Cache the compiler output of rust code with sccache, like we use ccache
for c code.
I've considered using sccache to completely replace ccache since it can
cache output of C/C++ code too. But let's not do it for now since ccache
doesn't need to run a daemon in the background that needs to be stopped
when shutting down / zapping. Also it would need more refactoring.
Reviewed-by: Luca Weiss <luca@z3ntu.xyz>
Link: https://lists.sr.ht/~postmarketos/pmbootstrap-devel/%3C20230806184729.4891-5-ollieparanoid@postmarketos.org%3E
With this code path, pmbootstrap would start a distccd + sshd in the
native chroot, and configure it so it runs the cross compiler. The
foreign arch chroots would then call this cross compiler from localhost
by calling the distcc client instead of gcc.
This code has been obsoleted by the much simpler crossdirect in 2019.
Let's finally remove it.
Fixes: issue 2179
Reviewed-by: Luca Weiss <luca@z3ntu.xyz>
Reviewed-by: Clayton Craft <clayton@craftyguy.net>
Link: https://lists.sr.ht/~postmarketos/pmbootstrap-devel/%3C20230613161437.570196-4-ollieparanoid@postmarketos.org%3E
Set the env var, so python programs running inside pmbootstrap chroot
don't buffer their output and only end up printing everything when they
are done. I've seen this with meson. This is bad for usability because
we don't see output, but also a problem because pmbootstrap kills
commands if they don't print any output for some time (default: 15 min).
Check if the pkgnames are sane in install_run_apk, right before running
apk. This makes sure that we really run it on all arguments that are
supposed to be packages / files and not options to apk.
Previously pmbootstrap would only show the packages that are about to be
installed. In case all packages were already installed, this would lead
to weird empty install messages:
(rootfs_asus-me176c) install
Show all packages that we want to install, even if they are already
installed in the given chroot.
(rootfs_asus-me176c) install device-asus-me176c
Previously to this patch, pmbootstrap would pass a full dependency tree
to "apk add". It would use a virtual package to ensure only the right
packages get added to /etc/apk/world. For example:
apk add -u --virtual .pmbootstrap postmarketos-base device-asus-me176c \
postmarketos-ui-sxmo-de-sway device-asus-me176c-nonfree-firmware w3m \
sfeed clickclack firefox-esr font-noto font-noto-emoji gnome-icon-theme \
imv megapixels mobile-config-firefox ttyescape postmarketos-base-nofde \
eudev openssh postmarketos-mkinitfs postmarketos-mvcfg postmarketos-keys \
...
Instead of doing that, only pass the packages we want to install and let
apk figure out the dependencies. Most of the time we can even avoid
using the virtual package now.
== Remaining edge case: locally built packages
apk will only upgrade a package with the same pkgver + pkgrel but a
different build date if the full path to an apk file gets passed as
argument. So if the user built a package locally that will be installed,
or one of its dependencies then we still need to use a virtual package
and possibly pass a dependency to apk. Replace
replace_aports_packages_with_path() with packages_get_locally_built_apks()
to get a list of such packages and adjust install() and
install_run_apk() to use it.
Make the code easier to read by moving split_to_add_del() to a separate
function and do some related refactoring. A future patch will use it
twice in install().
Move "arch = ..." to the top of the function while at it, since it's
needed later in the function in 2 places and is not needed for figuring
out packages_with_depends, to_add and to_del.
Remove "# Add depends to packages" because it's obvious from the
packages_with_depends variable name, and getting to_add/to_del is a
different action that stood under the same comment.
Split out the part that builds the apk commands and runs them out from
install() to a new function install_run_apk(). This makes install()
easier to read.
Since the previous commit that adds install_build(), all packages either
end up in to_add or to_del. Move the check for empty packages to the top
of the function, and directly check the packages variable.
I think it's worth keeping this check because it's shorter to add this
check once here than having it a few times in other place where we may
have or may not have something to install. And so we can avoid printing
an empty "install" message with no packages.
install_is_necessary used to do the following things:
1. Error out if there's no binary package but pmb was invoked as
"pmbootstrap install" and build_pkgs_on_install is disabled.
2. Build the package if necessary.
3. Return if a package "needs to be installed" (Boolean or Float).
The only caller of the function is pmb.chroot.apk.install. It would not
add the package to the long "apk add" command if according to 3. it does
not need to be installed.
When I implemented this a few years ago, I probably thought it would be
useful to not unnecessarily pass packages to apk. But this actually
makes it more complicated and doesn't have a benefit, apk is perfectly
capable of recognizing which packages it had already installed.
Replace the function with a much simpler pmb.chroot.apk.install_build,
which only does 1. and 2. Change the order of the package, arch
arguments to match called functions pmb.parse.apkindex.package and
pmb.build.package.
Put underscores between the words for consistency with other variable
names. Since the whole function is about dealing with packages, remove
the "packages_" prefix to make them shorter.
packages_toadd -> to_add
packages_todel -> to_del
Specifically, this fixes issues with process substitution in bash in the
chroot.
For example, the following lines from a NetworkManager build script:
... |
grep -Fx -f <(get_symbols_explicit) -v |
grep -Fx -f <(get_symbols_nm)
fail with:
ninja: job failed: /home/pmos/build/src/NetworkManager-1.35.92/tools/create-exports-NetworkManager.sh --called-from-build /home/pmos/build/src/NetworkManager-1.35.92
grep: /dev/fd/63: No such file or directory
grep: /dev/fd/63: No such file or directory
pmbootstrap netboot command exposes the generated vendor-codename.img
rootfs through nbd interface so that device can mount it and boot
postmarketOS without having any storage medium at all.
Co-authored-by: Luca Weiss <luca@z3ntu.xyz>
When installing dependencies for a package, conflicting (!) dependencies
are now deleted (with `apk del pkg`) whereas before a constraint for
their _absence_ was added (with `apk add !pkg`). Doing it the new way
around prevents creating deadlocks because a `!pkg` constraint will
prevent pkg from ever being installed without an explicit `apk del`
call.
Fixes: #2092
Replace "args.cache" with a global variable in order to
avoid passing "args" to all functions. This is a step to get rid of this
args-passed-to-all-functions pattern in pmbootstrap.
Replace "args.arch_native" with the direct function call in order to
avoid passing "args" to all functions. This is a step to get rid of this
args-passed-to-all-functions pattern in pmbootstrap.
Follow-up to 09794ef832 -
initfs file does not have flavor now, too.
Without this 'pmbootstrap initfs extract' fails with:
(rootfs_samsung-klte) % cp /boot/initramfs-postmarketos-qcom-msm8974 /tmp/initfs-extracted/_initfs.gz
cp: cannot stat '/boot/initramfs-postmarketos-qcom-msm8974': No such file or directory
Fix the issue of having the postmarketOS binary repository key deleted
from the chroots after "abuild undeps" removes the postmarketos-keys
package. This happens if e.g. building two device packages in a row (as
they depend on postmarketos-base, which depends on postmarketos-keys in
current pmaports.git master).
I've also considered installing the postmarketos-keys next to
alpine-base in new chroots. But this would introduce a bootstrap
problem, since you can't install the postmarketos-keys package unless
it already exists in the repository. We'd run into that when building
the next release.
Move code to install Alpine and postmarketOS keys into its own function,
so it can be called outside of pmb.chroot.init too (next patch).
Describe why this is needed in the first place and, while at it, tweak
the function to only copy the key if it does not exist in the target
directory.
kernel is named /boot/vmlinuz now, looking at the filename will no
longer tell us what flavor it is. This now will look at
/usr/share/kernel, which has always contained the kernel 'flavor', and
since we currently only install 1 kernel these days, guarding this with
pmaports.cfg should be unnecessary. In the worst case (if there are
multiple kernel 'flavors' installed), it'll just grab the first one and
return it.
The new mkinitfs does not have options for kernel flavor or version
A new pmaports.cfg variable, "supported_mkinitfs_without_flavors" is used to
determine how to invoke mkinitfs
gitlab runners seem to be registering binfmt now (?), so this fixes an
issue where register() might return immediately because the OS (or ??)
took care of registration, but the qemu-<arch> binary wasn't added to
the chroot.
It seems like the gitlab runners will now automatically register archs
when binfmt is mounted (I'm not completely sure, it's really
hard/annoying to confirm using gitlab CI jobs and no direct access to
the runner)
In any case, checking if it's already registered fixes a problem where
CI fails with "File exists" when it tries to register the arch.
Remove /in-pmbootstrap inside chroots in "pmbootstrap shutdown" instead
of having it at a specific part of "pmbootstrap install".
Reasoning:
* With current approach, it didn't get removed in the on-device
installer chroot.
* This is less error prone than calling it multiple times in
"pmbootstrap install"
Touch the file /in-pmbootstrap in chroots so that we can avoid
performing automated actions that should only happen on a real device
(like flashing the kernel).
Support branches, so pmbootstrap won't fail if v20.05 is selected:
ERROR: You have an outdated version of the 'apk' package manager installed
(your version: 2.10.5-r1, expected at least: 2.12.1-r0).
Move the logic for this check to pmb.helpers.apk.check_outdated and
adjust the test.
This fixes the CI failure in test_crossdirect_rust, which uses the
stable channel. (My bad for not creating this patch earlier, while at
the same time explaining in the creating pmbootstrap release instructions,
that this minimum apk version should be adjusted.)
Remove "_static" from the variable name, as this version isn't just
used to compare apk-tools-static's version (used to set up chroot), but
also for regular apk-tools before entering chroots.
Replace the "kill_as_root" argument with a much simpler "sudo" argument
and remove the now obsolete check for the output mode of "kill_as_root".
"kill_as_root" would only get set to True if both conditions are met:
a) command is running with sudo
b) command is running with an output mode ("log" or "stdout") where
pmb.helpers.run_core would kill it if it does not output anything
before a timeout is reached
The new "sudo" argument just indicates if the command is running with
sudo (a), regardless of the output mode (b).
Oliver Smith
Clayton Craft
Hugo Osvaldo Barrera
Song Fuchang
Mark Hargreaves
Johannes Marbach
Anjandev Momi
BO41
Alexey Min
Newbyte
Maxim Karasev
Caleb Connolly
Anri Dellal