Daniel Holth
Daniel Holth
IMO finding out which packages and index data we need to perform the tests is the most interesting part.
Sure, if it always can fall back to a cache miss, then the cache will be faster or slower depending on how good it is. We also do a number...
These errors `xz-5.2.8-h8cc25b3_0\.cph_tmpw8doj7mq\info-xz-5.2.8-h8cc25b3_0.tar.zst` would be libarchive-related (conda-package-handling 1.x series) since 2.x doesn't put the `.tar.zst`in a temporary directory. I'm not sure where multiprocessing is used, the recent parallel-download-and-extract from 22.11...
I wasn't able to trigger the error except when I was actually trying to uninstall a conda dependency. I think it may have been overly pessimistic when it was written....
It's not practical to change "conda install" itself so radically that it would install from pypi "by surprise". There are several projects that try to address this idea, including https://github.com/conda-incubator/conda-pypi...
I'm seeing this error ``` % cranko stage pycosat: 200 relevant commits error: error formatting argument 'bump_spec': cannot format as string ```
Does the `conda create --no-lock` flag help?
Yes, the `local_lock=none` options suggests locking was disabled. It's tricky if conda hangs instead of raising an exception however.
@travishathaway for example, figure out which conda test is the slowest, make it faster. Or remove redundant and extra tests. Update those tests. If the test run takes < 13.5h...