Subj : Re: Exceptions and Object Lifetimes
To   : Pete Diemert <pete.diemert@purposebuilt.com.au>
From : Brendan Eich <brendan@meer.net>
Date : Tue Dec 16 2003 04:50 pm

Pete Diemert wrote:

>1)  Exceptions - We have exposed our own object/funcs and so, would also
>like to expose errors that our functions can generate as catchable
>exceptions in JS.  However, from looking at JSAPI.H we cannot see a clear
>way to "throw" an exception from the host.  We have tried to add entries to
>JS.MSG and marked them as JSXN_ERR which seems to cause the exception object
>to get generated in scope but alas nothing is catching in JS.  Do we have to
>actually construct JS script to do the throw then evaluate the script in
>scope in order to get exceptions?
>  
>

I saw your followup saying you fixed your code.  That's good, but it's 
bad that you still have to hack js.msg.  That's our bug -- see 
http://bugzilla.mozilla.org/show_bug.cgi?id=215173.

>2)  Object Lifetimes - To make the object model in JS coherent with our own
>internal object model we allow the JS object to keep a "reference count" on
>our internal object so that JS controls its lifetime.  This works fine but
>we notice that the GC is VERY lazy and will choose to keep objects alive
>after they are clearly never going to be accessed again, waiting for
>shutdown or mem threshold, e.g.:
>
>function foo()
>{
>    var x = new SomeObject;
>    // x drops out of scope after func, never to be seen again but object
>ref will remain
>}
>
>  
>

That's how GC works, in general -- you shouldn't be depending on prompt 
finalization.

You should, however, run the GC more often, especially if there may be 
garbage to collect.  JS_MaybeGC from the branch callback every 2^n 
backward branches or returns, for some n around 12 or 13, and JS_MaybeGC 
or even JS_GC when you're destroying any large "container" object (such 
as browser windows, in the DOM embedding), are good ideas.

>The object does indeed get cleaned up at shutdown but we have a system that
>may instantiate some fairly big footprint objects such as forms, databases,
>recordsets, etc...  Understanding that the GC reserves the right to
>arbitrarily clean these objects is there some way we can get a callback
>stating "object no longer in use by any call object" in lieu of a
>JSFinalize() so that we can force the GC ourselves and prevent a rather
>large uneeded working set?
>  
>

No, that would require reference counting.

If you are tying up scarce resources, you need to require script writers 
to call a "release" method.  There's no substitute.  If you come to 
depend on a JS engine that does use ref-counting, you're depending on 
something not guaranteed by the language spec.  So a mandatory method is 
the way to go.

/be

.