Code Sample, a copy-pastable example if possible

d = pd.DataFrame()
try:
   d.set_value(0, "c", [1,2,4]) 
except Exception:
  print "Caught"

print d  # c is now in d

Problem description

The above call set_value (correctly) fails and throw an exception. However, the dataframe has been modified. It has a new 'c' column of type float. Because the operation failed, it probably should have remained unchanged.

Expected Output

print d should print an empty dataframe.

Output of pd.show_versions()

INSTALLED VERSIONS ------------------ commit: None python: 2.7.6.final.0 python-bits: 64 OS: Linux OS-release: 4.2.0-41-generic machine: x86_64 processor: x86_64 byteorder: little LC_ALL: None LANG: en_US.UTF-8 LOCALE: None.None pandas: 0.19.1 nose: 1.3.1 pip: 1.5.4 setuptools: 3.3 Cython: 0.25.1 numpy: 1.11.2 scipy: 0.18.1 statsmodels: 0.6.1 xarray: None IPython: 4.2.1 sphinx: None patsy: 0.2.1 dateutil: 2.6.0 pytz: 2016.7 blosc: None bottleneck: None tables: 3.1.1 numexpr: 2.2.2 matplotlib: 1.3.1 openpyxl: 1.7.0 xlrd: 0.9.2 xlwt: 0.7.5 xlsxwriter: None lxml: None bs4: 4.2.1 html5lib: 0.999 httplib2: 0.8 apiclient: None sqlalchemy: 1.0.15 pymysql: None psycopg2: 2.4.5 (dt dec mx pq3 ext) jinja2: 2.8 boto: 2.41.0 pandas_datareader: None

Comment From: jreback

The well-defined and document indexers have these types of guarantees (e.g. .loc), but these functions are being deprecated: https://github.com/pandas-dev/pandas/issues/15269

so unless someone puts forth a fix, this is won't fix.

Comment From: jreback

actually I spoke too soon about other operators. It depends if there are multiple operations or not. Though honestly you are pushing the boundaries here; these are not atomic operations. That said, if you'd like to push a fix would take it.

In [19]: d = pd.DataFrame()

In [20]: d.at[4,1] = [1,2,3]
ValueError: setting an array element with a sequence.

In [21]: d
Out[21]: 
    1
4 NaN

this is fine.

In [22]: d = pd.DataFrame()
    ...: 

In [23]: d.loc[4] = [1,2,3]
ValueError: cannot set a frame with no defined columns

In [24]: d
Out[24]: 
Empty DataFrame
Columns: []
Index: []

Comment From: jorisvandenbossche

@jreback the same happens with loc if you specify both dims:

In [29]: d = pd.DataFrame()

In [30]: d.loc[4, 3] = [1,2,3]
...
ValueError: setting an array element with a sequence.

In [31]: d
Out[31]: 
    3
4 NaN

Comment From: jreback

of course this is a 2 stage operation in implementation

Comment From: allComputableThings

... hmmm, so the caller should guess at the implementation to infer the behavior? Generally, any code that throws an exception if better off its it idempotent. Could we leave this as an open bug (clearly it shouldn't modify the dataframe, and is therefore a bug, even if the fix is not obvious).

Comment From: jorisvandenbossche

Given that the behaviour is also present in non-deprecated functionality (.at, .loc), I agree we can reopen this. It will still need someone diving into it to fix it.

Comment From: bt-

Is the result from this example related? I was surprised that this did not return an error when the left side of the last assignment does.

d = pd.DataFrame({'A': [1, 2, 3], 'B': [4, 5, 6]})
# Gives KeyError: "['C', 'D'] not in index"
# d.loc[:, ['A', 'B', 'C', 'D']] 
# Gives no error and adds columns 'C' and 'D' with NaNs
d.loc[:, ['A', 'B', 'C', 'D']] = pd.DataFrame({'A': [7, 8, 9], 'B': [10, 11, 12]})
d