19:15 <zmatt> huh, I never realized that a blocking write()/send() will only perform a partial write (i.e. at least 1 byte not less than requested) in cases where it would have returned an error if it hadn't already written some bytes

19:16 <zmatt> for some reason I thought it would work like read()/recv(), i.e. that it would behave non-blocking once *some* data has been transferred

19:18 <zmatt> well, that explains why send() doesn't support MSG_WAITALL ... it always behaves as if that flag were present :P