Trying to reproduce bug 766817 with http://www.websocket.org/echo.html. Using squid configured for authentication (standard config). When clicking the [connect] button, I'm getting assertion failure at http://hg.mozilla.org/mozilla-central/annotate/10e019421e6b/netwerk/protocol/http/nsHttpChannel.cpp#l1002
"SSL connect failed but not using SSL?");
Http status is 403, host is echo.websocket.org, port 80. mSSLConnectFailed is set to 'true' on the transaction.
Candidate for bug 715905?
it might be as simple as the ABORT being superfulous. iirc websockets uses connect to make a websockets tunnel through the proxy.. That path is the failed-to-make-tunnel path.. off the top of my head the error handling is going to be the same even in the absence of SSL.. but it obviously needs more rigor than that.
Created attachment 635841 [details] [diff] [review]
Patrick's comment is right about the issue. I grepped through the code to audit all uses of "SSL" and found a few other bugs, and a lot of stuff that should get renamed. Filing separate bugs for those so we can sort out which deserve aurora/beta, if any. Not sure if this one does, for instance (doesn't affect release builds).
Whoops--since this depends on bug 713026 and that isn't making it to aurora/beta, it's moot to consider backporting this as well.
Comment on attachment 635841 [details] [diff] [review]
Review of attachment 635841 [details] [diff] [review]:
the assert reason should be updated.. and maybe the general comments for the function too - I wouldn't go too crazy chasing the rabbit down the hole here though.
Comments for the assertion and the function generally got updated in bug 767516.
Push backed out for M5 orange:
Couldn't see you on #developers or see any try run URLs listed in-bug in the four bugs in the push, or I would have tried to find out if just part of it could have been backed out.
Jason, could we land just this one single patch?