The DataFrame.to_dict method is inconsistent when there are no columns and orient='records'.

[jack-pace]

Code Sample

import pandas as pd

df = pd.DataFrame([{}, {}, {}])
df.to_dict(orient='records')
# Output: []

Problem description

The DataFrame.to_dict method is inconsistent when there are no columns and orient='records'.

I think that this is a recent regression, since it was not a problem on pandas 0.22.0.
Searching through previous issues I see it might be related to a change in version 0.24.0.

I expect pandas to return a list of three empty dictionaries here: [{}, {}, {}].
That way, for example, doing a from_dict followed by a to_dict will round trip and vice versa. E.g.

pd.DataFrame.from_dict([{}, {}, {}], orient='records').to_dict(orient='records') == [{}, {}, {}]
# Output: False (I expected True)

Also, things like the length will be consistent:

df = pd.DataFrame([{}, {}, {}])
len(df)
# Output: 3
len(df.to_dict(orient='records'))
# Output: 0

Expected Output

See "Problem Description"

Output of pd.show_versions()

INSTALLED VERSIONS

commit: None
python: 3.7.3.final.0
python-bits: 64
OS: Linux
OS-release: 4.8.0-53-generic
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: en_GB.UTF-8
LOCALE: en_GB.UTF-8

pandas: 0.24.2
pytest: 3.3.2
pip: 19.1.1
setuptools: 41.0.1
Cython: None
numpy: 1.16.4
scipy: 1.3.0
pyarrow: None
xarray: None
IPython: 7.5.0
sphinx: 1.4.8
patsy: 0.5.1
dateutil: 2.8.0
pytz: 2019.1
blosc: None
bottleneck: None
tables: None
numexpr: None
feather: None
matplotlib: None
openpyxl: None
xlrd: None
xlwt: None
xlsxwriter: None
lxml.etree: None
bs4: None
html5lib: None
sqlalchemy: 1.2.9
pymysql: None
psycopg2: 2.7.5 (dt dec pq3 ext lo64)
jinja2: 2.10.1
s3fs: None
fastparquet: None
pandas_gbq: None
pandas_datareader: None
gcsfs: None

[WillAyd]

This is working as documented; the "records" orient simply provides records in the format of [{column -> value}, … , {column -> value}], and since you have no columns there is nothing to show here.

I think you might be looking for orient='index' instead if you wanted to preserve the index:

>>> df.to_dict(orient='index')
{0: {}, 1: {}, 2: {}}

[Voynec]

This was about preserving the shape, not preserving the index.

The documentation shows a list of column -> value dictionaries, one for each row.

I think it is totally reasonable - based on what you yourself cited - to expect:
A) The number of keys in each dict should match the number of DataFrame columns
B) The length of the list of records should match the number of DataFrame rows

Plus:
json.loads(pd.DataFrame.from_dict([{}, {}, {}], orient='records').to_json(orient='records')) == [{}, {}, {}]

True

I get that it is more critical for a serialization method to be able to reproduce the original data when deserialized.
But as far as I can tell, the documentation for orient='records' in DataFrame.to_json matches that of DataFrame.to_dict?
I don't get how it can be working "as documented" when the same text is being used to describe two different behaviors.

At the very least the documentation could explicitly mention that shape is not preserved for this particular edge case