ES4 draft: Error classes

Lars T Hansen lth at acm.org
Sun Mar 9 08:25:23 PDT 2008


On 3/9/08, Michael Daumling <mdaeumli at adobe.com> wrote:
> I think that adding backtrace information is overkill for the spec.
>  Collecting this information should be left to a debugging environment.
>
>  What I would suggest is something along the following lines. It should
>  be made clear that these properties must be present, but that the actual
>  value of these properties are implementation dependent. This creates a
>  reliable framework for returning extended error information to be used
>  in error logging or error display.
>
>  fileName
>
>  The initial value of the fileName prototype property is an
>  implementation-defined string that reflects the name of the source file
>  containing the script that created the Error instance.
>
>  The implementation of this property is optional. If not implemented, the
>  value of this property is the empty string.
>
>  line
>
>  The initial value of the line prototype property is the line number of
>  the executing code that created the Error instance. This is an integer
>  value, starting with the number 1.
>
>  The implementation of this property is optional. If not implemented,
>  than value of this property is zero.

I object to these suggestions on principle because they reveal
information about the caller of a function (the one invoking "new
Error") to the function (the Error constructor), which in turn reveals
it to arbitrary code (the code catching the exception).  Now you can
say that file name and line number is not very sensitive information,
but my claim is that it mildly reveals details about the structure of
the application and the computer the application originated on.  I
don't think this is excellent design.  (It also messes up tail calls,
and it is brittle because it only works one level up.)

Another problem is that, unless we expose this funcitonality ("grub
around in my caller and extract attributes of the code there") as a
primitive, then this is more functionality that "we" (language
implementers) can provide that "they" (language users) can't mimic.
Clearly there will be some functionality like that, but we should use
it sparingly.  And it seems clear to me that that functionality should
not be exposed as a primitive.

Better then, perhaps, to introduce a 'sourcecontext' nullary operator
that packages up this information and allows it to be passed to the
Error constructor:

  new Error("foo! this program is broken", sourcecontext)

(Inspired by __FILE__ and __LINE__ in C, of course.  And maybe those
are better names, all things considered.)

--lars

>
>  Michael
>  ----------------------------------------------
>
>
>  -----Original Message-----
>  From: es4-discuss-bounces at mozilla.org
>
> [mailto:es4-discuss-bounces at mozilla.org] On Behalf Of Lars Hansen
>  Sent: Friday, March 07, 2008 4:31 AM
>  To: Mike Shaver; es4-discuss Discuss
>  Subject: RE: ES4 draft: Error classes
>
>  > -----Original Message-----
>  > From: es4-discuss-bounces at mozilla.org
>  > [mailto:es4-discuss-bounces at mozilla.org] On Behalf Of Mike Shaver
>  > Sent: 7. mars 2008 03:34
>  > To: es4-discuss Discuss
>  > Subject: Re: ES4 draft: Error classes
>  >
>  > 2008/3/7 Ash Berlin <ash_es4 at firemirror.com>:
>  > > I forget all the problems, but from memory the main one is that:
>  > >
>  > > MyError = function() {};
>  > > MyError.prototype = Error.prototype;
>  > >
>  > > Doesn't do what you'd expect. And doing
>  > >
>  > > MyError.prototype = new Error();
>  > >
>  > > Is no good since then the filename and linenumber will be of that
>  > > line, not where you create the MyError instance.
>  >
>  > Right, and the idea of mutating an object when it's thrown in order to
>
>  > reset its stack information seems bogus.
>  >
>  > What about Error.prototype.throw, which would throw |this|, and act as
>
>  > a hook for the stack/location setting implementation behaviour?  We
>  > might want to permit an implementation to elide the
>  > Error.prototype.throw frame itself in whatever stack reflection is
>  > provided (though if E.p.throw has an internal error of some kind, it
>  > should probably be visible in the trace).
>
>  Not really endorsing any of these ideas as of yet, would like to see a
>  concrete proposal with details fleshed out.
>
>  The chances that the spec will *require* a useful backtrace (quasi-fixed
>  format and requirements about which frames may or may not be in it, say)
>  or source location information to be created for E.p.throw are probably
>  quite low.  Realistically what we could hope for is a statement of
>  intent and well-defined hooks for implementations who want to support
>  something, like the case is for the ControlInspector proposal (which is
>  optional).  Adding file/line info to errors thrown by the language
>  implementation looks easier to do across diverse runtimes, but I don't
>  know for sure.
>
>  --lars
>
>  >
>  > (We could also put it on Object.prototype, which would make
>  > |e.throw()| work for almost all values of e rather than just Errors,
>  > but that might be a bit beyond the pale.)
>  >
>  > Mike
>  > _______________________________________________
>  > Es4-discuss mailing list
>  > Es4-discuss at mozilla.org
>  > https://mail.mozilla.org/listinfo/es4-discuss
>  >
>  _______________________________________________
>  Es4-discuss mailing list
>  Es4-discuss at mozilla.org
>  https://mail.mozilla.org/listinfo/es4-discuss
>  _______________________________________________
>  Es4-discuss mailing list
>  Es4-discuss at mozilla.org
>  https://mail.mozilla.org/listinfo/es4-discuss
>



More information about the Es4-discuss mailing list