-
Notifications
You must be signed in to change notification settings - Fork 427
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
KeyError when using libmamba solver with conda-build #4991
Comments
@jaimergp Have you see this before? |
Nop. It looks like there's some kind of private mirror involved? I assume this is conda-libmamba-solver 23.7.0 too? conda-build hacks its way to build an |
It is conda-libmamba-solver 23.7.0. They get the error if they use conda-build is version 3.26.1 with libmamba solver, but the build works with classic w/ 3.26.1 conda-build. They have not yet tried libmamba with older conda-build. Not sure if this helps but:
|
I just merged a PR in conda-forge/staged-recipes that switches from That leads me to thinking that the error is triggered by something in their system. Can I get more details on:
The traceback hints at a problem with Thanks! |
@jaimergp I've added more info to the issue description above to help with debugging! |
This is happening right after the calling conda-build. |
It seems like something with channel aliases may be involved. After lots of testing with different condarc configurations, a minimal condarc where conda-build works has been found along with several minimal configurations that cause the error. This minimal condarc works:
All following condarc combinations DO NOT work.
In the following case, there is no channel alias but modified default channels
|
I've been trying to reproduce this with publicly available resources that mimic the reported configuration but I've had no luck despite matching almost everything 1:1. Details below.
FWIW, I also tried the most recent versions of every component (conda 23.9.0, conda-build 23.7.0, conda-libmamba-solver 23.9.1, libmambapy 1.5.1) and it works too. I'll keep trying other combinations but I would have expected to reproduce this more easily. |
One thing to notice is that the line raising conda-build/conda_build/render.py Lines 372 to 391 in c2bfe2b
"Vile hack" doesn't give me too much confidence and is also a "last resort" strategy, part of the
Thanks! |
@jaimergp from my experience with this bug, you should test in an isolated network environment where you can talk to a private conda-forge mirror, but nothing else. If you have general internet access, you probably won't trigger the bug. Edit: If network connectivity is indeed a factor, maybe we could integrate a tools like https://gitlab.freedesktop.org/hadess/resolvconf-override/ or https://github.com/pagekite/libunaccept into the test suite to test to ensure we test for regressions in this space. Monkey-patching the socket module is another option. |
After some fiddling it does reproduce with the proposed configuration above, but to hit the buggy code path you need a clean cache via |
What happened?
A customer has seen this traceback when using conda-build 3.26.1 with conda-libmamba-solver:
Conda Details
conda info
conda config
conda list
The text was updated successfully, but these errors were encountered: