Release 3.2 (gatesgarth)

This section provides migration information for moving to the Yocto Project 3.2 Release (codename “gatesgarth”) from the prior release.

Minimum system requirements

gcc version 6.0 is now required at minimum on the build host. For older host distributions where this is not available, you can use the buildtools-extended tarball (easily installable using scripts/install-buildtools).

Removed recipes

The following recipes have been removed:

  • bjam-native: replaced by boost-build-native

  • avahi-ui: folded into the main avahi recipe — the GTK UI can be disabled using PACKAGECONFIG for avahi.

  • build-compare: no longer needed with the removal of the packagefeed-stability class

  • dhcp: obsolete, functionally replaced by dhcpcd and kea

  • libmodulemd-v1: replaced by libmodulemd

  • packagegroup-core-device-devel: obsolete

Removed classes

The following classes (.bbclass files) have been removed:

  • spdx: obsolete — the Yocto Project is a strong supporter of SPDX, but this class was old code using a dated approach and had the potential to be misleading. The meta-sdpxscanner layer is a much more modern and active approach to handling this and is recommended as a replacement.

  • packagefeed-stability: this class had become obsolete with the advent of hash equivalence and reproducible builds.

pseudo path filtering and mismatch behaviour

pseudo now operates on a filtered subset of files. This is a significant change to the way pseudo operates within OpenEmbedded — by default, pseudo monitors and logs (adds to its database) any file created or modified whilst in a fakeroot environment. However, there are large numbers of files that we simply don’t care about the permissions of whilst in that fakeroot context, for example ${S}, ${B}, ${T}, ${SSTATE_DIR}, the central sstate control directories, and others.

As of this release, new functionality in pseudo is enabled to ignore these directory trees (controlled using a new PSEUDO_IGNORE_PATHS variable) resulting in a cleaner database with less chance of “stray” mismatches if files are modified outside pseudo context. It also should reduce some overhead from pseudo as the interprocess round trip to the server is avoided.

There is a possible complication where some existing recipe may break, for example, a recipe was found to be writing to ${B}/install for make install in do_install and since ${B} is listed as not to be tracked, there were errors trying to chown root for files in this location. Another example was the tcl recipe where the source directory S is set to a subdirectory of the source tree but files were written out to the directory structure above that subdirectory. For these types of cases in your own recipes, extend PSEUDO_IGNORE_PATHS to cover additional paths that pseudo should not be monitoring.

In addition, pseudo’s behaviour on mismatches has now been changed — rather than doing what turns out to be a rather dangerous “fixup” if it sees a file with a different path but the same inode as another file it has previously seen, pseudo will throw an abort() and direct you to a wiki page that explains how to deal with this.

MLPREFIX now required for multilib when runtime dependencies conditionally added

In order to solve some previously intractable problems with runtime dependencies and multilib, a change was made that now requires the MLPREFIX value to be explicitly prepended to package names being added as dependencies (e.g. in RDEPENDS and RRECOMMENDS values) where the dependency is conditionally added.

If you have anonymous Python or in-line Python conditionally adding dependencies in your custom recipes, and you intend for those recipes to work with multilib, then you will need to ensure that ${MLPREFIX} is prefixed on the package names in the dependencies, for example (from the glibc recipe):

RRECOMMENDS_${PN} = "${@bb.utils.contains('DISTRO_FEATURES', 'ldconfig', '${MLPREFIX}ldconfig', '', d)}"

This also applies when conditionally adding packages to PACKAGES where those packages have dependencies, for example (from the alsa-plugins recipe):

PACKAGES += "${@bb.utils.contains('PACKAGECONFIG', 'pulseaudio', 'alsa-plugins-pulseaudio-conf', '', d)}"
...
RDEPENDS_${PN}-pulseaudio-conf += "\
        ${MLPREFIX}libasound-module-conf-pulse \
        ${MLPREFIX}libasound-module-ctl-pulse \
        ${MLPREFIX}libasound-module-pcm-pulse \
"

packagegroup-core-device-devel no longer included in images built for qemu* machines

packagegroup-core-device-devel was previously added automatically to images built for qemu* machines, however the purpose of the group and what it should contain is no longer clear, and in general, adding userspace development items to images is best done at the image/class level; thus this packagegroup was removed.

This packagegroup previously pulled in the following:

  • distcc-config

  • nfs-export-root

  • bash

  • binutils-symlinks

If you still need any of these in your image built for a qemu* machine then you will add them explicitly to IMAGE_INSTALL or another appropriate place in the dependency chain for your image (if you have not already done so).

DHCP server/client replaced

The dhcp software package has become unmaintained and thus has been functionally replaced by dhcpcd (client) and kea (server). You will need to replace references to the recipe/package names as appropriate — most commonly, at the package level dhcp-client should be replaced by dhcpcd and dhcp-server should be replaced by kea. If you have any custom configuration files for these they will need to be adapted — refer to the upstream documentation for dhcpcd and kea for further details.

Packaging changes

  • python3: the urllib Python package has now moved into the core package, as it is used more commonly than just netclient (e.g. email, xml, mimetypes, pydoc). In addition, the pathlib module is now also part of the core package.

  • iptables: iptables-apply and ip6tables-apply have been split out to their own package to avoid a bash dependency in the main iptables package

Package QA check changes

Previously, the following package QA checks triggered warnings, however they can be indicators of genuine underlying problems and are therefore now treated as errors:

In addition, the following new checks were added and default to triggering an error:

  • shebang-size: Check for shebang (#!) lines longer than 128 characters, which can give an error at runtime depending on the operating system.

  • unhandled-features-check: Check if any of the variables supported by the features_check class is set while not inheriting the class itself.

  • missing-update-alternatives: Check if the recipe sets the ALTERNATIVE variable for any of its packages, and does not inherit the update-alternatives class.

  • A trailing slash or duplicated slashes in the value of S or B will now trigger a warning so that they can be removed and path comparisons can be more reliable — remove any instances of these in your recipes if the warning is displayed.

Globbing no longer supported in file:// entries in SRC_URI

Globbing (* and ? wildcards) in file:// URLs within SRC_URI did not properly support file checksums, thus changes to the source files would not always change the do_fetch task checksum, and consequently would not ensure that the changed files would be incorporated in subsequent builds.

Unfortunately it is not practical to make globbing work generically here, so the decision was taken to remove support for globs in file:// URLs. If you have any usage of these in your recipes, then you will now need to either add each of the files that you expect to match explicitly, or alternatively if you still need files to be pulled in dynamically, put the files into a subdirectory and reference that instead.

deploy class now cleans DEPLOYDIR before do_deploy

do_deploy as implemented in the deploy class now cleans up ${DEPLOYDIR} before running, just as do_install cleans up ${D} before running. This reduces the risk of DEPLOYDIR being accidentally contaminated by files from previous runs, possibly even with different config, in case of incremental builds.

Most recipes and classes that inherit the deploy class or interact with do_deploy are unlikely to be affected by this unless they add prefuncs to do_deploy which also put files into ${DEPLOYDIR} — these should be refactored to use do_deploy_prepend instead.

Custom SDK / SDK-style recipes need to include nativesdk-sdk-provides-dummy

All nativesdk packages require /bin/sh due to their postinstall scriptlets, thus this package has to be dummy-provided within the SDK and nativesdk-sdk-provides-dummy now does this. If you have a custom SDK recipe (or your own SDK-style recipe similar to e.g. buildtools-tarball), you will need to ensure nativesdk-sdk-provides-dummy or an equivalent is included in TOOLCHAIN_HOST_TASK.

ld.so.conf now moved back to main glibc package

There are cases where one doesn’t want ldconfig on target (e.g. for read-only root filesystems, it’s rather pointless), yet one still needs /etc/ld.so.conf to be present at image build time:

When some recipe installs libraries to a non-standard location, and therefore installs in a file in /etc/ld.so.conf.d/foo.conf, we need /etc/ld.so.conf containing:

include /etc/ld.so.conf.d/*.conf

in order to get those other locations picked up.

Thus /etc/ld.so.conf is now in the main glibc package so that there’s always an ld.so.conf present when the build-time ldconfig runs towards the end of image construction.

The ld.so.conf and ld.so.conf.d/*.conf files do not take up significant space (at least not compared to the ~700kB ldconfig binary), and they might be needed in case ldconfig is installable, so they are left in place after the image is built. Technically it would be possible to remove them if desired, though it would not be trivial if you still wanted the build-time ldconfig to function (ROOTFS_POSTPROCESS_COMMAND will not work as ldconfig is run after the functions referred to by that variable).

Host DRI drivers now used for GL support within runqemu

runqemu now uses the mesa-native libraries everywhere virgl is used (i.e. when gl, gl-es or egl-headless options are specified), but instructs them to load DRI drivers from the host. Unfortunately this may not work well with proprietary graphics drivers such as those from Nvidia; if you are using such drivers then you may need to switch to an alternative (such as Nouveau in the case of Nvidia hardware) or avoid using the GL options.

Initramfs images now use a blank suffix

The reference Initramfs images (core-image-minimal-initramfs, core-image-tiny-initramfs and core-image-testmaster-initramfs) now set an empty string for IMAGE_NAME_SUFFIX, which otherwise defaults to ".rootfs". These images aren’t root filesystems and thus the rootfs label didn’t make sense. If you are looking for the output files generated by these image recipes directly then you will need to adapt to the new naming without the .rootfs part.

Image artifact name variables now centralised in image-artifact-names class

The defaults for the following image artifact name variables have been moved from bitbake.conf to a new image-artifact-names class:

Image-related classes now inherit this class, and typically these variables are only referenced within image recipes so those will be unaffected by this change. However if you have references to these variables in either a recipe that is not an image or a class that is enabled globally, then those will now need to be changed to inherit image-artifact-names.

Miscellaneous changes

  • Support for the long-deprecated PACKAGE_GROUP variable has now been removed — replace any remaining instances with FEATURE_PACKAGES.

  • The FILESPATHPKG variable, having been previously deprecated, has now been removed. Replace any remaining references with appropriate use of FILESEXTRAPATHS.

  • Erroneous use of inherit += (instead of INHERIT +=) in a configuration file now triggers an error instead of silently being ignored.

  • ptest support has been removed from the kbd recipe, as upstream has moved to autotest which is difficult to work with in a cross-compilation environment.

  • oe.utils.is_machine_specific() and oe.utils.machine_paths() have been removed as their utility was questionable. In the unlikely event that you have references to these in your own code, then the code will need to be reworked.

  • The i2ctransfer module is now disabled by default when building busybox in order to be consistent with disabling the other i2c tools there. If you do wish the i2ctransfer module to be built in BusyBox then add CONFIG_I2CTRANSFER=y to your custom BusyBox configuration.

  • In the Upstream-Status header convention for patches, Accepted has been replaced with Backport as these almost always mean the same thing i.e. the patch is already upstream and may need to be removed in a future recipe upgrade. If you are adding these headers to your own patches then use Backport to indicate that the patch has been sent upstream.

  • The tune-supersparc.inc tune file has been removed as it does not appear to be widely used and no longer works.