Some time ago, PEP 475 was accepted in order to automatically retry system calls on EINTR, but the _ssl module wasn't touched.

However, it seems that OpenSSL socket calls can fail with EINTR:
https://stackoverflow.com/questions/24188013/openssl-and-signals

"it seems that OpenSSL socket calls can fail with EINTR: ..."

You are easily test by raising a signal: see Lib/test/test_eintr.py which stress Python functions by sending a signal every 100 ms while Python is blocked in a syscall.

If there is a bug, sure, it must be fixed ;-)

It was very tricky to handle EINTR properly in the socket module. While most Python functions are thin wrapper to syscalls with no timeout, almost every socket functions have a timeout. So I wrote sock_call() to factorize the code. This function calls select() internally to check if the socket is "ready", but also updates the timeout, etc.

A few other Python are retried on EINTR and update a timeout, like select.select(), time.sleep(), threading.Lock.acquire() and signal.sigtimedwait().

See bpo-23618 for handling EINTR in socket and bpo-23834 for sock_call().

Looks like this behavior is documented in the bio_should_retry(3) man page.

Basically bio_should_retry() tells if the error was non-fatal:

    int BIO_fd_should_retry(int i)
    {
        int err;

        if ((i == 0) || (i == -1)) {
            err = get_last_sys_error();

            return BIO_fd_non_fatal_error(err);
        }
        return 0;
    }