I take back my comment about write-blocking ("I don't see a check to see how much can be written without blocking"). I forgot to factor in that non-blocking applies to writing too.

So you don't have a problem with syswrite, but you do have a problem with sysread. If the sysread would block, it'll return undef, and your program will exit prematurely.

On Linux, setting the sockets and pipes to non-blocking doesn't seem to be required at all, but I am not sure it other Unixes will perform the same.

Nonsense. Imagine if the remote end does

print while <>;
[download]

If you try reading from the socket, your program will become permanently blocked.

If you try reading from the socket, your program will become permanently blocked.

The select call takes care of that. The program only reads when there is data available and the buffer is not full.

Yes, you're right, the select (or more precisely, the if vec) takes care of that.

So yes, you don't need to make the *read* handle non-blocking because of that check.

You do need to continue making the *write* handles non-blocking because you don't have a similar check around syswrite. (Note that the similar check for write handles involves checking how many bytes you can write without blocking.)

Sorry, I got confused by the lack of symmetry between your reading and writing code.