122472 - When a non-normal SOAP HTTP status is returned, the response is ignored.

Status: RESOLVED FIXED

(Core :: XML, defect)

Description

[Ray Whitmer]

In several cases, a server was smart enough to return a SOAP fault saying that a
service was not known, and then returned a code 500.  The Mozilla SOAP
implementation assumed a 500 was really bad and did not return the
more-expansive body of the response.  I cannot think of a case where the http
status is really significant.  What is significant is you got a body, in which
case, it tells you everything you need, or you didn't, in which case, the
particular http response tells you little.  So we just need to stop checking the
http response.

To test this, make a soap call to a service that is not deployed.  Either you
get an error status on the call or you get back a response that may have a
fault, or you get the former behavior, or the latter.  The former is what was
happening.  The latter is what we want to have happen.

Comment 1

[Heikki Toivonen (remove -bugzilla when emailing directly)]

I have no idea how to write such a test case at this point, so could you attach
a testcase?

Comment 2

[Ray Whitmer]

Test case: Take a functioning test of the name
/mozilla/extensions/xmlextras/tests/soap*.html.  Now, go to the "encode" call
and change the uri which identifies the service being invoked (not the
transportURI) to something not likely to be valid.  Now, the server receiving
the call will have to return an error.

I have not verified that servers other than my own return the alternate http
status, so I have set up the test "soapunscramble.html" to hit my own server
that I know demonstrates the problem, which I will check in as soon as a
spurious cvs lock is cleared.

Once you have the modified test, then the "invokeAsync" method returns an error
instead of returning a fault in a normal response.  The modified behavior will
be to return the fault and complete the call successfully (no js exception, and
get a SOAPResponse containing a fault, which the examples clearly handle).

Comment 3

[Ray Whitmer]

The fix for this was batched in with a number of others that piled up when a cvs
lock wouldn't clear.