Ensure autoinstall reads /etc/dkms/$module.conf

[bodgerer]

"dkms autoinstall" does not obey BUILD_DEPENDS (and presumably AUTOINSTALL) directives when placed in /etc/dkms/$module.conf, causing unexpected ordering and build failures.

The read_conf family of routines rely upon variables module and module_version (possibly others?) being already suitably set. Arguably they shouldn't, but this is a minimal change that corrects autoinstall's BUILD_DEPENDS processing.

There may be other places where this is needed.

Tested against dkms 3.1.4

[scaronni]

I think these are the only two variables required, as the function needs to run while iterating over $m + $v. It's also the only case where that function is called in a loop.

Thanks for the merge request!

[bodgerer]

Hi Simone, No problem : ) There may be another case lurking - I think I saw some attempts to read /etc/dkms/.conf from the "while read status mvka; do" loop, but didn't figure out where it was coming from Best, Mark

[bac0n]

I think parsing of /etc/dkms/*.conf should unset the complete array before trying assign new directives, as of now you have to match the number of array members as it assigns array members by index (basically merging), or you can get really odd behavior, e.g., overriding BUILT_MODULE_NAME or some other array, isn't really possible. It's couple of rev. behind, but I rewrote safe_source to unset the directives (arrays) before declaring them (I think it's a more sane/expected bevavior):

safe_source(){
    # $1 = file to source
    # $@ = environment variables to echo out
    mapfile -t < <(
        local -a "${@:2}"
        . "$1" > /dev/null 2>&1
        for e in "${@:2}"; do
            for i in $(eval echo "\${!$e[@]}"); do
                v=${e}[$i]
                printf '%s[%s]="%s"\n' "$e" "$i" "${!v}"
            done
        done
    )

    unset "${MAPFILE[@]%%[[=]*}" && \
    eval "${MAPFILE[@]}" || {
        echo "Error! Could not parse .conf directives." >&2
        return 1
    }
    # ....
}

[anbe42]

@bac0n Please file a new issue or PR, otherwise this comment on a closed PR tends to get forgotten ... Thanks!