Add arro3 to package list

[kylebarron]

Description

arro3 is a new Python library for working with Apache Arrow data. See documentation.

arro3 is distributed with namespace packaging. This means that users who only need to represent Arrow data (say as interop between Arrow-based Python packages) only need the core module. So this PR creates three new recipes: arro3-core, arro3-io, and arro3-compute.

cc @hoodmane who suggested contributing this.

I'm expecting to make an arro3 0.3 release in the next few days and will bump these versions to 0.3 when that's released. (It can either be this or a follow up PR; I saw the 0.27.0a1 tag was published and figured I should make this PR sooner than later.)

Checklists

• Add a CHANGELOG entry
• Add / update tests
• Add new / update outdated documentation

[hoodmane]

It'll be a few more weeks at least before we release 0.27 I think so no huge rush.

[hoodmane]

Thanks @kylebarron!

[agriyakhetarpal]

Not a blocker, but it would be great if you could add licensing fields to the recipes too.

[kylebarron]

I added license fields and updated to arro3 0.3.0.

[hoodmane]

CI is failing because top-level isn't supposed to contain a dot. Either these should be top-level: [arro3] or we need to fix find_imports. @ryanking13 perhaps it would make sense to change find_imports to support this:

+def _add_prefixes(s: set[str], mod: str) -> None:
+    [current, *rest] = mod.split(".")
+    s.add(current)
+    for part in rest:
+        current += f".{part}"
+        s.add(current)
+
+
 def find_imports(source: str) -> list[str]:
    source = dedent(source)
    try:
         mod = ast.parse(source)
     except SyntaxError:
         return []
-    imports = set()
+    imports: set[str] = set()
     for node in ast.walk(mod):
         if isinstance(node, ast.Import):
             for name in node.names:
                 node_name = name.name
-                imports.add(node_name.split(".")[0])
+                _add_prefixes(imports, node_name)
         elif isinstance(node, ast.ImportFrom):
             module_name = node.module
             if module_name is None:
                 continue
-            imports.add(module_name.split(".")[0])
+            _add_prefixes(imports, module_name)
     return list(sorted(imports))

[kylebarron]

I really like namespace packaging but it does tend to break tools 😅

Either these should be top-level: [arro3]

For a namespace module arro3.core, import arro3 will do nothing because there's no __init__.py file in the arro3 directory. It's not until you reach arro3/core/__init__.py that there's anything to import.

[ryanking13]

Yes, I think it makes sense to change find_imports. Changing top-level: [arro3] would result in importing arro3 loading all three arro3 subpackages, which might not be the intended behavior.

[kylebarron]

Changing top-level: [arro3] would result in importing arro3 loading all three arro3 subpackages

No, with namespace packaging, calling import arro3 does nothing. No code is loaded until importing one of the submodules:

In [1]: import arro3

In [2]: arro3.core.Array
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
Cell In[2], line 1
----> 1 arro3.core.Array

AttributeError: module 'arro3' has no attribute 'core'

In [3]: import arro3.core

In [4]: arro3.core.Array
Out[4]: arro3.core._core.Array

[hoodmane]

Looks good @kylebarron, please add a changelog entry and then we will merge.

[kylebarron]

Added to changelog!

[agriyakhetarpal]

Thanks, @kylebarron!