User Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.6; rv:17.0) Gecko/20100101 Firefox/17.0 Build ID: 20121128204232 Steps to reproduce: I am using Mozilla/5.0 (Macintosh; Intel Mac OS X 10.6; rv:17.0) Gecko/20100101 Firefox/17.0 (FireFox 17.0.1). Open javascript console, and set a non-standard HTTP request header: xhr = new XMLHttpRequest(); xhr.open("POST", url); xhr.setRequestHeader("X-appended-to-this", "True"); xhr.setRequestHeader("X-appended-to-this", "False"); .... Do rest of xhr Actual results: Open the firebug tool to inspect the XHR request. The header "X-appended-to-this" was sent with the value "False". Of course, the server also received "X-appended-to-this" with the value "False". Expected results: I expected (based on http://www.w3.org/TR/XMLHttpRequest/#request-method) that the header "X-appended-to-this" would be sent with the value "True; False". I don't know if there is a good reason for not following the W3C draft.
Same for the XHR Living Standard.
> I don't know if there is a good reason for not following the W3C draft. Probably that it wasn't implemented that way initially back before there was ever a draft. Jonas, just flipping the last arg of nsIHttpChannel::SetRequestHeader would do the right thing here, right? (Though note the value would be "True, False".)
Yup, sounds about right. And write a test :)
Created attachment 709327 [details] [diff] [review] Patch + test case. This turns out to be not quite so simple as just changing a boolean, due to a) privileged callers, and b) overwriting default headers, but with some IRC discussion we think we've sorted it out. A new test case is included, and the old tests should all be passing (here's a good Try run, minus some known oranges: https://tbpl.mozilla.org/?tree=Try&rev=b08b1629020e).
Comment on attachment 709327 [details] [diff] [review] Patch + test case. One other thing that needs to be done: mAlreadySetHeaders needs to be cleared in Open(). See http://www.w3.org/TR/XMLHttpRequest2/#the-open-method step 18 Please add a test for that? Basicall call open(), set a header, call open() again, set the header again, then check that this overrode instead of merging. You'll probably need to test with a default-set header like Accept or something, since the second open() will wipe out the HTTP channel. r=me with that fixed and the test added.
Created attachment 709539 [details] [diff] [review] Final patch. Indeed, without that someone setting 'Accept' would get it merged after a second open() call. Added a testcase to the new test for that, and fixed the issue. Nice green run (https://tbpl.mozilla.org/?tree=Try&rev=59613714b24b).
https://hg.mozilla.org/integration/mozilla-inbound/rev/a14a65fe91d5
https://hg.mozilla.org/mozilla-central/rev/a14a65fe91d5
Benedict, do you mind if I submit your test to the spec's test suite?
No, that's fine, go ahead.
I've added this bug to the compatibility doc. Please correct the info if I'm wrong. https://developer.mozilla.org/en-US/docs/Site_Compatibility_for_Firefox_21
(In reply to :Ms2ger from comment #9) > Benedict, do you mind if I submit your test to the spec's test suite? Ms2ger, couldn't find anything that seemed relevant at https://github.com/w3c/web-platform-tests/pulls/Ms2ger - did you submit it?
No, never got around to it, I think. Feel free to do it yourself, or else I can do it at some point.
Ms2ger: submitted at https://github.com/w3c/web-platform-tests/pull/239 (somewhat rewritten) - merge?
David: Last month, I've added "If this method is called several times with the same header, the values are merged into one single request header." to https://developer.mozilla.org/en-US/docs/Web/API/XMLHttpRequest#setRequestHeader%28%29 Do you think it needs more docs?
Oh... and I missed kohei comment too. My mistake.
David: no problem, better a double check than none :-)
or
