W3C home > Mailing lists > Public > public-media-capture@w3.org > April 2013

Error handling - what to change

From: Harald Alvestrand <harald@alvestrand.no>
Date: Wed, 17 Apr 2013 14:02:12 +0200
Message-ID: <516E8F44.5050308@alvestrand.no>
To: "public-media-capture@w3.org" <public-media-capture@w3.org>
I've been working for a while on getting together a specific proposal 
for what to do with error handling.
I don't think we need to do that much..... what I propose is below.

If you like commenting in Google docs, go here:


If you like nicely formatted PDF, look at the attachment.

Otherwise, just keep on reading...


We keep to our general approach of errors, exceptions and error events.
We try to align as closely as possible to the DOM specification of these 
GetUserMedia spec:

- Change the NavigatorUserMediaError to be a subclass of DOMError. This 
means just that the "name" attribute is specified in the superclass. 
"message" is referred to, but not defined in, DOMError, so its status 
seems unclear. If this is deemed to be present, the only extra attribute 
is "constraintName", which is gUM specific.
- Change the enum NavigatorUserMediaErrorName to have error names in the 
style of section 3.4 of the DOM spec: CamelCase with uppercase initial.
- Ask the DOM team to add PermissionDenied and ConstraintNotSatisfied to 
their error list (per their expressed desire). No corresponding code.

- In text, state that all exceptions thrown are DOMExceptions.
- Change all references to codes, like "INVALID_STATE_ERR", to be the 
equivalent strings, like "InvalidStateError" (this is the only exception 
thrown in this spec.)

PeerConnection spec:

- Remove RTCError and make RTCSdpError a subclass of DOMError. Replace 
all references to RTCError with DOMError.
- Change the enum RTCExceptionName and RTCErrorName to have error names 
in the style of section 3.4 of the DOM spec, or remove them once 
SessionDescriptionError is registered in the DOM spec.
INVALID_MEDIASTREAM_TRACK. We should use the Javascript builtin 
TypeError instead, as per DOM guidance.
- Use DOM's InvalidStateError instead of INVALID_STATE
- Use DOM's InvalidModificationError instead of INCOMPATIBLE_CONSTRAINTS
- Ask the DOM team to register a SessionDescriptionError instead of 

- In text, state that all exceptions thrown are DOMExceptions, except 
for TypeError.

Present spec: error handling
There are 3 different things in the spec that happen because of what 
we'd call errors. Some of them may happen for other reasons as well 
(like events signalling a non-error situation), but let's focus on the 
error situation.

Error callbacks are called when a specific request to the API results in 
an unsuccessful outcome. The normal case should then be that the state 
of the underlying object does not change, and the error happens exactly 
once per request.
Exceptions are thrown when something bad happens in an API call. A 
typical case is the "type error" mismatch, usually generated from 
libraries (DOM Error 14). When an exception is thrown, the state of the 
object should be unchanged.
Events are dispatched at the objects presented at the API in various 
situations. These events have to carry information enough to guide the 
app in what to do about them, and some of them need diagnostics - such 
as the SDP line or construct that caused an error in a 
SetLocalDescription call.

Being inventive is rarely good. If we can be consistent with other 
specs, programmers will have a better time learning our APIs, and will 
use them more consistently.

Other specifications we should align with
The most important specs to align with are the HTML5 spec and the DOM spec.
 From the HTML5 spec, we inherit the whole Event mechanism including the 
EventHandler interface.

 From the DOM spec, we should align with errors.

DOMError is here:
It is intended for "specs that want to introduce error handling by other 
means than exceptions" - the only defined member is a name field.

the DOM spec also describes the EventTarget:

DOM exceptions are here:
The core exception definition is here:

The DOM Exception has a "name" field and a "message" field, inherited 
from the WebIDL spec 
(http://dev.w3.org/2006/webapi/WebIDL/#idl-exceptions), in addition the 
IDL says that it has a "code" - it doesn't say what the code is when 
there is no code for a message.

Received on Wednesday, 17 April 2013 12:02:46 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 6 January 2015 21:24:40 UTC