ENH: Support using repr() instead of str() to print non-numeric Series elements, especially strings

[gwerbin]

Is your feature request related to a problem?

Sometimes, it is useful for debugging to be able to see the contents of a Series with repr() instead of str() applied to each element.

Specifically with strings, it is often useful to see non-ASCII Unicode characters escaped, and to see the surrouning quotation marks, to detect whitespace at the start and end.

Another use case might be some situation where you have other nontrivial objects in a Series, and their str representations are somehow less useful than their repr representations.

For example, print(pd.Series(['a', 'b', 'c'], dtype='string')) yields:

0    a
1    b
2    c
dtype: string

But it would be helpful if it could yield:

0    'a'
1    'b'
2    'c'
dtype: string

Describe the solution you'd like

One possibility is a new module option, maybe display.string_as_repr (just for strings) or display.print_as_repr (for all non-numeric entries), to enable this behavior.

API breaking implications

None that I know of.

Describe alternatives you've considered

You could use .apply(repr) or similar, to see this output in an ad-hoc way.

Additional context

This was originally a StackOverflow question: https://stackoverflow.com/q/71832297/2954547

[rhshadrach]

Thanks for the report! I wonder what the downsides would be in making this default behavior.

ser = pd.Series(['  a', 'a', ' a'])
print(ser)
print(ser.apply(repr))

0      a
1      a
2      a
dtype: object
0    '  a'
1      'a'
2     ' a'
dtype: object

I could see trying to perform logic while not realizing there are space characters giving rise to issues.

[gwerbin]

Something else I thought of: perhaps this could also be added as a parameter to DataFrame.to_string? Not sure what the core devs' opinions are on this kind of thing.

I posted a closely related feature request in #46755, where I discussed some more-general solutions that also might subsume this request.

But yes, I think changing the default to repr would be great for a lot of purposes. I just don't know if there are adverse implications.

[MarcoGorelli]

I'm not keen on adding an extra parameter (.apply(repr) should work fine), but changing the default sounds nice - I've added that to the agenda for this month's dev meeting

[rhshadrach]

Agreed on not adding a parameter. Looking at this again, I think using repr as the default would be confusing. When one does print(obj), the __str__ method, not __repr__, is called. It seems like we'd be going against this standard.

On the other hand, repr on a Series still calls __str__:

print(repr(pd.Series(['  a', 'a', ' a'])))
# 0      a
# 1      a
# 2      a
# dtype: object

and that looks like a bug to me.

[EkinsMatthew]

Hi all, wondering where we are on this. I see from the related post #50194 it was brought up at the Dev meeting. I too would really appreciate repr being the default display method for strings.

Here is a link to a stackoverflow post I made a yesterday showing a good example of how this makes analysis on string columns with escaped characters very difficult.

[rhshadrach]

It does seem like Python and NumPy both use repr in containers consistently.

class Foo:
    def __str__(self):
        return 'str'

    def __repr__(self):
        return 'repr'

foo = Foo()

print(foo)
# str

print([foo])
# [repr]

print((foo,))
# (repr,)

print({foo: foo})
# {repr: repr}

print(np.array([foo]))
# [repr]

print(pd.Series([foo]))
# 0    str
# dtype: object

This suggests to me pandas should do the same, but I wonder what issues (especially for downstream libraries that extend pandas) this might cause.

cc @MarcoGorelli @mroeschke

[mroeschke]

I think this is a sensible change. Not sure how easy it is to migrate from str to repr though. Maybe it's a candidate for a 3.0 breaking change?

[rhshadrach]

The only other way I see doing this is via an option. +1 on making this a breaking change.