Zhenbin Xu wrote:
-----Original Message-----
From: Jonas Sicking [mailto:[EMAIL PROTECTED]
Sent: Friday, June 20, 2008 2:05 PM
To: Zhenbin Xu
Cc: Sunava Dutta; Ian Hickson; [email protected]; IE8 Core AJAX
SWAT Team
Subject: Re: <New: Tracking Issues in XHR that we raised>RE: <Was:
Further LC Followup from IE> RE: Potential bugs identified in XHR LC
Test Suite
Zhenbin Xu wrote:
-----Original Message-----
From: Jonas Sicking [mailto:[EMAIL PROTECTED]
Sent: Friday, June 20, 2008 1:19 PM
To: Zhenbin Xu
Cc: Sunava Dutta; Ian Hickson; [email protected]; IE8 Core AJAX
SWAT Team
Subject: Re: <New: Tracking Issues in XHR that we raised>RE: <Was:
Further LC Followup from IE> RE: Potential bugs identified in XHR LC
Test Suite
Zhenbin Xu wrote:
Jonas, I don't feel you have summarized our position properly. We
said it should be exception but we are willing to accommodate other
implementations for the spec to have a leeway there and avoiding
protracted discussions.
I assume you mean by the "null or exception" proposal? I think many
people have made it quite clear that they think that is the least
good
solution as it doesn't produce interoperability across browsers.
We have absolutely no problem for the spec to
clearly state that exception is the best API that should be
followed.
Hehe, yes, that has been quite clear :)
It is backed by technical arguments on my replies. Let's expand
more
there if you feel those are inadequate.
Yes please do, I'm curious as to what those technical arguments are.
The
one I've heard so far is concern about site compatibility with
returning
null. I.e. you guys are concerned that sites will break if an
exception
isn't thrown.
[Zhenbin Xu] We are concerned because returning null is not a
consistent,
predictable programming model. It is a deviation from other part of
the XHR
design, as well as the state machine approach that entire spec is
based on.
Which other parts of the spec is it inconsistent with. I definitely
agree that consistency is important. Looking at the spec it seems like
returning null is consistent with .responseText, but not consistent
with
.status and .getResponseHeader().
Ideally we should have consistency across all these properties.
[Zhenbin Xu] .statusText, .getAllResponseHeaders(), send() all raise
INVALID_STATE_ERR as well.
So if we can change .responseText to also throw an exception then I'd be
fine with having .responseXML also throw.
/ Jonas
