REGR: DataFrame.eval not respecting inplace argument in pandas 1.4

[simonjayhawkins]

import pandas as pd
import numpy as np

# Example for eval
pdf = pd.DataFrame({"A": range(1, 6), "B": range(10, 0, -2), "C": range(11, 16)})
pser = pdf.A
pdf.eval("A = B + C", inplace=True)
# pser != pdf.A after 1.4

Originally posted by @Yikun in #47188 (comment)

[Yikun]

Thanks!

[simonjayhawkins]

moving to 1.4.4

[CloseChoice]

first bad commit: [ 03dd698 ] BUG: DataFrame.setitem sometimes operating inplace (#43406)

cc @jbrockmendel

[jbrockmendel]

I'm unclear on the problem. we want (pser == pdf.A).all(), right? Thats what I see on mainWoops, nope.

[jbrockmendel]

In core.computation.eval L387 we do target[assigner] = ret. Changing that to

if inplace:
    target.loc[:, assigner] = ret
else:
    target[assigner] =ret

fixes the OP example. Haven't run the test suite with this, or even looked that closely at the code to see if it is reasonable.

[jorisvandenbossche]

@Yikun we are fixing this in #47550, but in general I would recommend to not rely on this too much (updating Series by performing inplace on the DataFrame). There are plans to change that behaviour (#36195), and actually also to deprecate the inplace keyword (#16529)

[Yikun]

@jorisvandenbossche Thanks for your info!