BUG: Improve error message for multi-char sep and quotes in Python engine (#14582)

If there is a field counts mismatch, check whether
a multi-char sep was used in conjunction with quotes.
Currently, that setup is not respected and can result
in improper line breaks.

Closes gh-13374.

Author: gfyoung

doc/source/whatsnew/v0.19.2.txt

@@ -30,6 +30,7 @@ Bug Fixes
 - Compat with ``dateutil==2.6.0``; segfault reported in the testing suite (:issue:`14621`)
 - Allow ``nanoseconds`` in ``Timestamp.replace`` as a kwarg (:issue:`14621`)
 - Bug in ``pd.read_csv`` where reading files fails, if the number of headers is equal to the number of lines in the file (:issue:`14515`)
+- Bug in ``pd.read_csv`` for the Python engine in which an unhelpful error message was being raised when multi-char delimiters were not being respected with quotes (:issue:`14582`)
 
 
 

pandas/io/parsers.py

@@ -2515,6 +2515,11 @@ def _rows_to_cols(self, content):
 
             msg = ('Expected %d fields in line %d, saw %d' %
                    (col_len, row_num + 1, zip_len))
+            if len(self.delimiter) > 1 and self.quoting != csv.QUOTE_NONE:
+                # see gh-13374
+                reason = ('Error could possibly be due to quotes being '
+                          'ignored when a multi-char delimiter is used.')
+                msg += '. ' + reason
             raise ValueError(msg)
 
         if self.usecols:

pandas/io/tests/parser/python_parser_only.py

@@ -7,6 +7,7 @@
 arguments when parsing.
 """
 
+import csv
 import sys
 import nose
 
@@ -204,3 +205,19 @@ def test_encoding_non_utf8_multichar_sep(self):
                                        sep=sep, names=['a', 'b'],
                                        encoding=encoding)
                 tm.assert_frame_equal(result, expected)
+
+    def test_multi_char_sep_quotes(self):
+        # see gh-13374
+
+        data = 'a,,b\n1,,a\n2,,"2,,b"'
+        msg = 'ignored when a multi-char delimiter is used'
+
+        with tm.assertRaisesRegexp(ValueError, msg):
+            self.read_csv(StringIO(data), sep=',,')
+
+        # We expect no match, so there should be an assertion
+        # error out of the inner context manager.
+        with tm.assertRaises(AssertionError):
+            with tm.assertRaisesRegexp(ValueError, msg):
+                self.read_csv(StringIO(data), sep=',,',
+                              quoting=csv.QUOTE_NONE)