Trying to fix Mac distributable #353
Conversation
for more information, see https://pre-commit.ci
|
This includes changes in #346. |
|
@rly Is this ready for review? (some conflicts now that the other one is through) |
| - numpy # may have x64/arm64 mismatch issues so install from conda-forge | ||
| # install these from conda-forge so that dependent packages get included in the distributable | ||
| - jsonschema = 4.18.0 # installs jsonschema-specifications | ||
| - pydantic[email] = 2.0.2 # installs email-validator |
There was a problem hiding this comment.
Yes, I was wondering if the conda-forge installation pathway would resolve a lot of our challenges with this
Thanks for trying that out
There was a problem hiding this comment.
This is an alternative to pip install email_validator which is used in the win/linux envs. I don't know which one is better, but they both seem to work. Do you have a preference? We should probably be consistent with the envs as much as we can.
There was a problem hiding this comment.
I believe that conda-forge is more reliable when there are many nested (as in hidden imports) or non-Python (as in bindings) assets attached to the package
Also architecture specific releases - PyPI does a lot of that magically and you just cross your fingers but as you know, it's something you specifically setup in the feedstock (with or without certain amount of developer pains)
Though PyPI can host non-pure Python, in experience it tends to work best for pure Python code
So I'm fine with pydantic coming from conda-forge, yes. Also v2.x of Pydantic is partially written in Rust whereas v1.x is pure Python so setting up for conda-forge might be best here
[M1 Mac environment] Try neuroconv specific debug
|
Well, I must be going crazy, but at least I have evidence... Last commit on #358 (hash Last commit on this PR, following the acceptance and merging of that PR into this one, (hash |
|
Aha! The difference between the runs is whether or not the conda environment in the CI was cached (which is done for speed) So something changed from the cached versions that now means we have quite the depth do our PyInstaller spec |
|
On no! I'm glad you're onto the cause. I think I fixed the Windows / Linux window rendering issues. Just a stupid mistake because of inconsistencies between development and production hooks (#360) :( |
Fix #89