BUG: df.from_records() is all kinds of broken

• the columns argument gets clobbered:
  In [13]: pd.DataFrame.from_records({1:["foo"],2:["bar"]},columns=['a','b']).columns
  Out[13]: Index([1, 2], dtype=int64)
• if index is specified, and result_index computed, it will get clobbered
  later on.
• if index is specified as a list of labels, with the first few matching columns names
  and the others not, then sdict will get mutilated, because the removal of columns
  occurs within the try clause rather then after success.
  EDIT - ignore, I misread the code.
• There's duplication against the main Dataframe ctor which also accepts dicts and arrays,
  and The exclusion and float coercsion which are unique to from_records() would be useful
  to have generally available in the main ctor anyway.
• It's unclear if the columns argument should be specified relative to the original
  data, or relative to the data modulo excluded columns
• Doesn't support duplicate column names (Although that's checked with
  a warning)
• the docstring specifies the datatypes for data , which do not include a dict.
  But the code specifically checks and handles dict input, This is a minor thing, but
  using columns along with dict is not well defined because of key (non-)ordering,
  so the original docstring spec seems more sane.

[wesm]

Indeed this code has been fairly neglected. I'll see if I can easily fix some of the issues you raised

[wesm]

A major issue here is that the index argument is overloaded to take either index field names or an array to be used as the index. This was a pretty bone-headed decision made a long time ago

agreed, that was a bit too flexible. But if the loop for removing the columns
gets put in an else caluse to the try/except that's taken care of.

I couldn't figure out a way to get the main ctor to plug data in as columns rather
then rows after sorting out the rest.

[wesm]

I'm adding a docs clarification about what columns means (i.e: renaming columns is not possible with this function at the moment) and leaving this issue open for more cleanup at a later time (and fixing the bug from the first bullet point).

I'm pretty sure if you move the if/else setting result_index above
the check on index , you can close the 2rd as well.

[wesm]

Test case?

no need, I see you've removed the if entirely.

[wesm]

What still needs to get fixed here?

support for duplicate columns is still missing. other then that I think it's good.