How can I deal with the case where I have a depend...
# general
b
How can I deal with the case where I have a dependency that forgot to add a build requirement to its build-requires, e.g., forgetting to add torch and expects --no-build-isolation alternatively? (other than forking it and adding the dep to build-requires myself)
b
Forking is exactly the right answer! You can use the fix immediately and push a PR async to help them - and everyone else in your situation - out.
Just use a VCS dep on your fork.
b
That's exactly what I'm doing 🙂 . I think we're gonna need to vendor a lot of libs at this rate lol. Probably for the best anyway
Thanks
b
Your welcome. And there is no other way currently that I'm aware of.
😞 1
Its not much of a sad face though, at least in terms of perf usually. If you're already building an sdist, usually the clone vs download sdist tarball isn't a big latency factor compared to the build itself. You've already lost if you need to build.
b
Ah. Anyway, I looked into open prs in the repo I'm staring at and they explicitly want
--no-build-isolation
since torch has all the cuda-specific variants.
b
Ok, then the only solution I know of is you need to pre-build wheels and host.
Lots of mid / big shops due this for a host of other reasons as a matter of policy / principle.
b
We used to do that too.
We dont need to anymore with this specific one for weird reasons (as I've hinted at in other threads). Pretty sure I can remove this
python_requirement
in a week or two actually.
b
K