homepage

In both Python 2.7 and Python 3.x, csv.Error is documented as:
 Raised by any of the functions when an error is detected.
As far as I can tell from using the module and looking at the code, this is completely incorrect. There is actually a single instance of csv.Error being used: the instantiation of csv.Dialect (which converts TypeError raised from _csv._Dialect() into csv.Error, a comment notes that this is "for compatibility with py 2.3").
And the only way to hit that code paths seems to be subclassing `Dialect` and putting incorrect values in the various attributes (providing them to `csv.reader` raises a TypeError).
I believe the documentation to csv.Error should be changed to:
1. Mark it as effectively deprecated
2. Indicate that the only situation in which it it may be raised is when initializing a subclass of csv.Dialect

Correction: csv also seems to raise csv.Error if the file contains NUL bytes:
 Error: line contains NULL byte

Thanks for the report.
Most of the csv.Error usages are in Modules/_csv.c. You can search for "_csvstate_global->error_obj" in the module's source code. I think the documentation is correct as is.