> when does the receiving side evaluate the code? There are different solutions. E.g. I can send its length information with the data, and call parse-evaluate when that many bytes arrived. (I coded this up, and it is working, too). However, the function I posted for evaluating computer generated code does not use input streams, but properly terminated strings are pasted to the console, that is, it uses messages for the input. The return path is also safe: whenever the evaluation is done, the next Julia instructions are executed, which save the output. I did not find problems with this approach, except with the aforementioned triple-quoted strings, which is not a big deal. Multi-line input and output, error messages, etc. all work reliably.
The problem I see is the unpredictable behavior of the REPL. With an older Julia version I get some output twice. Can I rely on the current behavior of the REPL (when the display() results don't appear on STDOUT), or in another nightly version they would appear again? Some kind of documentation on the REPL I/O would really help us.
