There is a race condition between when the CONFIG_SITE list is generated and
then used here via the cat to create the config.cache and in *all* configure
scripts when reading from CONFIG_SITE.
The race in this case is that the sstate setscene task runs on a package that
contains a site config file (ncurses in this case) and then bitbake decides
that it needs to rebuild and cleans out the site config file, so it existed
for siteinfo_get_files() to find in SITECONFIG_SYSROOTCACHE and then was removed
for the rebuild. When bintuils tried to run the do_configure() task which creates
the binutil's version of config.cache it reads from CONFIG_SITE which now contains
the non-existant site config file. (confused yet ;-)?
Currently the configure script does a test -r to ensure the file is readable
before using it, therefore having the cat succeed regardless of the file being
available is consistent behaviour.
(From OE-Core rev: ffd8f05e5548500199c9b04a174067811ad2c5e7)
Signed-off-by: Saul Wold <sgw@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
39214157da