Comment 9 for bug 1564571

OK then, my environment must've gotten weird by the time I've noticed the -soname missing.

It is a fact, though, that new cmake caused something to change since this was working before. I can't currently say what that is.

And it's not a change in how the test is invoked - below's a diff between the last passing and failing runs (http://autopkgtest.ubuntu.com/packages/u/unity8/xenial/amd64/) - 3.5 did add another LD_LIBRARY_PATH (I suppose that's an issue in our cmake module), but they actually both have the same value, and I've verified this doesn't even matter anyway, since the link is done via a relative path, so LD_LIBRARY_PATH doesn't really matter at all...
http://pastebin.ubuntu.com/15575765/

It's not about the library path after all, but rather about QML2_IMPORT_PATH - it's fine if it's using the locally built one - fails if using the installed one.

FWIW we *did* want to use the installed ones in autopkgtest (a local one shouldn't even be built - our issue).

As for SONAME being set in the .so - I believe the problem only appears after you've packaged the file and it was stripped. And SHARED or MODULE doesn't seem to make a difference now.

I've now compared the output from objdump for libUbuntuGestures.so as installed from https://launchpad.net/ubuntu/+source/unity8/8.12+16.04.20160323.2-0ubuntu1 and https://launchpad.net/ubuntu/+source/unity8/8.12+16.04.20160330-0ubuntu1

And here's the twist: the previous one, built with cmake 3.2, does have the SONAME, while the new one, built with cmake 3.5.1, does not. So I think this might actually be fallout of both building the packages *and* using cmake 3.5.1 during testing...

TBH I'm quite lost with this. Am looking more into what's happening.

In the mean time, we're skipping this test to unblock things.