Fix two small issues that cause Windows problems #21
Conversation
1. The return value of `accept` for Windows is not -1 on failure, but `INVALID_SOCKET` (which is unsigned). Failures were getting reported as error code 0. 2. If the first argument to `accept` is `NULL` then the second is also required to be
|
Yes. Cool. Better to consider some loosely-related fixes separately than in one big PR. OMG, is the Windows |
|
Good thinking, there are probably more. I'll update the PR to fix the rest :p |
As the previous commit, the `::socket` function returns `SOCKET` (unsigned) on Windows. All other areas that use `check_ret` have been confirmed to return -1 on error on Windows as well (in general, Winsock2 functions that return int return SOCKET_ERROR (-1) on failure)
|
I counted roughly a dozen places that |
|
Cool. I'll merge this shortly and take a look. I also grep'ed for any additional "< 0" comparisons, but I think we're close. I may have been aware of this when I first wrote the library, but forgotten it since then! There are a lot of explicit comparisons to the invalid socket constant, which I don't normally do in Linux programming. I used Windows a lot back then, but not really at all any more. |
|
I made one minor change... As Pushed into |
This is also included in #17 but I figured I'd make it standalone as well since it is outside the context of TLS support.