Might be needed for ext4 filesystems formatted with the metadata_csum
option.

ref: https://bugs.archlinux.org/task/54091

Consistently handle hooks as symlinks by relying on the exit status of
readlink instead of an lstat and always, additionally, examine the hooks
basenames to ensure that we aren't warning about foo deprecating foo
(because the hook is a symlink to somewhere else that the hook is
actually stored on disk.

ref: https://bugs.archlinux.org/task/55323

ref: https://bugs.archlinux.org/task/54460

https://sourceware.org/git/?p=glibc.git;a=commit;h=eedca9772e99c72ab4c3c34e43cc764250aa3e3c

Add third argument to add_full_dir: strip_prefix.
The strip_prefix will be stripped off from the destination
path (path in the initramfs image) when adding files.

Rationale: Make it easier to add rootfs overlay hooks when
generating images.

add_full_dir can be invoked in this way:

add_full_dir /path/on/parent/rootfs-overlay * /path/on/parent/rootfs-overlay

The above invocation will add all content of */path/on/parent/rootfs-overlay*
into */* in the initramfs image.

Signed-off-by: Erik Stromdahl <erik.stromdahl@gmail.com>

By passing the filter glob in all recursive invocations of
add_full_dir, the filtering is performed on all subdirs and
not just the top dir.

This makes the implementation correspond to the documented
behavior in the man page.

Signed-off-by: Erik Stromdahl <erik.stromdahl@gmail.com>

allows for backwards compat by detecting and converting old-style
string-based configs.

ref: https://bugs.archlinux.org/task/53343

This is harmless fallout from 5249350c, which tries harder to re-add
aliases of builtins.

Not all root devices are block devices, after all. Rely on mount doing
the right thing (or not).

ref: https://bugs.archlinux.org/task/45118
ref: https://bugs.archlinux.org/task/35529

On compressed images, bash 4.3 gets salty:

$ lsinitcpio initrd.gz >/dev/null
/usr/bin/lsinitcpio: line 78: warning: command substitution: ignored null byte in input
/usr/bin/lsinitcpio: line 90: warning: command substitution: ignored null byte in input

Tidy up all of our hexdump comparisons via command substitution to use
process substitution instead.

ref: https://bugs.archlinux.org/task/52768

crc32 is a builtin module in the Arch kernel but also a crypto module.
Resolve by this only noting that a module is added if the value in the
has is 1, and not 2 (indicating a builtin).

While we're here, change the f2fs softdep of crc32 to crypto-crc32.

ref: https://bugs.archlinux.org/task/49380

this is no longer needed -- we get softdeps from the kernel.

systemd v232 only activates systemd-vconsole-setup via udev. Follow this
to make sure things work properly. This also obsoletes the modules-load
hack.

https://github.com/systemd/systemd/issues/4699

This really ought to be fixed in the module, but here we are again
guessing at module requirements.

ref: https://bugs.archlinux.org/task/49380

builderrors -> _builderrors -- this makes binary dep resolution
properly a builderror. As seen in:

https://bugs.archlinux.org/task/51818

Arch's binaries are all in /usr/bin, so let's not bother searching in
duplicates of this directory.

It's been too long and I've forgotten how to hack on mkinitcpio...

...and mine too, now that he points it out.

These were curiously omitted from the documentation.

ref: https://bugs.archlinux.org/task/50470

If the MODULES contain a GPU driver like i915, wait for it
to be loaded. This avoids the font configuration being lost.

ref: https://bugs.archlinux.org/task/50700

Arch might not really support some of these, but mkinitcpio can.

This prevents having unquoted expansion in the bsdcpio commandline

This refactors parse_cmdline into a few chunks of code, mainly to
separate the work of parsing from the handling of the actual parsed
parameters. By default, parse_cmdline_item is used as the callback for
handling the parameters, but this could be overriden by other code
sourcing init_functions if desirable.

Our test harness passes more tests, but we leave behind some expected
failures to describe the cases where it still fails. Fortunately I've
not yet been able to find any cases which warrant --expect-parse-fail.

Some of these tests currently fail including hard failures where the
shell quits entirely (this would lead to a kernel panic). A followup
commit will rewrite the parse_cmdline function to improve it and fix
these deficiencies.

Version extraction only works on x86 (i686 and x86_64) kernels.

ref: https://bugs.archlinux.org/task/43256

rootfstype is handled by the kernel, so we should handle it, too.

ref: https://bugs.archlinux.org/task/45117

ref: https://bugs.archlinux.org/task/44805

ref: https://bugs.archlinux.org/task/49380

This is true of more and more modules with the advent of aliases, e.g.

  $ modinfo -F filename crc32c
  /lib/modules/4.6.2-rampage/kernel/arch/x86/crypto/crc32c-intel.ko.gz
  /lib/modules/4.6.2-rampage/kernel/crypto/crc32c_generic.ko.gz

modinfo outputs softdeps which we can add if they exist (failing
quietly on not-found). this obviates the need for some of our own
module quirks.

The output of modinfo can take many forms, as it's nearly freeform text
in the  .modinfo section of the kernel module. The basic form is:

  softdep:        pre: foo

But it might also be:

  softdep:        pre: foo bar post: baz

Or:

  softdep:        pre: foo
  softdep:        post: bar

So just parse the entire line, discarding anything that ends with a ':'.