ENH: the error behaviour in pandas operation should be consistent rename erros are ignore by default whereas drop erros are raise

[Furqan-s]

Feature Type

• Adding new functionality to pandas

• Changing existing functionality in pandas

• Removing existing functionality in pandas

Problem Description

Feature Description

can we make the pandas function consistent in how they raise errors drop should be default to ignore and not likely to break.

Alternative Solutions

def drop(labels=None, axis=0, index=None, columns=None, Level=None, inplace=false, errors='ignore'):

Additional Context

No response

[Gwalib]

Hello!! Well, I noticed that, this leads to unexpected errors for users when dropping non-existent columns or index labels, even though renaming silently skips missing keys.

We can either change the default:

i.e. errors='raise' to errors='ignore' in DataFrame.drop():

Or we can add a warning, if this causes inconsistency.
I would like to work on this issue! Let me know if it’s okay for me to work on it, and whether changing the default behavior of errors in DataFrame.drop() is acceptable from a compatibility perspective.

[petern48]

I prefer to set rename's default to errors="raise" instead. I was debugging my code for a while with zero idea that the problem was in rename. When I set errors="raise" manually, it instantly made it clear that the bug was that I was using rename incorrectly. I made the same mistake as the issue I tagged above, where I didn't specify "columns" in the arguments, so it was silently trying to rename indexes without telling me. I feel like it's a very easy to make that mistake since it's feels more natural to rename columns instead of indexes.

[Gwalib]

Thanks!! It helped me understand both sides of this.

You're absolutely right that silently passing in rename can definitely hide bugs, especially when someone forgets to set axis='columns'. At the same time, I also find .drop()'s errors='raise' default a bit too aggressive, especially in exploratory or dynamic settings.

So this feels like a bigger API consistency discussion.

So, what would make sense now?->

• Keep current behavior for now (to avoid breaking users)
• Add a warning when rename() is used without matching keys, encouraging users to explicitly set errors
• Possibly align both functions in a future major version (e.g., pandas 3.0)

I would like to assist with creating a warning message or updating the doc regarding this!