adding new row to frame breaks types

[behzadnouri]

xref #8596

>>> df
   a
0  1
>>> df.dtypes
a    int64
dtype: object
>>> df.loc[1, 'a'] = 2
>>> df
   a
0  1
1  2
>>> df.dtypes
a    float64
dtype: object

in an intermediate step nan is inserted, and the data are casted to floats

[jreback]

type conversion is not done here. as it can be expensive. imagine doing this repeatedly (not recommended anyhow).

This is a side-effect of this type of enlargement. a user can use concat if they want type preservation. However, you can fix the impl to avoid the 2-step creation with missing then filling.

Is that the intention here? as an enhnacement?

[behzadnouri]

agreed that a code change to handle this should not be expensive in terms of performance cost; added the issue just because #8596 is a similar issue, so there may be a solution which works for both cases.

[goretkin]

can concat guarantee that the new row is added in-place?

[shoyer]

can concat guarantee that the new row is added in-place?

concat never updates in-place.

[jorisvandenbossche]

This actually seems a duplicate of #6485 (BUG/ENH: dtype change on enlargement).