Re: libobjc2 1.8.1 test failures while building on Canonical's buildfarm

[David Chisnall]

On 27 Dec 2015, at 01:11, Ivan Vučica <address@hidden> wrote:
> 
> (1) Is this a known and expected bug in clang-3.5?

Not sure about 3.5, but some clang releases don’t emit quite the right property 
introspection metadata.  I thought that we’d worked around the test failure, 
but perhaps not.

> (2) Is clang-3.5 actually unsupported? 

Yes, pretty much everything will work, but code that uses introspection on 
properties won’t have quite the 

> (3) Can I work around these failures somehow?

Can you try with the version of the test from master?  I made some fixes to 
that test this week and found a few bugs (in particular, it was broken on 
32-bit x86, which might be the issue that you’re hitting).  Of course, in 
fixing the test I 

> (4) Can I disable tests somehow, and instead declare "this code is known to 
> be broken; use at your own risk"? (I'd do this only for versions of Ubuntu 
> that don't ship with a sufficiently new clang.)

Yes, -DTESTS=OFF on the cmake command line ought to do it.

> (5) If not I could upload debian source package for whatever minimum version 
> of clang you believe is required for libobjc2 1.8.1 and have Canonical's 
> build farm sort things out. What version is required?

I think 3.4 or later should work.  Newer is generally better.  We use the 
system clang (3.4) on FreeBSD 10.

Once the ARM issue is fixed (and possibly once I’ve added AArch64 code paths, 
if I get around to it), I’m planning on pushing out a 1.9 release soon.  It 
might be worth waiting for that.

David

-- Send from my Jacquard Loom