[pacman-dev] [PATCH 1/7] libmakepkg: disallow empty arch

[Morgan Adamiec]

On Wed, 20 Jun 2018 at 00:52, Allan McRae <allan at archlinux.org> wrote:
>
> On 20/06/18 05:54, Eli Schwartz wrote:
> > On 06/19/2018 08:28 AM, Allan McRae wrote:
> >> On 09/06/18 05:33, Eli Schwartz wrote:
> >>> On 06/08/2018 02:18 PM, morganamilo wrote:
> >>>> We already ensure arch is an array but if arch is never defined
> >>>> this error never triggers. Add an explicit check for a missing arch.
> >>>
> >>> Good catch! We'd usually abort with
> >>> ==> ERROR: foo is not available for the 'x86_64' architecture.
> >>>
> >>> but not if we're doing some IGNOREARCH operation that doesn't check
> >>> this, e.g. --source or --printsrcinfo or --packagelist
> >>>
> >>> Or, makepkg --ignorearch.
> >>>
> >>
> >> Hrm...  I just saw this after replying the patch was OK.
> >>
> >> With this patch:
> >>
> >> $ /home/allan/arch/code/pacman/scripts/makepkg --ignorearch
> >> ==> ERROR: arch is not allowed to be empty.
> >>
> >>
> >> I think our current check is OK here and this additional check is too
> >> stringent.
> >>
> >> Allan
> >
> > I'll agree with morganamilo here, if someone uses arch=(x86_64) and I
> > --ignorearch it to use it on armv7h, that's one thing, but if they don't
> > define it at all then --printsrcinfo succeeds but makepkg on its own fails.
> >
> > IMHO that makes the patch worthwhile for linting --printsrcinfo, and in
> > the process if it results in --ignorearch failing for the same issue,
> > then I think it's a good thing on the grounds of technical correctness.
> >
> > The PKGBUILD is fundamentally wrong to be missing required variables, it
> > shouldn't be "okay as long as you call makepkg with certain switches to
> > infer the arch", it doesn't work in the common case. Therefore it should
> > be fixed, and we should ensure that --printsrcinfo fails early.
> >
> > Yes, anyone who tries to build the package will notice the issue. But
> > unfortunately many people who try writing PKGBUILDs will only ever run
> > --printsrcinfo and not try building it -- this patch prevents that from
> > occurring.
> >
>
> man makepkg:
>
> -A, --ignorearch
> Ignore a missing or incomplete arch field in the build script.
>
> The documentation defines the behaviour of --ignorearch as currently
> specified.
>
> Allan

While --ignorearch is documented correctly, flags such as
--printsrcinfo will not complain about a missing arch, which is
defined as a required field. I also believe --ignorearch should be
changed to not ignore a missing arch either.

Anyway, tonight I will start working on v2 of this patch set. I'll
remove this commit unless you change your mind.