On Wed, 2017-02-01 at 10:40 +0000, Robbie Gemmell wrote:
> On 18 January 2017 at 11:05, Robbie Gemmell <[email protected]
> > wrote:
> > On 18 January 2017 at 01:21, Alan Conway <[email protected]>
> > wrote:
> > > On Tue, 2017-01-17 at 14:20 -0500, Alan Conway wrote:
> > > > On Wed, 2017-01-11 at 10:54 -0500, Andrew Stitcher wrote:
> > > > > On Wed, 2017-01-11 at 12:32 +0000, Robbie Gemmell wrote:
> > > > > > ...
> > > > > > Running it again on the 14.04.4 instance, node ubuntu-2, it
> > > > > > failed
> > > > > > as
> > > > > > before (seemingly using Go 1.6, vs 1.6.2 on the 16.04.1
> > > > > > instance,
> > > > > > which was node H12)
> > > > > 
> > > > > I only disabled the PHP binding build, so any problems with
> > > > > the go
> > > > > bindings on Ubuntu 1404 were not addressed.
> > > > > 
> > > > > We can't simply disable the go bindings as they are in active
> > > > > development, so someone with a clue will have to actually
> > > > > understand
> > > > > what the problem is (not me!).
> > > > > 
> > > > 
> > > > I'll look into this.
> > > > 
> > > 
> > > The problem seems to be specific to certain go versions. I've
> > > done some
> > > work to make the code build with golang and gcc Go compilers,
> > > that may
> > > or may not address the failure for go 1.6. If not I'll have to
> > > get an
> > > older compiler and figure it out.
> > > 
> > 
> > Two builds resulting from your commits failed, on different nodes.
> > One
> > instance was 14.4.4, the other 14.4.5, both using go 1.6.
> > 
> > Robbie
> 
> I looked/googled into this a bit given the intent to release
> imminently.
> 
> The issue seems to be with use of the -race option during the tests,
> which it appears didn't work in some versions of Go in Ubuntu:
> https://bugs.launchpad.net/bugs/1487010
> https://bugs.launchpad.net/bugs/1496503
> 
> From bindings/go/CMakeLists.txt, it appears the build is currently
> using -race if the go version doesn't match 'gccgo'.
> 
> It would seem we could do various things:
> 1. Decide its fine as-is, anyone affected can [maybe?] fix their
> installation to work, CI will continue to sporadically fail (we could
> identify all the working nodes and pin them).
> 2. Figure out a way to detect this situation automatically and not
> set
> -race when it wont work.
> 3. Add a manual flag to control whether -race is used, perhaps
> default
> it to false to get builds working (or again for CI, pin the working
> nodes).
> 4. Remove use of -race entirely.
> 5. Other?

Thanks Robbie. I don't want to stop running -race tests entirely, but
until we can auto-identify a safe setting for them, I can make them
optional default-off and enable manually in my tests. Will do that
now...


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to