OpenEuphoria

1. Non Blocking Support

• Posted by C Bouzy <CBouzy at Egisca.com> Feb 15, 2007
• 695 views

Hello All,

I started working on Nexus Radio 2 and I just wanted to know if EU
will have non blocking support in the near future, or is anyone working
on this? 



"If you do what you have always done, you will get what you always got."

3. Re: Non Blocking Support

• Posted by Jonas Temple <jtemple at yhti.net> Feb 15, 2007
• 610 views

Chris,

I'm currently using tdll by Daniel Kluss and found that it works quite well!

I'm about to release a new version of FROG that uses Euphoria's new task
routines and Daniel's tdll.  Users will be able to run multiple SQL statements
against a host at the same time.

I did have to modify Win32Lib (if that's what you use) to put a call to
task_yield() in the doEvents() routine.

Jonas Temple
http://www.innovativesys.net

5. Re: Non Blocking Support

• Posted by Jonas Temple <jtemple at yhti.net> Feb 15, 2007
• 615 views

CChris wrote:
> > <a href="http://www.innovativesys.net">http://www.innovativesys.net</a>
> 
> Did you plainly insert a task_yield() call at line 33376 or is it
> anything subtler?
> 
> CChris

Actually, looking at what I did, I insert this into two places.  And, of course,
I can't remember why I did the first!  :)

The first is in eventLoop() procedure:

-- Gather some entropy for the random integer routine
        w32Seed = remainder((w32Seed * peek4u(msg+16))
                            + 1 + peek4u(msg+20) * peek4u(msg+24),
                            #FFFFFFFF)

        w32Proc( xTranslateMessage, { msg } )
        w32Proc( xDispatchMessage, { msg } )
        task_yield()

The second is in doEvents():

global procedure doEvents(integer id)
    atom msg, hWnd

    if vWinMainState != kStarted then
    return
    end if

    if id = 0 then
    hWnd = 0
    else
    if validId(id) = w32False then
        return
    end if
    hWnd = getHandle(id)
    end if

    -- Allocate a message buffer
    msg = w32acquire_mem(0, SIZEOF_MSG)

    if w32Func( xPeekMessage, { msg, hWnd, 0, 0, PM_REMOVE } ) then
        w32Proc( xTranslateMessage, { msg } )
        w32Proc( xDispatchMessage, { msg } )
        task_yield()
    end if
    w32release_mem(msg)

end procedure

And to be honest, I haven't noticed any performance depreciation using
task_yield() in these routines.

Sorry I can't give you the exact line numbers, my copy of Win32lib has some
"tweaks".

Just in case you're curious, here's how I modified my code (which was not a
lot!)

* included tdll.e (of course)
* changed my open_dll() call to open_tdll()
* For functions that I wanted non-blocking, I replaced the define_c_func() calls
with define_t_func() calls.  In my case the only change was replacing the "c" in
define_c_func with "t".  The parameters to the define_t_func were identical to
define_c_func.  I then replaced the c_func() calls with t_func().  Again, the
only change was the "c" to "t".  I even left all the rest of the C functions with
the standard define_c_func/c_func Euphoria routines and they worked fine!

In my main program I coded a routine to kick off the task and then wait until it
was complete:

task = task_create(routine_id("execute_task"), {invoked_from, ctl_index})
    task_schedule(task,1)
    -- Process message loop until end of task
    while find(task, task_list()) do
        doEvents(0)
        task_yield()                    
    end while

Carefully placed task_yield() calls, both in the main program and in Win32Lib
seemed to keep things in order.

I did have to rework a HUGE portion of my code to eliminate my over-use of
globals.  But that was my issue, not tdll.  I just had to ensure that one task
would not interfere with the values in another task.

I was having some problems at first before I figured out how to place calls to
task_yield() and Daniel suggested his APC DLL which he said was better than TDLL
but I could not figure out how to use the library and never got a response back
from him on a request for documentation.  I just found the TDLL calls easier to
understand.

HTH,

Jonas Temple
http://www.innovativesys.net

7. Re: Non Blocking Support

• Posted by C Bouzy <CBouzy at Egisca.com> Feb 15, 2007
• 605 views

Hello Jonas,

Thank you for your feedback. I will give your method a try and see if it works.

C. Bouzy

"If you do what you have always done, you will get what you always got."

9. Re: Non Blocking Support

• Posted by CChris <christian.cuvier at agriculture.gouv.fr> Feb 15, 2007
• 630 views

Jonas Temple wrote:
> 
> CChris wrote:
> > > <a href="http://www.innovativesys.net">http://www.innovativesys.net</a>
> > 
> > Did you plainly insert a task_yield() call at line 33376 or is it
> > anything subtler?
> > 
> > CChris
> 
> Actually, looking at what I did, I insert this into two places.  And, of
> course,
> I can't remember why I did the first!  :)
> 
> The first is in eventLoop() procedure:
> 
> }}}
<eucode>
>         -- Gather some entropy for the random integer routine
>         w32Seed = remainder((w32Seed * peek4u(msg+16))
>                             + 1 + peek4u(msg+20) * peek4u(msg+24),
>                             #FFFFFFFF)
> 
>         w32Proc( xTranslateMessage, { msg } )
>         w32Proc( xDispatchMessage, { msg } )
>         task_yield()
> </eucode>
{{{

> 
> The second is in doEvents():
> 
> }}}
<eucode>
> global procedure doEvents(integer id)
>     atom msg, hWnd
> 
>     if vWinMainState != kStarted then
>     return
>     end if
> 
>     if id = 0 then
>     hWnd = 0
>     else
>     if validId(id) = w32False then
>         return
>     end if
>     hWnd = getHandle(id)
>     end if
> 
>     -- Allocate a message buffer
>     msg = w32acquire_mem(0, SIZEOF_MSG)
> 
>     if w32Func( xPeekMessage, { msg, hWnd, 0, 0, PM_REMOVE } ) then
>         w32Proc( xTranslateMessage, { msg } )
>         w32Proc( xDispatchMessage, { msg } )
>         task_yield()
>     end if
>     w32release_mem(msg)
> 
> end procedure
> </eucode>
{{{

> 
> And to be honest, I haven't noticed any performance depreciation using
> task_yield()
> in these routines.
> 
> Sorry I can't give you the exact line numbers, my copy of Win32lib has some
> "tweaks".
> 
> Just in case you're curious, here's how I modified my code (which was not a
> lot!)
> 
> * included tdll.e (of course)
> * changed my open_dll() call to open_tdll()
> * For functions that I wanted non-blocking, I replaced the define_c_func()
> calls
> with define_t_func() calls.  In my case the only change was replacing the "c"
> in define_c_func with "t".  The parameters to the define_t_func were identical
> to define_c_func.  I then replaced the c_func() calls with t_func().  Again,
> the only change was the "c" to "t".  I even left all the rest of the C
> functions
> with the standard define_c_func/c_func Euphoria routines and they worked fine!
> 
> In my main program I coded a routine to kick off the task and then wait until
> it was complete:
> 
> }}}
<eucode>
>     task = task_create(routine_id("execute_task"), {invoked_from, ctl_index})
>     task_schedule(task,1)
>     -- Process message loop until end of task
>     while find(task, task_list()) do
>         doEvents(0)
>         task_yield()                    
>     end while
> </eucode>
{{{

> 
> Carefully placed task_yield() calls, both in the main program and in Win32Lib
> seemed to keep things in order.  
> 
> I did have to rework a HUGE portion of my code to eliminate my over-use of
> globals.
>  But that was my issue, not tdll.  I just had to ensure that one task would
> not interfere with the values in another task.
> 
> I was having some problems at first before I figured out how to place calls
> to task_yield() and Daniel suggested his APC DLL which he said was better than
> TDLL but I could not figure out how to use the library and never got a
> response
> back from him on a request for documentation.  I just found the TDLL calls
> easier
> to understand.
> 
> HTH,
> 
> Jonas Temple
> <a href="http://www.innovativesys.net">http://www.innovativesys.net</a>

Sure does. And perhaps the next win32lib version could have a task_yield() 
in doEvents(), provided there is a zero time way to prevent the call from 
Eu intrepreters that don't support task_yield.
I didn't test, but OTOH I'd hink the task_yield() n the message loop is 
overkill and could eat some CPU as it would be executed often for no benefit.
. You appear to be well placed o test whether it is necessary to leave that 
call there.

Thanks for the detailed reply.

CChris

11. Re: Non Blocking Support

• Posted by Robert Craig <rds at RapidEuphoria.com> Feb 16, 2007
• 635 views

Jonas Temple wrote:
> CChris wrote:
> > Sure does. And perhaps the next win32lib version could have a task_yield() 
> > in doEvents(), provided there is a zero time way to prevent the call from 
> > Eu intrepreters that don't support task_yield.
> > I didn't test, but OTOH I'd hink the task_yield() n the message loop is 
> > overkill and could eat some CPU as it would be executed often for no
> > benefit.
> > . You appear to be well placed o test whether it is necessary to leave that
> > 
> > call there.
> 
> Maybe Rob could comment on the overhead of repeating task_yield()?  

With no tasks created (other than the single main task 
that all programs now have), on my 1.8 GHz P4, I can do 
4 million task_yields per second using exw.exe. (1/4 microsecond each). 
So if you were somehow doing, say, 100,000 events per second 
(10 microseconds each) the overhead of 1/4 of a microsecond, 
would be 2.5%

Regards,
   Rob Craig
   Rapid Deployment Software
   http://www.RapidEuphoria.com