DEPR: read_csv keywords: keep_date_col, delim_whitespace

[jbrockmendel]

Issue Description

• delim_whitespace: "Equivalent to setting sep='\\s+'" -> seems redundant
• dialect: does anyone use this? looks like it overrides several other keywords
• cache_dates: once ENH/API: resolution inference in vectorized datetime parsing #55564 is done, this option should not have any behavior impact, so seems unnecessary (also in to_datetime)
• parse_dates: not the entire keyword, just the list-of-list and dict-of-list variants. seems like users can handle those cases in post-processing.
• keep_date_col: only relevant in the parse_dates cases above.

Others? ive never used verbose, na_filter, keep_default_na, or na_values.

[lithomas1]

+1 on delim_whitespace, cache_dates, verbose, na_filter, keep_default_na.
-1 on na_values (I think some people use it at least)

I also don't think keeping dialect hurts to much either.

[jbrockmendel]

I could also be convinced to deprecate converters. seems just as easy to do that in post-processing

[rmhowe425]

take

[rmhowe425]

@lithomas1 @jbrockmendel Can I file a PR for this issue?

It looks like the Needs Triage tag has been removed, but we're still deciding on which parameters we'll be deprecating?

[rmhowe425]

Also thinking it would make sense to create a new decorator in pandas.util._decorators.py to handle the deprecation of multiple keyword arguments. That way we can minimize the number of conditional statements in the actual implementation for read_csv.

I already see that there is an IF statement being used to handle the deprecation for infer_datetime_format.

[lithomas1]

You're free to deprecate everything that we agreed on.

For doing many deprecations, it might be worth studying
https://github.com/pandas-dev/pandas/pull/48849/files#diff-c2c56de582ca256e79e43c1ef22dadfe0628b713bba932b1f599458105c5bc84L497-L508

This was the old deprecation machinery for read_csv.

[rmhowe425]

@lithomas1 thx.

The link provided appears to be the final, formal deprecation of certain features of read_csv. For this particular PR, we would want to alert users of upcoming changes by leveraging calls to warnings.warn wouldn't we?

Something similar to this: https://github.com/pandas-dev/pandas/pull/53710/files#diff-1a2e3df0db7dd8bddc2ec4bff9de8a7a55e328e6c32e2cecde761dc9549fcd46R9351