This is probably more a question about programming methods than an request for getting some specific code going. Bear in mind I have come from an embedded programming background which is very much run to complete software.

I have finished a little Qt4 C++ program that sends and receives serial data for programming embedded microcontrollers. I send out little packets of data over a USB virtual com port (the virtual com port sends and receives data in exactly the same way as a normal ttySx com port).

Now, I have a big array of memory content that I parse and then send out 16 bytes at a time with a checksum. I then wait for an acknowledgement byte back from the microcontroller. When I say wait, I really mean wait. The program is just stuck in an endless loop waiting for the ack byte.

I am sure that, apart from the fact that if no serial message comes back, this is just not the way to write PC programs. In an embedded routine I would either have a timer go off if nothing was received and report an error or be calling a serial handling routine from my main loop. There doesn't seem to be this main loop way of writing code in PC based GUI applications.

It seems in C++ global variables are really frowned upon. I could, I suppose, have a timer go off and check the routine that does all this parsing and handling, going through this large state machine a stage at a time, but I think that would necessitate having loads of global variables.

Is it easy either to explain or point me to an article on how this is done in this sort of application (it's using KDevelop and Qt4)? I have tried Googling around, but it's such a verbose question it doesn't work too wll in a search engine!

Many thanks.

You can use for example a blocking read with a time out like that:
"poll" function will return a soon as something append on device "tty_dev" (error, incoming data...), or return after "timeout" in ms.
You can monitor for event on several device with only one "poll" since the function take an array of "struct pollfd", one per device to monitor.

In GUI programming, I think there is usually a way of hooking into an 'idle' condition such that your callback is called whenever the GUI is idle. You could create and register a callback that uses the select() function to determine whether there is data available in your serial device. This is a standard facility provided by the OS for such circumstances. You would probably be well served to find an online tutorial or two that explains the use of the select() call, as well as the small number of functions that are used in conjunction with it.
--- rod.

In GUI programming with packages like QT or WX -- and most others -- the "main loop" is generally handled by the library itself to abstract event handling and other low level things like drawing to the screen.

In your case and almost ALL cases you want to handle non-GUI specific things like database, serial and socket connections outside of your GUI code so the interface stays responsive and the user productive while other tasks are performed. You can do this with other processes, threads or simply call a function in your serial handling class to poll if data is in the buffer.

This is generally done in an MVC model and keeps the code clean, maintainable and extensible. In simple cases, you could keep this all balled up into one like you have it now and just present a modal window with a progress bar and cancel button. Obviously, that's not what you want.

In short, look up MVC GUI programming techniques.

Thanks everyone, that's given me lots to go on. I'll do some reading up.

I'd vote for using the idle loop to dispatch to certain functions depending on a state machine. The state machine will at least need static variables in the idle loop.

Ignore the fools who scream about using globals; while it is bad practise to use globals merely for convenience, coming up with even worse schemes for the sake of avoiding globals is just ridiculous. To make things slightly neater, you can create one struct to hold globals.