Tcl Source Code

View Ticket
Login
Ticket UUID: 1113452
Title: Tk_MainLoop processes WM_DEVICE_CHANGE infinitely often.
Type: Bug Created on: 2005-01-31 21:00:41
Submitter: kodecharlie Assigned to: jan.nijtmans
Subsystem: 02. Event Loops Severity: Minor
Priority: 5 Medium Last modified: 2014-02-23 16:52:14
Status: Open Closed by: nobody
Resolution: None Closed on:
Version: obsolete: 8.4.9
Description:
Here is what happens:

0. In MacroMedia's Director MX 2004, I launch a thread 
(via _beginthread, called from an xtra) that starts a 
Tcl interpreter (with Tk, TclX et al built in). 
1. The Task Manager in Windows XP shows that CPU 
consumption 
rises to 50-60%.  (Note that if I use the same Tcl/Tk 
libraries outside of the context of a Director xtra, no 
such penalty occurs.) 
2. I then load the xtra and debugged version of Tcl/Tk 
8.4.9 
into VC++.  I set breakpoints at Tk_MainLoop, 
Tcl_DoOneEvent­, 
and Tcl_WaitForEvent. 
3. What I observe is that after initialization, the Tcl 
thread essentially goes into a polling mode in which 
it tries to process a WM_DEVICE_CHANGE (0x0219) 
message. 
The sequence of steps is this (Note that all messaging 
here is done on the thread; ie, no window is involved.): 


* Tcl_DoOneEvent is called. 
* Tcl_WaitForEvent is called. 
* PeekMessage is called and finds a 
WM_DEVICE_CHANGE with 
the wParam set to DBT_DEVNODES_CHANGED (0x0007). 
* GetMessage is called and finds the 
WM_DEVICE_CHANGE msg. 
* TranslateMessage is called and posts back the 
WM_DEVICE_CHANGE msg to the messaging queue for 
the 
thread. 
* DispatchMessage is called with the 
WM_DEVICE_CHANGE msg. 


At this point, program execution goes back to 
Tk_MainLoop, and then Tcl_DoOneEvent is called and 
the whole sequence of steps continues ad infinitum. 

NOTE: I have observed this same behavior for
an undefined message with ID 0x0118.  To correct
this infinite message-processing loop, I inserted
the following code in tcl8.4.9/win/tclWinNotify.c:

        result = GetMessage(&msg, NULL, 0, 0);
        while (msg.message == WM_DEVICECHANGE || 
msg.message == 0x0118) {
          result = GetMessage(&msg, NULL, 0, 0);
        }

This "worked", in the sense that said inifinite
processing loop no longer occurs.

Although I'm not sure, I believe that launching a Tcl 
thread 
in this manner triggers the WM_DEVICE_CHANGE 
message, and 
that the Tcl_DoOneEvent processing logic just doesn't 
handle 
this case gracefully.  IE, things "still work", but you 
sacrifice a huge amount of CPU horsepower in the 
process. 

kodeCharlie
User Comments:
hobbs added on 2005-02-02 12:33:23:
Logged In: YES 
user_id=72656

Not enough information is given in the above trace.  The
TranslateMessage and DispatchMessage must somehow be getting
into Tk, which isn't returning an expected error code. 
Otherwise, why does it come back to Tcl unless we are
receiving multiple such messages?

davygrvy added on 2005-02-02 16:42:57:
Logged In: YES 
user_id=7549

Maybe a window within the thread that is owned by macromedia
is the one reposting?

dkf added on 2014-02-23 16:52:14:

0x0118 is “the undocumented WM_SYSTIMER, which appears to be used for caret blinks”.