Subj : Re: Loading code into runtime; and error reporting issues
To   : Alan Kemp <alan@alankemp.com>
From : Brendan Eich <brendan@meer.net>
Date : Wed Feb 04 2004 03:47 pm

Alan Kemp wrote:

>I traced through the code untill I finally got to
>js_DecompileValueGenerator, and the place its getting tripped up is here:
>
>/*
>* At this point, pc may or may not be null, i.e., we could be in
>* a script activation, or we could be in a native frame that was
>* called by another native function. Check pc and script.
>*/
>if (!pc)
>    goto do_fallback;
>
>Where by upon going to do_fallback the string "undefined" is returned which
>is then used as the function name.
>
>Upon further investigation I have discovered that its only script functions
>that I am calling from my c++ code that get names "undefined" in the error
>report.  If I call a funtion that then calls another function (which doesn't
>exist) the error report correctly shows the name of the function I tried to
>call.
>
>Is this expected behaviour?  Or am I likely doing something wrong elsewhere?
>

Good debugging!  Yes, this is expected.  If you use JS_CallFunctionName 
to call a function named "f" in some object, e.g., there is no script 
that generates the callee expression, as there is when you evaluate a 
script such as "f()". It's a fact, whether you call it a bug or feature, 
that is not likely to change.

If you may be calling a non-function, say an undefined property, you 
might use JS_GetProperty first to get "f"'s value, then 
JS_CallFunctionValue to invoke the value returned by JS_GetProperty if 
and only if JS_TypeOfValue says that value is JSTYPE_FUNCTION.

/be

.