Drop python 3.8, add 3.11 & 3.12 to testing

[charlesbluca]

With RAPIDS planning to drop python 3.8 support in 23.06 and discussion happening in dask/community#315 around Dask doing the same relatively soon, now seems like a good time to begin exploring dropping 3.8 support/testing ourselves.

[ayushdg]

A few more places we can update:

dask-sql/dask_planner/Cargo.toml

Line 16 in ab7340b

pyo3 = { version = "0.18.3", features = ["extension-module", "abi3", "abi3-py38"] }

dask-sql/setup.py

Line 42 in ab7340b

python_requires=">=3.8",

Can be done in a followup but we could remove a bunch of instances in the code for type hints with from typing import ... as it should just work py3.9 and above

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (dbb36a4) 85.55% compared to head (78df128) 85.56%.

❗ Current head 78df128 differs from pull request most recent head ae174c1. Consider uploading reports for the commit ae174c1 to get more accurate results

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1143      +/-   ##
==========================================
+ Coverage   85.55%   85.56%   +0.01%     
==========================================
  Files          77       77              
  Lines        4258     4255       -3     
  Branches      758      794      +36     
==========================================
- Hits         3643     3641       -2     
+ Misses        446      445       -1     
  Partials      169      169              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[charlesbluca]

There are some intermittent mamba env solve failures I'm trying to work through (could be worth getting a minimal example together to open an issue in setup-miniconda), but seems like in general the main blocker here is lack of mlflow 3.11 packages in conda-forge; going to try rerunning conda-forge/mlflow-feedstock#120 to see if there are any blockers there.

[charlesbluca]

In an ideal world, we would be using the full python versions here, but since manipulation of string variables in GHA is difficult I think adding this comment to make things more easily grep-able is a fair compromise

[charlesbluca]

Looks like the blockers now are lack of availability for sasl on python 3.11 (required for our Hive testing stack) and some issues with cibuildwheel requiring protoc which are unrelated to the changes here.

I'll open a separate PR to address the issues around cibuildwheel, but not really sure what to do about the Hive testing - it seems like the GH page for sasl hasn't been updated in some time, so might not be feasible to make the changes required to publish 3.11 builds, and I wasn't able to get things working without it, though I did take a start with the following patch:

diff --git a/tests/integration/test_hive.py b/tests/integration/test_hive.py
index 1a86082c..85ab763f 100644
--- a/tests/integration/test_hive.py
+++ b/tests/integration/test_hive.py
@@ -138,7 +138,7 @@ def hive_cursor():
             "IPAddress"
         ]
         port = 10000
-        cursor = sqlalchemy.create_engine(f"hive:https://{address}:{port}").connect()
+        cursor = sqlalchemy.create_engine(f"hive:https://{address}:{port}", connect_args={"auth": "NOSASL"}).connect()
 
         # Create a non-partitioned column
         cursor.execute(

[ayushdg]

Thanks for looking into this. Assuming sail cannot be patched in time for 3.11 we might want to evaluate skipping hive tests for 3.11.
For CIBW happy to sync to get the builds around protobuf working.

[sarahyurick]

I think gpuCI should pass now after merging main.

[charlesbluca]

Going to guess that the issues in the conda build are caused by the recent release of anyio=3.7; going to check if I can trivially reproduce to file an issue and unblock things on our end for now