Summary:
In #15964, I learned that `errno` is only meaningful if the function call fails. E.g., on some macos, a successful `fork()` sets `errno` to `EINVAL` in child process. This commit changes the `SYSCALL` macro so error checking is only done when an error happens. This means checking whether `rv == -1` for most calls, but is checking `rv == nullptr` for `inet_ntop`.

Now `SYSCALL` accepts a second argument `success_cond`, which should be an expression returning whether the call succeeded. `SYSCHECK_ERR_RETURN_NEG1` is the shorthand for checking if rv is `-1`.

Any suggestion on better macro names is welcomed.
Pull Request resolved: #15986

Reviewed By: janewangfb

Differential Revision: D13661790

Pulled By: pietern

fbshipit-source-id: 9551b14b9f88805454a7bfb8e4d39e0f3aed8131

Summary:
fixes pytorch#14314
Pull Request resolved: pytorch#15964

Differential Revision: D13639034

Pulled By: soumith

fbshipit-source-id: 44592762aa46982e5d3616d55b5666a2c2ce9105

Summary:
In pytorch#15964, I learned that `errno` is only meaningful if the function call fails. E.g., on some macos, a successful `fork()` sets `errno` to `EINVAL` in child process. This commit changes the `SYSCALL` macro so error checking is only done when an error happens. This means checking whether `rv == -1` for most calls, but is checking `rv == nullptr` for `inet_ntop`.

Now `SYSCALL` accepts a second argument `success_cond`, which should be an expression returning whether the call succeeded. `SYSCHECK_ERR_RETURN_NEG1` is the shorthand for checking if rv is `-1`.

Any suggestion on better macro names is welcomed.
Pull Request resolved: pytorch#15986

Reviewed By: janewangfb

Differential Revision: D13661790

Pulled By: pietern

fbshipit-source-id: 9551b14b9f88805454a7bfb8e4d39e0f3aed8131