bitbake: data_smart.py: clear expand_cache in _setvar_update_overridevars

At the end of this function, self.overrides is cleared, which means we'll
construct a new self.overrides after this call. And a new self.overrides
will give out different expand_cache, so the old expand_cache should also
be cleared to avoid any wrong value.

Currently, there's a problem revealed by recent recipe specific virtual
provider patch. If we enable multilib and set "OVERRIDES:prepend" in local.conf,
things don't work.

Here's the error message:

  ERROR: Nothing PROVIDES 'lib32-gcc-cross-x86_64'

Below are reproduce steps:
1. Add in local.conf the following lines:
   MACHINE ?= "qemux86-64"
   require conf/multilib.conf
   MULTILIBS ?= "multilib:lib32"
   DEFAULTTUNE:virtclass-multilib-lib32 ?= "core2-32"
   OVERRIDES:prepend = "some-override:"
   (Note that using :append and :remove also reproduces the issue.)
2. bitbake -n lib32-sysstat
   (bitbake -n core-image-minimal also reproduces the issue)

The expandWithRefs calls getVar, which fills expand_cache. So when setting
OVERRIDES:prepend, this will fill the expand_cache. When overridevars are updated,
if we don't clear expand_cache, we'll retrieve wrong values.

Previously, things happened to work because there's a call to expand
PREFERRED_PROVIDER_virtual/${TARGET_PREFIX}gcc, which in turn expands
TARGET_VENDOR. Now what we expand is PREFERRED_PROVIDER_ virtual/cross-cc,
so the problem is revealed.

(Bitbake rev: 7375d32e8c1af20c51abec4eb3b072b4ca58b239)

Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Signed-off-by: Steve Sakoman <steve@sakoman.com>

bitbake/lib/bb/data_smart.py

@@ -635,6 +635,7 @@ class DataSmart(MutableMapping):
                 nextnew.update(vardata.contains.keys())
             new = nextnew
         self.overrides = None
+        self.expand_cache = {}
 
     def _setvar_update_overrides(self, var, **loginfo):
         # aka pay the cookie monster