COMPAT/BUG: Ignore numexpr < 2.0 #5858
Conversation
| @@ -79,6 +79,10 @@ class NumExprEngine(AbstractEngine): | |||
| has_neg_frac = True | |||
|
|
|||
| def __init__(self, expr): | |||
| import pandas.computation.expressions | |||
There was a problem hiding this comment.
is their a reason you can't do this at the module level
and just check a variable on the engine? (importing in a function not a gr8 idea); maybe need to move the Numexpr import check to pandas.init (like the numpy version check)
There was a problem hiding this comment.
I moved the import to the top. I think this module needs to respect _USE_NUMEXPR from computation/expressions (or I guess we could move that to pandas/computation/__init__.py)
There was a problem hiding this comment.
I take it back, I'm moving it to __init__.py of computation, used in many places.
|
@jtratner can you fixup....? |
ghost
assigned 
|
doing it now - thanks for the second ping. |
|
I'll merge this once I get it to pass - still a few kinks to work out |
|
u prob need 2 change a build maybe 2.6 |
|
wow - never seen one minor change break so many tests! |
|
something weird going on that requires going deeper, going to circle back tomorrow |
Otherwise ignore it (issue with API change pre-2.0)
in pandas/computation/__init__.py vs. previous behavior of different modules checking in different ways.
|
I'm working on it. There's a few weird test failures that I'm not grokking |
|
closing.. @cpcloud putting up a new PR |
|
okay thanks @cpcloud |
|
np! |
in both computation.expressions and computationr.engine. Issues with API
compatibility.
Fixes #5857, #5855, #5845.