Correct way to fire graphics updates from a background thread

[m***@public.gmane.org]

Hi all,
just to have confirmation of what I am doing.
My app is multithreaded ; the main() loop and some other working threads, receiving / sending data and doing calculations.

The question: what is the correct way to fire a graphics update from a thread ?

My work (omitted all sanity checks):

- main:
.....
static int fd_refresh[2];
.....
pipe(fd_refresh);
SetNonblocking(fd_refresh[0]);
SetNonblocking(fd_refresh[1]);
Fl::add_fd(fd_refresh[0], Do_refresh_callback, NULL);
.....
#define CMD_REFRESH 0
.....
static void Do_refresh_callback(int fd, void *data)
{
unsigned int val;
while(read(fd, &val, sizeof(val)) > 0)
{
switch(val)
{
case CMD_REFRESH:
// HERE I can draw to screen !
break;
.....
}
}
.....
}

- A thread receives data and wants to update a drawing. It cannot do it directly,
so it writes to fd_refresh[1] an integer which values represent what I want to do
(full redraw(), partial updates, etc.) .
Fltk then will call Do_refresh_callback and there I will be safe drawing anything.

Is this sequence correct ?
Now it's working without (apparent) problems.

Thanks,
regards

Lucio

[m***@public.gmane.org]

Post by m***@public.gmane.org

Post by m***@public.gmane.org
The question: what is the correct way to fire a graphics update from

a thread ?
Others will probably chime in and tell you more about programming with
threads and FLTK.
Ah... That quite possibly means me then...

:-)

Post by m***@public.gmane.org
Firstly, a caveat: Rendering to the display device from any thread other than the "main" thread is not robust (may work on some systems, will fail in horrible, hard to debug ways, on others, so do not do that.)

I can confirm this. REALLY horrible ways.

Post by m***@public.gmane.org
Instead, all drawing is done from the "main" thread. In particular, note that creating/deleting/hiding/showing windows *has* to be done by the "main" thread.
However, if you create your widgets in the "main" thread, you can alter them from other threads quite safely, if you use the fltk locking methods. You can signal updates between threads using Fl::awake(); there's no need to get fancy with pipe's > and so forth (fltk is handling that internally) and it is more portable.
- in the "main" thread, before you create any windows, call Fl::lock(); *just once* to enable thread support. (Do not do any more locking/unlocking in the main thread, let fltk manage that for you.)
Fl::lock();
my_widget->change_this();
my_other_widget->change_that(value);
etc.;
Fl::unlock();
Fl::awake();
And that should update the widgets in a thread-safe way, then flag the main thread to manage the redraw sequence.

In my first app I used Fl::lock, but when the app is growing I have to be careful not to create deadlocks, as threads can communicate with each other.
Nothing impossible, but I found the "pipe" method a little easier to debug.

Post by m***@public.gmane.org
Note that Fl::awake can also be passed a callback function - the "main" thread will then execute that callback in its context,
allowing the worker threads to do some operations that would not otherwise be allowed
(e.g. you may be able to use this to create new windows under the control of a child thread, since the actual creation will occur in the context of the "main" thread.)

This is interesting. I missed it.

Post by m***@public.gmane.org
However, that is often not necessary, the simple case I outlined above covers a lot of use cases.

Albrecht and Ian, thank you for your help.
If you are coming to Italy let me know, beer ready for you !

Thanks,
regards

Lucio

[MacArthur, Ian (Selex ES, UK)]

Post by m***@public.gmane.org
In my first app I used Fl::lock, but when the app is growing I have to
be careful not to create deadlocks, as threads can communicate with
each other.

Sure; but the Fl::lock() is a mutex, so you must only take it immediately before you want to modify the widgets, and you release it as soon as possible.

You never hold the lock for any longer than is necessary, and you do nothing but fltk GUI operations within the lock, so there should be no scope for deadlocking.

That said, if you need to be handling incoming data asynchronously (and at a high rate) you want to be separating that out from the GUI update process anyway - see Edzard's note on queuing up data for processing, then dealing with that at "user rates", not at input rates.




Selex ES Ltd
Registered Office: Sigma House, Christopher Martin Road, Basildon, Essex SS14 3EL
A company registered in England & Wales. Company no. 02426132
********************************************************************
This email and any attachments are confidential to the intended
recipient and may also be privileged. If you are not the intended
recipient please delete it from your system and notify the sender.
You should not copy it or use it for any purpose nor disclose or
distribute its contents to any other person.
********************************************************************

[m***@public.gmane.org]

Post by MacArthur, Ian (Selex ES, UK)
Sure; but the Fl::lock() is a mutex, so you must only take it immediately before you want to modify the widgets, and you release it as soon as possible.
You never hold the lock for any longer than is necessary, and you do nothing but fltk GUI operations within the lock, so there should be no scope for deadlocking.
That said, if you need to be handling incoming data asynchronously (and at a high rate) you want to be separating that out from the GUI update process anyway -
see Edzard's note on queuing up data for processing, then dealing with that at "user rates", not at input rates.

Nice ideas ; actually the separation is done from the pipe, which buffers update requests.
Explicitly buffering is much more elegant. Thanks.

regards
Lucio