CI and Sandbox: Temporarily remove use of uv

[amotl]

uv provided some hiccups in dependency resolutions, so we are trying without, in order to check if this improves the situation, or if it's unrelated.

• CI: Update to explicitly use llvmlite>=0.40 #32
• Dependencies: Explicitly use PyCaret >= 3 #35

Most prominently, the update to a higher version of pandas still does not succeed after many cycles to make it so.

• Update pandas requirement from <2.1 to <2.3 #27

[charliermarsh]

So sorry to bother but for my own prioritization, are you able to share the resolution that didn't succeed? Is it:

cratedb-toolkit[datasets]==0.0.29
ipyleaflet<0.20
kaleido<0.3
llvmlite>=0.40
pandas<2.1
pandas<2.3
plotly<5.25
pycaret>=3,<3.4
pydantic<3
refinitiv-data<1.7
sqlalchemy<2.1
sqlalchemy-cratedb==0.40.1

[amotl]

Dear Charlie,

thanks a stack for your reply, and sorry for missing it.

for my own prioritization

TLDR; Any issues on our ends have been resolved well. @zanieb educated us about astral-sh/uv#8157, but I am no longer sure this was even the problem here. Thank you both for your time and diligence on supporting us and tracking this back.

With kind regards,
Andreas.

Details

The issue has been resolved and we swapped uv back in per GH-39. The problem originated because of the constraint limit pandas<2.1, which, when resolved per GH-27, did no longer cause any issues, afair.

able to share the resolution that didn't succeed?

Because you are listing pandas<2.1, I guess your enumeration is on the right track. Please note, however:

• You are listing pandas twice, as you also included pandas<2.3.
• Please also note we added and changed constraints about llvmlite, pycaret, and kaleido while trying to resolve woes on CI, see Update kaleido requirement from <0.3 to <0.5 #29, CI: Update to explicitly use llvmlite>=0.40 #32, Revert "Update kaleido requirement from <0.3 to <0.5" #33, Dependencies: Explicitly use PyCaret >= 3 #35.

Thoughts

Apparently, this tripped us because of some resolution issue around llvmlite, the first one @konstin is enumerating within the Known Cases section at astral-sh/uv#8157. I guess it is pandas/numpy, which is pulling in llvmlite?

In this spirit, I think the constellation of dependencies can be made much smaller to be reproducable, if it's just revolving about this small subset of dependencies that goes south?

Conclusion

Wrapping up, I think the noise was more about our own sillyness than about a serious issue we observed with uv. If you think differently, and our constellation of dependencies can support you to work on astral-sh/uv#8157 in any way, please let us know any time. We can easily travel back in time of the repository history to get you a dependency dump using pipdeptree, or any other incantation of uv with relevant debug options you may advise.