Arch Linux

This is a makepkg bug, not a core/linux bug. util/pkgbuild.sh provides extract_function_variable, and the attr_regex used does not account for valid bash like `conditional && groups=('base')`.

It assumes that the variables it `grep`s for are preceded by nothing other than whitespace.

This is not a makepkg bug; it is an inherent limitation. makepkg has to actually parse the PKGBUILD and it will never be able to parse all valid bash. The linux PKGBUILD is simply too complex for --printsrcinfo to parse correctly. The only real solution is to stop trying to make the linux PKGBUILD reusable for arbitrary kernel builds and simplify it.

I didn't really expect much as far as actual proper parsing of bash. :D

However, this raises an interesting question. Currently, makepkg says the wrong thing when a variable follows a conditional. But even with Erich's proposed fix, this would still be wrong when the pkgbase is changed.

So, should makepkg err on the greedy side, and read more potential variables, or on the conservative side, and read fewer? If the former, maybe the regex should be extended to match anything. If the latter, maybe we should look for `if` conditionals and exclude them...

I'm inclined to think, if anything, we should be greedy since the practical purpose of .SRCINFO is allow e.g. AUR resolution, proper dependency ordering -- more, possibly extraneous information is better than missing information.

I think, the proper solution would be to disallow (or discourage) the whole complexity of bash in PKGBUILDs -- like it's already done for something like

if [ $CARCH = 'x86_64' ]; then
depends+=(lib32-something)
elif [ $CARCH = 'i686' ]; then
depends+=(libsomething)
fi

Meanwhile, I'm in favour of making makepkg more "greedy" as eschwartz proposed.

Discouraging PKGBUILDs that try to do too much here is the only solution. Makepkg will not do more heavy parsing.