r.david.murray: ah, sure :) However, I'm not sure a test case is absolutely required for this issue for two reasons:

 - the fix is trivial: it's a one-liner that enables a SSL mode that explicitely authorizes SSL_write to be called a second time with a a different memory pointer than the first time. Since memory pointers are opaque to Python programmers anyway, I doubt it could break code (unless you'd expect the failure, of course :) )

 - tests about SSL in non-blocking mode are almost inexistant, I think. The only one I could find tests the handshake. See issue #3890 for instance. Probably because writing tests in non-blocking mode isn't easy.

However, my test may be correct, I'm just not sure it will pass everywhere :)