I thought that was kinda odd myself. But how does moving the include resolve the issue? Wouldn't the stub still be needed in either case?

Indeed that include should not be outside a class. Best way to do this is to put all that it in a ::config class and include it in any other role class that needs it. Otherwise it is polluting the global scope. Finally, I am not sure why RSpec is complaining about this. Why is it parsing that role anyway ?

In T78344#844493, @akosiaris wrote:

Best way to do this is to put all that it in a ::config class and include it in any other role class that needs it.

+1 that was my thought when I saw this.

In T78344#844493, @akosiaris wrote:

Indeed that include should not be outside a class. Best way to do this is to put all that it in a ::config class and include it in any other role class that needs it. Otherwise it is polluting the global scope. Finally, I am not sure why RSpec is complaining about this. Why is it parsing that role anyway ?

rspec puppet include manifests/site.pp which has:

import 'role/*.pp'

And thus include the rspec test ends up loading the phabricator role and ultimately the non existing password class :-)

Ah, OK, cause in my RSpec tests I test per module, avoiding this issue entirely. Granted I haven't gotten around to testing roles but moving all those role manifests to a module should not be too hard (famous last words)

Change 179930 had a related patch set uploaded (by Alexandros Kosiaris):
Move the top level variables in phabricator role in a class

https://gerrit.wikimedia.org/r/179930

Patch-For-Review

Change 179930 merged by Alexandros Kosiaris:
Move the top level variables in phabricator role in a class

https://gerrit.wikimedia.org/r/179930