ENB: support f-strings in pd.eval

[dss010101]

Pandas version checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• I have confirmed this bug exists on the main branch of pandas.

Reproducible Example

import pandas as pd, numpy as np
import timeit
class Data:
    def __init__(self, data, tag = 'xxx') -> None:
        self.data = data
        self.tag1 =  'xxx'
        
df = pd.DataFrame(np.random.randint(0,10,100).astype(str).reshape(50,2),columns=list('AB'))
d = Data(data=df)

eval("data.A + f':{tag1}'", d.__dict__)

pd.eval("data.A + f':{tag1}'", resolvers=(d.__dict__,))

Issue Description

In the above, eval works fine. However pd.eval throws the following exception

AttributeError: 'PandasExprVisitor' object has no attribute 'visit_JoinedStr'

Expected Behavior

If eval is able to evaluate this correctly, then pandas.eval should i feel.

For context, the reason for code like this, is that f strings dont seem to work with pandas series. so i cant do something like:

Installed Versions

Replace this line with the output of pd.show_versions()

[topper-123]

We haven't implemented f-string in pd.eval. Couldn't you add the xxx variable directly into the string?

[dss010101]

We haven't implemented f-string in pd.eval. Couldn't you add the xxx variable directly into the string?

yes ofcourse. but this example is greatly simplified to demonstrate the issue. the actual use case is a more generic piece of code that deals with expression evaluation for inputs may be python popo's or pandas dataframe

[topper-123]

The string inside pd.eval isn't normal python code like the plan eval, because it supports e.g. @ and backslashes. Can you give an example it wouldn't work/be inconvenient with current setup?

[dss010101]

IIt's been a while since i opened the ticket so i can not recall exactly the use cases that was failing...
but we have configuration file that defines calculated fields by using an expression. the expression may refer to datasets + columns (i.e. df.A above), or other variables, either single variables/constants or variables that can be object properties or even function calls.
The scope of known dataset is a python class representing the data model or domain - so you can build an expression by using any properties or methods on that data model. Again those properties may be single values, constants or timeseries (dataframes).

an example would be something like:

{
    "calc_field1",  
     expression: "{port.spread} * {port.mkt_val} + ( {rolling_avg(200, rate)} / 100)"

    "calculed_fiedl2",  
     expression: "{port.name}_batch_{batch__id}"
}

i assume everything in curly braces can be evaluated against the data model instance, and everything outside is simply a constant. so after little parsing/massaging (if need be), i generate a string to be evaluated. in above made up example, 'port' is dataframe on the object

[topper-123]

I think your example is geared towards str.format and passing that to python's eval (which certainly can be a fine strategy, nothing wrong with that).

A setup using pd.eval would end up like something like pd.eval("port.spread * port.mkt_val + (rolling_avg(200, rate) / 100)", local_dict=dict(port=df, rolling_avg=rolling_avg, rate=rate)) or similar.

I think passing variables to local_dict is the way to go for pd.eval and should give the needed end result.