ENH: Allow update to use an on keyword. Allow one to many update.

[jseabold]

Scratch an itch I had. Use case

df = DataFrame([[np.nan, 'A'],
                [np.nan, 'A'],
                [np.nan, 'A'],
                [1.5, 'B'],
                [2.2, 'C'],
                [3.1, 'C'],
                [1.2,' B']], columns=['number', 'name'])

df2 = pd.DataFrame([[3.5, 'A']], columns=['number', 'name'])

df.update(df2, on='name')

The tests fail though because when you reset the index, the column order is not preserved and self = self[col_order] doesn't seem to do what I'd expect it to do. Maybe there's a better way to go about doing all of this?

Also, I dropped reindex_like because you were just iterating through columns of NaNs in other.

[jreback]

can you put the output you are expecting up as well?

[jseabold]

NaNs "updated" to 3.5.

[jreback]

In [10]: df.loc[df.name.isin(df2.name)] = df2

In [11]: df
Out[11]: 
   number name
0     3.5    A
1     3.5    A
2     3.5    A
3     1.5    B
4     2.2    C
5     3.1    C
6     1.2    B

[jseabold]

This is just a toy case. What if on is multiple columns? This also only updates where there are NaNs in df. There might be overlap in df and df2 and this sanity checks that.

[jreback]

true.....update really needs to be rewritten actually (and pushed into internals).....but that's a bit more drastic....

shouldn't you merge these then set?

[jseabold]

Sounds like half a dozen of one to me... Merge, conditionally set _x for each column then delete _y and rename everything?

[jreback]

moving a bit away from adding things to .update and stale