reads wrong config file while checking BUILD_DEPENDS availability

[anbe42]

While investigating a spurious autopkgtest failure in Debian (of the only(?) dkms package in Debian using BUILD_DEPENDS), I noticed that dkms claims the build dependency is not available after it has been successfully built and installed immediately before this check.

So far I haven't been able to reproduce that in run_tests.sh

It could be a recently introduced regression, since I think the test succeeded in the past. But it could also be a very fragile path that leads to an error here. If there is a second kernel, the failure seems to be only on the first.

What I noticed so far is:

• is_module_built gets 4 arguments and while checking availability of a build-dependency, $1/$2 refer to the B-D being checked and differ from $module/$module_version. But when is_module_built calls read_conf_or_die, this happens with the implicit $module/$module_version parameters and a $conf override inherited from somewhere. As a result, some wrong module gets checked for being built returning a failure.
• If I prefix that command with module=$1 module_version=$2 and the correct dkms.conf (from the build-dependency) gets loaded, this modifies the internal state (BUILT_MODULE_NAME MAKE etc.) and the whole thing explodes during build.

[scaronni]

@anbe42 do you want to tackle this in the short term or can I proceed with tagging and releasing 3.1.6?

[scaronni]

Nevermind, I'll just tag the release, let's address it after. thanks.