Re: [Twisted-Python] Twisted 16.3.0 Prerelease 2 Announcement

2016-07-12 Thread Paweł Miech
> In an earlier e-mail you mentioned that you were using Python 3. Is that still true? I can reproduce this in Python 2.7.11 and Python 3.5.2. In both of them Chrome responds with ERR_SPDY_INADEQUATE_TRANSPORT_SECURITY. When I test with curl with verbose flag I see that it also shows information

Re: [Twisted-Python] Twisted 16.3.0 Prerelease 2 Announcement

2016-07-12 Thread Cory Benfield
> On 11 Jul 2016, at 22:04, Paweł Miech wrote: > > This seems to suggest that Ubuntu 16.04 (the system I'm testing) does not > support ciphers required by HTTP2. But nginx article about HTTP2 lists ubuntu > as only linux like system that is able to support HTTP2 over ALPN which is > required

Re: [Twisted-Python] Twisted 16.3.0 Prerelease 2 Announcement

2016-07-12 Thread Cory Benfield
> On 11 Jul 2016, at 20:22, Glyph Lefkowitz wrote: > > So pyOpenSSL/Cryptography doesn't have SSL_get_current_cipher anywhere? get_current_cipher isn’t helpful. In particular, it puts us in an awkward place where we have a connection that has been negotiated for HTTP/2, but we cannot use it.

Re: [Twisted-Python] Twisted 16.3.0 Prerelease 2 Announcement

2016-07-12 Thread Tristan Seligmann
On Tue, 12 Jul 2016 at 09:43 Cory Benfield wrote: > For that reason, you should adjust your code to use > OpenSSLCertificateOptions or, even better, use the TLS endpoint directly. > > The exported name of this class is actually just "CertificateOptions", fwiw.

Re: [Twisted-Python] Twisted 16.3.0 Prerelease 2 Announcement

2016-07-12 Thread Paweł Miech
> DefaultOpenSSLContextFactory should have been deprecated a long time ago. It’s insecure, and in particular does not set a cipher string, so it uses DEFAULT. That will have all kinds of messed up priorities. For that reason, you should adjust your code to use OpenSSLCertificateOptions or, even bet

Re: [Twisted-Python] Twisted 16.3.0 Prerelease 2 Announcement

2016-07-12 Thread Cory Benfield
> On 12 Jul 2016, at 09:33, Paweł Miech wrote: > > If you google for "ssl in twisted" you will also find articles that recommend > it. Since so many people use it, maybe it could be updated to be more secure? > If it does not make sense to update it then perhaps it would be good to > deprecat

Re: [Twisted-Python] Twisted 16.3.0 Prerelease 2 Announcement

2016-07-12 Thread Paweł Miech
> Agreed. I’m planning to begin the deprecation process, though it will take a little while as we need to remove all uses of it from within the Twisted codebase itself, as well as from the documentation. That turns out to be a bigger task than expected! +1 One final point that I glossed over earl

Re: [Twisted-Python] Twisted 16.3.0 Prerelease 2 Announcement

2016-07-12 Thread Cory Benfield
> On 12 Jul 2016, at 17:42, Paweł Miech wrote: > > > Agreed. I’m planning to begin the deprecation process, though it will take > > a little while as we need to remove all uses of it from within the Twisted > > codebase itself, as well as from the documentation. That turns out to be a > > big

Re: [Twisted-Python] Twisted 16.3.0 Prerelease 2 Announcement

2016-07-12 Thread Glyph Lefkowitz
> On Jul 12, 2016, at 12:43 AM, Cory Benfield wrote: > > DefaultOpenSSLContextFactory should have been deprecated a long time ago. 2 years ago, to be precise: https://twistedmatrix.com/trac/ticket/6923 Someone fixing this would be tremendously useful. -glyph __

[Twisted-Python] removing twistedchecker buildbot for the time being

2016-07-12 Thread Glyph Lefkowitz
Right now it seems the difference-computation logic on the twistedchecker buildbot has just broken completely. It's introducing useless noise into the build results because it makes every actually-passing build into a big red 'X' on the pull request status page. I think I'm going to remove it.

Re: [Twisted-Python] removing twistedchecker buildbot for the time being

2016-07-12 Thread Adi Roiban
On 13 July 2016 at 00:37, Glyph Lefkowitz wrote: > Right now it seems the difference-computation logic on the twistedchecker > buildbot has just broken completely. It's introducing useless noise into > the build results because it makes every actually-passing build into a big > red 'X' on the pu