Re: Variable key replaces original warnings

[Richard Purdie <richard.purdie@linuxfoundation.org>]

On Fri, 2015-07-31 at 10:54 +0100, Richard Purdie wrote:
> On Thu, 2015-07-30 at 17:03 +0200, Olof Johansson wrote:
> > Hi all,
> > 
> > We've started to see issues on master like
> > 
> >    WARNING:
> >      Variable key          USERADD_PARAM_${PN}             (; --system --home / --no-create-home --comment 'Storage manager daemon' --gid storage storage)
> >      replaces original key USERADD_PARAM_recording-indexer (--system --home / --no-create-home --comment 'apache httpd' --gid www www).
> > 
> > (my alignment...)
> > 
> > The warning happens when doing
> > 
> >  WWWUSER_PACKAGE ?= "${PN}"
> >  USERADD_PARAM_${WWWUSER_PACKAGE} := "--system --home / --no-create-home --gid www www"
> > 
> >  USERADD_PARAM_${PN}_append = "; --system --home / --no-create-home --gid username username"
> > 
> > (in reality, the first two lines are inherited from a class, and the third in
> > the recipe itself.)
> > 
> > With bitbake -e I see what looks like the expected value, with both users being
> > created. If I change USERADD_PARAM_${WWWUSER_PACKAGE} (second line) to
> > USERADD_PARAM_${PN} no warning is printed, so this only seems to be a problem
> > in cases where multiple unique non-expanded vars expand to the same expanded
> > var.
> > 
> > I made a unit test that should demonstrate the problem (I think), and it fails
> > because of the warnings:
> > 
> > diff --git a/bitbake/lib/bb/tests/data.py b/bitbake/lib/bb/tests/data.py
> > index e9aab57..e7716cc 100644
> > --- a/bitbake/lib/bb/tests/data.py
> > +++ b/bitbake/lib/bb/tests/data.py
> > @@ -386,6 +386,15 @@ class TestKeyExpansion(unittest.TestCase):
> >              self.assertTrue(logContains("Variable key VAL_${FOO} (A) replaces original key VAL_foo (B)", logs))
> >          self.assertEqual(self.d.getVar("VAL_foo", True), "A")
> >  
> > +    def test_append(self):
> > +        self.d.setVar("TEST_${BAR}", "Bar")
> > +        self.d.setVar("TEST_${FOO}_append", "Foo")
> > +        with LogRecord() as logs:
> > +            bb.data.expandKeys(self.d)
> > +            self.assertFalse(logContains("Variable key TEST_${FOO} (Foo) replaces original key TEST_foo (Bar)", logs))
> > +        self.assertEqual(self.d.getVar("TEST_${FOO}", True), "BarFoo")
> > 
> > 
> > If you have any ideas on what this issue can be or how to fix it, please
> > let me know :)
> 
> The data store changes recently caused a number of these to appear. The
> issue has always been there, as has this warning, the new datastore code
> just exposes it a bit more consistently than previously.
> 
> The reason we do this is that users with conflicting (different values)
> of the same variable name usually have a metadata issue they're unaware
> of.
> 
> The fix is to ensure you only set a variable in one canonical way.
> OE-Core did show a few of these but I fixed them before merging the data
> store changes.

The OE-Core change I made for a similar case was:

http://git.yoctoproject.org/cgit.cgi/poky/commit/meta/classes/update-rc.d.bbclass?id=3b627bb28c4ea2ea33050ad4884c6351e2d6ebad

i.e. moved to anonymous python. Not ideal, but more explicit.

Cheers,

Richard