Message 289691 - Python tracker (original) (raw)

also, IMHO, we should open an issue named 'raise ValueError instead of OverflowError when a negative int is invalid', and leave for that new issue some the changes (which you commented about in the PR).

I didn't want to open it without your approval. here is what I thought should be the contents of the issue (feel free to edit my draft as you wish and open the issue yourself, if you think we should open it):

In various C functions, a negative int argument is invalid. in most of these functions, ValueError is raised for a negative int, or at least for a small negative int (i.e. a negative int that fits in some signed integer C type).

as Serhiy mentioned in http://bugs.python.org/issue15988#msg289679 and in code review comments in https://github.com/python/cpython/pull/668, it might be that a ValueError should always be raised in such functions, upon receiving a negative int (even upon receiving a big negative int, such as (-1 << 1000)).

the following functions were identified as relevant only while working on #15988, so there are probably (a lot?) more relevant functions.

functions where OverflowError is raised only for big negative ints: - Objects/bytesobject.c - bytes_new - Objects/bytearrayobject.c - bytearray_init - functions that would be fixed as part of #29819? * Modules/_io/stringio.c - _io_StringIO_truncate_impl * Modules/_io/bytesio.c - _io_BytesIO_truncate_impl

functions where OverflowError is raised for any negative int (so a change would probably break some users code): - Objects/longobject.c: * PyLong_AsUnsignedLong * PyLong_AsSize_t - Modules/_ctypes/_ctypes.c - PyCArrayType_new

Note: while patching these functions, one should also make sure that the patched functions produce consistent error messages, as explained in #15988 (in which it was decided to leave these functions (which currently produce inconsistent error messages) to this issue.)