: https://flit.readthedocs.io/ > * Add new pep517 plugin that uses python3-build + python3-installer to > build > Python modules as wheels and then unpack the wheel. This is the new > interface used by modern Python packaging tools. (Closes: #984824) Could be, but it't not :) > Full entry (in cases something didn't look suspicious to me) at > https://packages.qa.debian.org/d/dh-python/news/20211214T010008Z.html > > Florian: Any idea which of these might have broken this? It's "Add support for environment markers.". Environment markers are basically conditional dependencies, e.g. for a given Python version or OS only. The crash happens here: https://salsa.debian.org/python-team/tools/dh-python/-/blob/5.20211217/dhpython/pydist.py#L522 because "dependency" is None. That's coming from here: https://salsa.debian.org/python-team/tools/dh-python/-/blob/5.20211217/dhpython/pydist.py#L516 i.e. is the return value of "guess_deps". Looking at that, we spot our error message about dataclasses: https://salsa.debian.org/python-team/tools/dh-python/-/blob/5.20211217/dhpython/pydist.py#L313-317 ...but also a strange commented out "return pname" which has been around for 7 years: https://salsa.debian.org/python-team/tools/dh-python/-/commit/de7db9bd0021cee65e477f11cd69cfd93350bd2e Due to the commented out return, the function ends and implicitly returns None. The code which crashes due to that None value is very recent: https://salsa.debian.org/python-team/tools/dh-python/-/commit/832f01c9c7da22a1d00887d8ce0e6be9893bdf3e So I suppose it would be good if someone could forward this do dh-python, I use the Debian bugtracker too seldom to know what the best way to do this would be (but if preferable, I can open a new bug). > And is the suggestion to add "dataclasses python3-dataclasses" line to > debian/py3dist-overrides something we should simply do or is this an > issue which we should make the dh-python maintainers aware of? I must > admit, I still don't understand what this is dh-python argues about. > Maybe someone has some more Python insight for a non-pythonista? Perhaps both. So what dh-python seems to be doing here is to look at the dependencies declared in a "requirements.txt" (a somewhat common way to specify dependencies for Python projects) and check whether the Debian packaging mirrors those requirements as well. Here is the line in question: https://github.com/qutebrowser/qutebrowser/blob/v2.4.0/requirements.txt#L5 So this declares that qutebrowser needs the "dataclasses" backport for Python versions older than 3.7 (it still supports Python 3.6, though that will change as soon as 3.0.0 is finished: https://github.com/qutebrowser/qutebrowser/issues/6905). However, the Debian package for qutebrowser declares "Depends: python3 (>= 3.7)", so that dependency actually is irrelevant for Debian. Should dh-python handle that case? Well, maybe, but it sure shouldn't crash. Florian -- m...@the-compiler.org | https://www.qutebrowser.org https://bruhin.software/ | https://github.com/sponsors/The-Compiler/ GPG: 916E B0C8 FD55 A072 | https://the-compiler.org/pubkey.asc I love long mails! | https://email.is-not-s.ms/
signature.asc
Description: PGP signature
