On Sat, 22 Aug 2020 at 03:26:11 +1000, Joel Sing wrote:
> On 20-08-20 15:36:23, Aaron Bieber wrote:
> > Hi!
> > 
> > Here is an update to bring us to 1.15.
> > 
> > I have run this through a "mini bulk" - there was one failure, but the port
> > that failed (net/minio/server) builds fine by hand. The log is here:
> >   https://deftly.net/server.log
> > 
> > I have tested:
> > 
> > | Arch  | Go port build | Bulk | Apps | Tests            |
> > |-------+---------------+------+------+------------------|
> > | amd64 | Y             | Y    | Y    | ALL TESTS PASSED |
> > | arm64 | Y             | N    | Y    | Running...       |
> > | armv7 | -             | N    | N    |                  |
> > | i386  | -             | N    | N    |                  |
> > 
> > My armv7 machine is updating. I will report back once I run the build/test 
> > on
> > it.
> > 
> > I don't have an i386 machine.
> > 
> > edd@ has a diff for syncthing that I also tested (build and runtime). With 
> > this
> > diff and edd's, syncthing is back up and running!
> 
> I presume this was broken already and did not break moving from Go 1.15 beta1 
> to
> release?

Because of the quic lib syncthing is using, it's actually broken on beta1 as
well. More info here:
  https://github.com/lucas-clemente/quic-go/issues/2614
  https://forum.syncthing.net/t/syncthing-broken-by-go-1-15-quic/15467/14
  https://github.com/lucas-clemente/quic-go/issues/2727

TL;DR: The quic lib does an unsafe cast on internal crypto/tls structures which
is highly dependent on the version of Go.
 
> 
> > OK?
> 
> Thanks - I was in the process of doing this and have already tested on i386.
> ok jsing@, providing regress passes on arm64 and armv7 (and without PLIST.orig
> of course!).

Awesome, armv7 and arm64 both complete regress as expected!

arm64 has the same hit-or miss success as it did with beta1.

Reply via email to