Re: GTK - Proper Way to Code Pipes

[Daniel Yek <dyek real com>]

At 02:27 PM 12/5/2006, Christopher Bland wrote:
> >> Based on suggestions from this mailing list I've implemented a pipe to
> >> pass data from my other processes to the main GTK thread. I noticed
> >> that when writing to the pipe there was a long delay before it was
> >> handled. I'm assuming this is because it was a pending event.
> >>

> <snip...>


> I'm pretty sure this is exactly what I've been seeing. Within my app I 
> have a 'clicked' event
> attached to a button which basically starts all of my processing. Within 
> this callback I call
> my main C++ routine which creates my objects and starts doing work. My app 
> is multi-
> threaded but I had such a headache trying with the threads_enter/leave and 
> someone
> suggested using pipes and use g_io_add_watch to "watch" a pipe.
>
> Within one of my C++ routines I write to the pipe, but the "watch" routine 
> does not get
> called during execution of my threads. When control is given back to the 
> GUI thread (hence
> my call to gtk_main_iteration()) then the "watch" routine gets fired off.
>
> I wanted to use the "watch" routine to move my progressbar and update the 
> statusbar in my
> application and I'm having such a hard time getting it working with my 
> threads.
>
> >typically people will open (pipename, O_WRONLY | O_NONBLOCK) just to
> >open the file with out blocking, and then fcntl(fd...) to set it back
> >to blocking mode... then you might go ahead and select() the fd to know
> >if the pipe is ready to receive one byte of data without blocking
> >(maybe select()/write()ing one byte at a time until the buffer to write
> >is finished... always returning to the main loop when it would block).
>
> I'm doing this as well. I basically followed the following example:
> http://wwwtcs.inf.tu-dresden.de/~tews/Gtk/catch_signals.c

Did you notice any abnormal CPU usage?

That web site is really useful, but it didn't point out that you need to 
handle G_IO_HUP, G_IO_NVAL, (and maybe G_IO_PRI). Handling these helped my 
troubles, but I don't know if these have anything to do with your trouble.



> >Remember that this heavily depends on what you are going to do with
> >the pipe, do you want the reader to block in read() all the time ?
> >do you want the writer to block in write() and have the reader
> >test for a readable condition on the file descriptor before reading
> >a byte ? do you want to use nonblocking read/write operations on both
> >ends and just buffer the data in the pipe ? it can get complex.
> >
> >Cheers,
> >                  -Tristan
>
> I really would like to be able to write to the pipe from my threads, have 
> the GUI read from
> the pipe and then update my progress/status bars based on the data within 
> the pipe. If
> anyone has any suggestions I would greatly appreciate it. I thought I was 
> on the road to
> success but now I'm starting to doubt myself.
>
> Thanks.




Daniel Yek