feat(python): add to_repr methods to DataFrame and Series

[ghuls]

No description provided.

[ghuls]

Closes: #7732

@alexander-beedie I also added missingpl.from_repr documentation link.

[alexander-beedie]

I'm trying to come up with a different name for to_repr, as it's not the actual repr.

Something like df.to_init_repr() or pl.to_init_repr(df) instead?

[ritchie46]

I'm trying to come up with a different name for "to_repr", as it's not the actual repr.

Yes, I agree. I think to_init_repr is better. 👍

Really cool feature @ghuls.

[ghuls]

I'm trying to come up with a different name for to_repr, as it's not the actual repr.

Something like df.to_init_repr() or pl.to_init_repr(df) instead?

I am fine with changing the name, but can you explain me why it is not the actual repr?

As far as I know __repr__ functions should construct a string that can reinstate the object.

import numpy as np

In [9]: a = np.array([[48984, 4894], [4568, 48968], [468135, 49849]])

In [10]: a.__repr__()
Out[10]: 'array([[ 48984,   4894],\n       [  4568,  48968],\n       [468135,  49849]])'

In [11]: print(a.__repr__())
array([[ 48984,   4894],
       [  4568,  48968],
       [468135,  49849]])

In [13]: from numpy import array

In [14]: eval(a.__repr__())
Out[14]: 
array([[ 48984,   4894],
       [  4568,  48968],
       [468135,  49849]])

Do you plan to find another name for pl.from_repr too?

[alexander-beedie]

I am fine with changing the name, but can you explain me why it is not the actual repr?

The short version would be: if it's not what you get back from calling repr(obj), then it's not the object's repr.

As far as I know __repr__ functions should construct a string that can reinstate the object.

The result from an object's repr is defined specifically as "a string containing a printable representation of an object¹." So, something useful/informative that can reasonably be returned in an interactive console, notebook, logs, etc.

Now, for many types it is also a good idea to make the repr usable as if you could eval/init it (and on any smaller-scale custom types I develop at work I make the effort to do so); also from the python docs: "for many types, this function makes an attempt to return a string that would yield an object with the same value when passed to eval()" (but note the use of "many", not "all", and "makes an attempt").

Objects like DataFrames (or anything else that fronts large data) don't usually return eval-style reprs because anything other than a trivial frame would return an absurdly large string, rendering it unprintable and/or less useful as the "official" representation of the given object/data. Hence Polars doesn't, Pandas doesn't, PyArrow doesn't, DB interfaces don't, etc. The general trend in these cases is to have a partial table representation that provides a useful sense of the object (its size, the schema, some data, and so on).

Do you plan to find another name for pl.from_repr too?

No - because from_repr does parse the DataFrame repr, as advertised ;)
(And should probably handle the Series repr too, in the near future 💭)

Footnotes

1. https://docs.python.org/3/library/functions.html#repr ↩

[ghuls]

The result from an object's repr is defined specifically as "a string containing a printable representation of an object1." So, something useful/informative that can reasonably be returned in an interactive console, notebook, logs, etc.

I thought that __str__ was meant for that. (although for a lot of objects __str__ and __repr__ is the same).

Changed name to to_init_repr.