Re: [Tcl-soc2008] Problem with Tcl_EvalObjv

[Jeff H. <je...@ac...>]

Remigiusz Jan Andrzej Modrzejewski wrote:
> Not the first, probably not the last - problem with Tcl_EvalObjv. Now 
> I've got this here:
> http://dev.lrem.net/tcldtrace/browser/trunk/generic/dtrace.c#L680
> When executing this test:
> http://dev.lrem.net/tcldtrace/browser/trunk/tests/dtrace.test#L484
> I get a segfault. This is during the second call of callback. Everything 
> looks quite the same as in the first call. Some gdb dump:

691 	static int listProbe (
692 	        dtrace_hdl_t *handle,
693 	        const dtrace_probedesc_t *pdesc,
694 	        void *varg
695 	        )
696 	{
697 	    list_arg *arg = (list_arg*) varg;
698 	    Tcl_Obj *objv[3];

int i;

699 	
700 	    objv[0] = arg->proc;
702 	    objv[1] = formatProbeDesc(arg->hd, pdesc);
704 	    objv[2] = arg->args;

for (i = 0; i < 3; i++) { Tcl_IncrRefCount(objv[i]); }

706 	    if (Tcl_EvalObjv(arg->hd->interp, 3, objv, 0) != TCL_OK) {
707 	        /* What now?! */

Tcl_AddErrorInfo(arg->hd->interp, "\n    (running listProbe callback)");
Tcl_BackgroundError(arg->hd->interp);
return -1; // ERROR CONDITION RESULT??

708 	    }

for (i = 0; i < 3; i++) { Tcl_DecrRefCount(objv[i]); }

710 	    return 0;
711 	}

> The test should fire the callback two times for probes 
> syscall::exec:entry and syscall::exec:return.
> 
> BTW - how should I signalize an error within the callback? Like the 
> failed Tcl_EvalObjv (if it would ever fail without bringing the whole 
> thing down)?

I chose background error as that does the bgerror style reporting, as is 
probably necessary in this error situation.  Otherwise you need to set 
the interp result and trigger a TCL_ERROR return code.

Jeff