22 Jul
2016
22 Jul
'16
4:21 p.m.
On Fri, Jul 22, 2016, at 11:35, Victor Stinner wrote:
Oh, the first one is a regression that I introduced in the implementation of the PEP 475 (retry syscall on EINTR). I don't think that it can be triggered in practice, because socket handles on Windows are small numbers, so unlikely to be seen as negative.
The problem as I understand it isn't that handles will be seen as negative, the problem is that the error return will be seen as *non*-negative.
I just fixed it: https://hg.python.org/cpython/rev/6c11f52ab9db
Does INVALID_SOCKET exist on non-windows systems? (It's probably safe to compare against -1, the relevant functions are defined in POSIX as returning -1 rather than an unspecified negative value)