Re: [PATCH] daemon: Check for HAVE_CHROOT instead of CHROOT_ENABLED.

[Mark H Weaver]

Manolis Ragkousis <address@hidden> writes:

> There is an issue with the chrooted builds on Hurd. In
> nix/libstore/build.cc CHROOT_ENABLED
> is false because HAVE_UNSHARE && HAVE_SYS_MOUNT_H are not defined on Hurd.
>
> The part that we are interested from build.cc is at line 1768. The
> part of the code in that if, is used by the daemon for chrooted
> builds. At this specific part of the daemon we only need
>  HAVE_CHROOT to be defined, which is true on hurd, for the builds to
> work. With this workaround,
> things seem to work for me until now. WDYT?

This won't be sufficient, because if you set up a chroot, you'll also
need to do some of the things within the "#if CHROOT_ENABLED" starting
on line 1997, such as setting up local instances of /dev and /etc.

Here's what I'd suggest: instead of using HAVE_CHROOT on line 1768,
instead split CHROOT_ENABLED into two separate flags: CHROOT_ENABLED and
maybe CONTAINER_ENABLED.  All existing occurrences of CHROOT_ENABLED
would initially be replaced with CONTAINER_ENABLED.  Then, define
CHROOT_ENABLED (now a newly defined name) to depend on a smaller set of
requirements, maybe just HAVE_CHROOT && HAVE_SYS_MOUNT_H.

We should assume that CHROOT_ENABLED is a requirement for
CONTAINER_ENABLED (which should ideally be made explicit in the code by
adding CHROOT_ENABLED to the right-hand-side expression for
CONTAINER_ENABLED), so there are only three cases to consider:

1. CONTAINER_ENABLED and CHROOT_ENABLED are both true.
2. CONTAINER_ENABLED is false and CHROOT_ENABLED is true.
3. CONTAINER_ENABLED and CHROOT_ENABLED are both false.

We should ensure that the code works properly in all three cases.

So, look at the code within "#if CONTAINER_ENABLED" and decide which
parts should still be run if CONTAINER_ENABLED is false but
CHROOT_ENABLED is true.

What do you think?

    Mark