BUG: Inconsistent behavior: Occasional crash on same code and data

[arshadparvez]

[x] I have checked that this issue has not already been reported.
[x] I have confirmed this bug exists on the latest version of pandas.

Description
When reading csv with pd.read_csv, giving both names= and dtypes= arguments, the program occasionally crashes. Surprisingly, sometimes the application completes its execution cycle normally - without any change in code or data.

I had initially used only the dtype argument, but the first row of my data would be considered as column headers (the csv, otherwise has no headers). Instead of using headers=None, I added names argument, which on one hand made all my rows read as data while on the other set the desired column headers in DataFrame.

But after this change, the program's behavior has become inconsistent; occasional crashing while normal execution at other times.

Reproducing Code Example

Example Code and Data
Minimal code along with a small data file, that demonstrate the reported behavior, is:

import pandas as pd

df = pd.read_csv('data\\test3.csv', names={'pVal1', 'pClass'}, dtype={'pVal1':int, 'pClass':bool})
print (df)

The data file test3.csv contains:

1,TRUE
2,FALSE
3,TRUE

The problem goes away if we remove either of the two arguments (names, dtype). With any of the following statements, the program executes normally:

df = pd.read_csv('data\\test3.csv', names={'pVal1', 'pClass'})

or

df = pd.read_csv('data\\test3.csv', dtype={'pVal1':int, 'pClass':bool})

Error message:

Demonstration of Inconsistent Behaviour

I am posting the result of two consecutive executions; first successful, the second failing with error:

Successful Execution
(ml) C:\Users\pyuser\projs\ml>py err.py
pVal1 pClass
0 1 True
1 2 False
2 3 True

Crash with Error
(ml) C:\Users\pyuser\projs\ml>py err.py
Traceback (most recent call last):
File "pandas_libs\parsers.pyx", line 1152, in pandas._libs.parsers.TextReader
._convert_tokens
TypeError: Cannot cast array from dtype('int64') to dtype('bool') according to t
he rule 'safe'

During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "err.py", line 3, in
df = pd.read_csv('data\test3.csv', names={'pVal1', 'pClass'}, dtype={'pVal1
':int, 'pClass':bool})
File "C:\Users\pyuser\projs\ml\lib\site-packages\pandas\io\parsers.py", line 676
, in parser_f
return _read(filepath_or_buffer, kwds)
File "C:\Users\pyuser\projs\ml\lib\site-packages\pandas\io\parsers.py", line 454
, in _read
data = parser.read(nrows)
File "C:\Users\pyuser\projs\ml\lib\site-packages\pandas\io\parsers.py", line 113
3, in read
ret = self._engine.read(nrows)
File "C:\Users\pyuser\projs\ml\lib\site-packages\pandas\io\parsers.py", line 203
7, in read
data = self._reader.read(nrows)
File "pandas_libs\parsers.pyx", line 860, in pandas._libs.parsers.TextReader.
read
File "pandas_libs\parsers.pyx", line 875, in pandas._libs.parsers.TextReader.
_read_low_memory
File "pandas_libs\parsers.pyx", line 952, in pandas._libs.parsers.TextReader.
_read_rows
File "pandas_libs\parsers.pyx", line 1084, in pandas._libs.parsers.TextReader
._convert_column_data
File "pandas_libs\parsers.pyx", line 1160, in pandas._libs.parsers.TextReader
._convert_tokens
ValueError: cannot safely convert passed user dtype of bool for int64 dtyped dat
a in column 0

Software and Library Versions

Numpy/Python version information:

(ml) C:\Users\pyuser\projs\ml>python --version
Python 3.8.2

(ml) C:\Users\pyuser\projs\ml>python -c "import numpy; print(numpy.version.version
)"
1.18.4

(ml) C:\Users\pyuser\projs\ml>python -c "import pandas; print(pandas.version)"
1.0.5

1.18.4 3.8.2 (tags/v3.8.2:7b3ab59, Feb 25 2020, 23:03:10) [MSC v.1916 64 bit (AM
D64)]

[MJafarMashhadi]

I could reproduce this on the latest commit on master (506eb54) and also observed the mentioned non-deterministic behavior.

[MarcoGorelli]

Thanks @arshadparvez for the report

For a copy-and-pasteable version:

import pandas as pd
from io import StringIO

pd.read_csv(
    StringIO(
        """1,TRUE
2,FALSE
3,TRUE
"""
    ),
    names={"pVal1", "pClass"},
    dtype={"pVal1": int, "pClass": bool},
)

[jorisvandenbossche]

One should not specify the names as a python set, as those are not ordered. I suppose that it the reason of the occasional failure, if it tries to convert the first column (instead of second) to a bool dtype.

We should probably disallow sets for the names argument

[MJafarMashhadi]

/take