Conference bulova::decw_jan-89_to_nov-90

3247.0. "Crash in FetchWidget, _XInputArrival" by STKHLM::GRANVALD () Thu Aug 23 1990 02:31

	Hi,

	We have experienced a rather nasty problem in V2 of DECWindows.
	
	The application we are building has to use EFN's out of cluster 1
	to syncronize with other stuff. Adding to that we are using a 
	set of timers and XtAppPending, XtAppProcessEvent.
	The AST routines does only alter an EF, NO Xt or Xlib calls are made.

	This was working all right on V1 (the development is done there),
	but moving it to the target systems running VMS 5.3-1 (VS3100,SPX),
	it blows up 8 times out of 10. This is what happends;

	- The timers to poll the event queue are set up.
	- We start of a DwtFetchWidget, were the hierarcy is rather deep.
	- Before the fetch completes the timer AST hits and we set an
	  event flag and return

	- Crash,

  [_XInputArrival - input buffer pointer not empty during processing]
  [               - state: 2   flags: 0  stopDepth: 0]
  %SYSTEM-F-BADPARAM, bad parameter value
  %TRACE-F-TRACEBACK, symbolic stack dump follows
  module name     routine name                     line     rel PC     abs PC
							  00126DDE   00126DDE
							  001689B6   001689B6


	- If we disable AST delivery before the call to FetchWidget it
	  works nicely, so far...

	BUT...

	- Could this happend at any call to the toolkit ?
	  i.e we have to disable AST delivery before all calls to the toolkit...

	- What does the error say?

	- Will this be fixed? (or maby it is fixed)

	- The error has not been posted in the QAR system, I'll try to
	  strip down the application first

        - Any ideas?
    
    
    
    /Anders
    

How are you allocating your event flags?  Are you using the LIB$*_EF
routines?  If not (I'm sure DECwindows does) you're likely to be using the
same event flags as DECwindows since DECwindows thinks it's OK to use
them (since LIB$GET_EF says so).

Are you using the VMS-reserved event flags (24-31 I think)?  If so, all bets
are off what happens.

Why must you use cluster 1?  You should code the application to get the
event flags it needs with LIB$GET_EF.

-Mike

>How are you allocating your event flags?  Are you using the LIB$*_EF
>routines?  If not (I'm sure DECwindows does) you're likely to be using the
>same event flags as DECwindows since DECwindows thinks it's OK to use
>them (since LIB$GET_EF says so).
    
    We are using the LIB$*_EF routines.
    
>Why must you use cluster 1?  You should code the application to get the
>event flags it needs with LIB$GET_EF.
    
    Well, we have to syncronize the activity in the user interface with
    other layers of software, who in some cases give us the flag to use
    (all of them are allocated using LIB$*_EF routines).
    The LIB$GET_EF also seem to allocate flags 63 and down.
    If we were independent of the other layers we could use another 
    scheme, but such is life...
    
    But why is it working using V1 (VMS 5.2)? It's the same code!
    
    /Anders
    

    An QAR has been sent in.