fix: Scalar typetracer after reduction #529
Conversation
| @@ -532,6 +530,11 @@ def f(self, other): | |||
| meta = op(other, self._meta) | |||
| else: | |||
| meta = op(self._meta, other) | |||
| if meta.ndim: | |||
There was a problem hiding this comment.
This applies to the scenario, which apparently we hadn't anticipated, that an operation on a scalar produces an array.
|
Codecov ReportAttention: Patch coverage is
❗ Your organization needs to install the Codecov GitHub app to enable full functionality. Additional details and impacted files@@ Coverage Diff @@
## main #529 +/- ##
==========================================
- Coverage 93.06% 92.91% -0.16%
==========================================
Files 23 23
Lines 3290 3389 +99
==========================================
+ Hits 3062 3149 +87
- Misses 228 240 +12 ☔ View full report in Codecov by Sentry. |
| @@ -570,6 +573,15 @@ def f(*args): | |||
| args = tuple( | |||
| ak.Array(arg.content) if isinstance(arg, MaybeNone) else arg for arg in args | |||
| ) | |||
| args = tuple( | |||
There was a problem hiding this comment.
Here we account for getting the low-level typetracer through during column optimisation. This is the questionable part.
martindurant
changed the title
Fixes #528
This appears to fix the linked issue, but I'm not really sure why a typetracer object is being passed around rather than a high-level Array.