Jim Pivarski

Okay, I have a minimally reproducing example that doesn't use Awkward. The reason I missed it before is because https://github.com/scikit-hep/awkward/issues/1635#issuecomment-1228804025 didn't load the extension type. The reason we need two...

Okay! This has been reported as [ARROW-17539](https://issues.apache.org/jira/browse/ARROW-17539) and since there's no action to do here (Awkward doesn't need to be changed, as far as we know), I'll close this issue....

The imports in `awkward/_v2/__init__.py` (and v1) are order-dependent: some class definitions use others (either by inheritance or by accessing constants), and the `__init__.py` file makes a set of them available...

As we discussed today, this will go after the 1.10.0 release and before the git-split.

It looks like all of these changes are generated by this one change to .pre-commit-config.yaml: ```diff @@ -19,6 +19,12 @@ repos: - id: requirements-txt-fixer - id: trailing-whitespace - repo: https://github.com/PyCQA/isort...

It's the same error in Codecov as before: ``` ##[debug]Evaluating condition for step: 'Codecov' ##[debug]Evaluating: (success() && (matrix.python-version == '3.9')) ##[debug]Evaluating And: ##[debug]..Evaluating success: ##[debug]..=> true ##[debug]..Evaluating Equal: ##[debug]....Evaluating Index:...

I don't think RaggedTensors can allow for missing values. However, if the policy is to turn missing numbers into `np.nan` and missing lists into empty lists, then UnmaskedArrays would just...

> Need to think about the C++ layer; the Forth machine returns a `ContentPtr` in a number of places. Some of these are trivial Numpy arrays, but the `bytecodes` getter...

We should be able to remove `Index` from `ForthMachine`. I'll take a look at it later, but that's a separable problem from everything else.

In the way AwkwardForth is used, the output of `toNumpyArray` and `toIndex` are used, they're immediately wrapped with `np.asarray`, so we only need one way to get a buffer out....