Andrew Geissler | 3b8a17c | 2021-04-15 15:55:55 -0500 | [diff] [blame] | 1 | Moving to the Yocto Project 3.2 Release (gatesgarth) |
| 2 | ==================================================== |
Andrew Geissler | 6ce62a2 | 2020-11-30 19:58:47 -0600 | [diff] [blame] | 3 | |
| 4 | This section provides migration information for moving to the Yocto |
Andrew Geissler | 3b8a17c | 2021-04-15 15:55:55 -0500 | [diff] [blame] | 5 | Project 3.2 Release (codename "gatesgarth") from the prior release. |
Andrew Geissler | 6ce62a2 | 2020-11-30 19:58:47 -0600 | [diff] [blame] | 6 | |
| 7 | .. _migration-3.2-minimum-system-requirements: |
| 8 | |
| 9 | Minimum system requirements |
| 10 | --------------------------- |
| 11 | |
| 12 | ``gcc`` version 6.0 is now required at minimum on the build host. For older |
| 13 | host distributions where this is not available, you can use the |
| 14 | ``buildtools-extended-tarball`` (easily installable using |
| 15 | ``scripts/install-buildtools``). |
| 16 | |
| 17 | |
| 18 | .. _migration-3.2-removed-recipes: |
| 19 | |
| 20 | Removed recipes |
| 21 | --------------- |
| 22 | |
| 23 | The following recipes have been removed: |
| 24 | |
| 25 | - ``bjam-native``: replaced by ``boost-build-native`` |
| 26 | - ``avahi-ui``: folded into the main ``avahi`` recipe - the GTK UI can be disabled using :term:`PACKAGECONFIG` for ``avahi``. |
| 27 | - ``build-compare``: no longer needed with the removal of the ``packagefeed-stability`` class |
| 28 | - ``dhcp``: obsolete, functionally replaced by ``dhcpcd`` and ``kea`` |
| 29 | - ``libmodulemd-v1``: replaced by ``libmodulemd`` |
| 30 | - ``packagegroup-core-device-devel``: obsolete |
| 31 | |
| 32 | |
| 33 | .. _migration-3.2-removed-classes: |
| 34 | |
| 35 | Removed classes |
| 36 | --------------- |
| 37 | |
| 38 | The following classes (.bbclass files) have been removed: |
| 39 | |
| 40 | - ``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. |
| 41 | |
| 42 | - ``packagefeed-stability``: this class had become obsolete with the advent of hash equivalence and reproducible builds. |
| 43 | |
| 44 | |
| 45 | pseudo path filtering and mismatch behaviour |
| 46 | -------------------------------------------- |
| 47 | |
| 48 | pseudo now operates on a filtered subset of files. This is a significant change |
| 49 | to the way pseudo operates within OpenEmbedded - by default, pseudo monitors and |
| 50 | logs (adds to its database) any file created or modified whilst in a ``fakeroot`` |
| 51 | environment. However, there are large numbers of files that we simply don't care |
| 52 | about the permissions of whilst in that ``fakeroot`` context, for example ${:term:`S`}, ${:term:`B`}, ${:term:`T`}, |
| 53 | ${:term:`SSTATE_DIR`}, the central sstate control directories, and others. |
| 54 | |
| 55 | As of this release, new functionality in pseudo is enabled to ignore these |
| 56 | directory trees (controlled using a new :term:`PSEUDO_IGNORE_PATHS` variable) |
| 57 | resulting in a cleaner database with less chance of "stray" mismatches if files |
| 58 | are modified outside pseudo context. It also should reduce some overhead from |
| 59 | pseudo as the interprocess round trip to the server is avoided. |
| 60 | |
| 61 | There is a possible complication where some existing recipe may break, for |
| 62 | example, a recipe was found to be writing to ``${B}/install`` for |
| 63 | ``make install`` in ``do_install`` and since ``${B}`` is listed as not to be tracked, |
| 64 | there were errors trying to ``chown root`` for files in this location. Another |
| 65 | example was the ``tcl`` recipe where the source directory ``S`` is set to a |
| 66 | subdirectory of the source tree but files were written out to the directory |
| 67 | structure above that subdirectory. For these types of cases in your own recipes, |
| 68 | extend ``PSEUDO_IGNORE_PATHS`` to cover additional paths that pseudo should not |
| 69 | be monitoring. |
| 70 | |
| 71 | In addition, pseudo's behaviour on mismatches has now been changed - rather |
| 72 | than doing what turns out to be a rather dangerous "fixup" if it sees a file |
| 73 | with a different path but the same inode as another file it has previously seen, |
Andrew Geissler | 09209ee | 2020-12-13 08:44:15 -0600 | [diff] [blame] | 74 | pseudo will throw an ``abort()`` and direct you to a :yocto_wiki:`wiki page </Pseudo_Abort>` |
Andrew Geissler | 6ce62a2 | 2020-11-30 19:58:47 -0600 | [diff] [blame] | 75 | that explains how to deal with this. |
| 76 | |
| 77 | |
| 78 | .. _migration-3.2-multilib-mlprefix: |
| 79 | |
| 80 | ``MLPREFIX`` now required for multilib when runtime dependencies conditionally added |
| 81 | ------------------------------------------------------------------------------------ |
| 82 | |
| 83 | In order to solve some previously intractable problems with runtime |
| 84 | dependencies and multilib, a change was made that now requires the :term:`MLPREFIX` |
| 85 | value to be explicitly prepended to package names being added as |
| 86 | dependencies (e.g. in :term:`RDEPENDS` and :term:`RRECOMMENDS` values) |
| 87 | where the dependency is conditionally added. |
| 88 | |
| 89 | If you have anonymous python or in-line python conditionally adding |
| 90 | dependencies in your custom recipes, and you intend for those recipes to |
| 91 | work with multilib, then you will need to ensure that ``${MLPREFIX}`` |
| 92 | is prefixed on the package names in the dependencies, for example |
Andrew Geissler | c926e17 | 2021-05-07 16:11:35 -0500 | [diff] [blame] | 93 | (from the ``glibc`` recipe):: |
Andrew Geissler | 6ce62a2 | 2020-11-30 19:58:47 -0600 | [diff] [blame] | 94 | |
| 95 | RRECOMMENDS_${PN} = "${@bb.utils.contains('DISTRO_FEATURES', 'ldconfig', '${MLPREFIX}ldconfig', '', d)}" |
| 96 | |
| 97 | This also applies when conditionally adding packages to :term:`PACKAGES` where |
Andrew Geissler | c926e17 | 2021-05-07 16:11:35 -0500 | [diff] [blame] | 98 | those packages have dependencies, for example (from the ``alsa-plugins`` recipe):: |
Andrew Geissler | 6ce62a2 | 2020-11-30 19:58:47 -0600 | [diff] [blame] | 99 | |
| 100 | PACKAGES += "${@bb.utils.contains('PACKAGECONFIG', 'pulseaudio', 'alsa-plugins-pulseaudio-conf', '', d)}" |
| 101 | ... |
| 102 | RDEPENDS_${PN}-pulseaudio-conf += "\ |
| 103 | ${MLPREFIX}libasound-module-conf-pulse \ |
| 104 | ${MLPREFIX}libasound-module-ctl-pulse \ |
| 105 | ${MLPREFIX}libasound-module-pcm-pulse \ |
| 106 | " |
| 107 | |
| 108 | |
| 109 | .. _migration-3.2-packagegroup-core-device-devel: |
| 110 | |
| 111 | packagegroup-core-device-devel no longer included in images built for qemu* machines |
| 112 | ------------------------------------------------------------------------------------ |
| 113 | |
| 114 | ``packagegroup-core-device-devel`` was previously added automatically to |
| 115 | images built for ``qemu*`` machines, however the purpose of the group and what |
| 116 | it should contain is no longer clear, and in general, adding userspace |
| 117 | development items to images is best done at the image/class level; thus this |
| 118 | packagegroup was removed. |
| 119 | |
| 120 | This packagegroup previously pulled in the following: |
| 121 | |
| 122 | - ``distcc-config`` |
| 123 | - ``nfs-export-root`` |
| 124 | - ``bash`` |
| 125 | - ``binutils-symlinks`` |
| 126 | |
| 127 | If you still need any of these in your image built for a ``qemu*`` machine |
| 128 | then you will add them explicitly to :term:`IMAGE_INSTALL` or another |
| 129 | appropriate place in the dependency chain for your image (if you have not |
| 130 | already done so). |
| 131 | |
| 132 | |
| 133 | .. _migration-3.2-dhcp: |
| 134 | |
| 135 | DHCP server/client replaced |
| 136 | --------------------------- |
| 137 | |
| 138 | The ``dhcp`` software package has become unmaintained and thus has been |
| 139 | functionally replaced by ``dhcpcd`` (client) and ``kea`` (server). You will |
| 140 | need to replace references to the recipe/package names as appropriate - most |
| 141 | commonly, at the package level ``dhcp-client`` should be replaced by |
| 142 | ``dhcpcd`` and ``dhcp-server`` should be replaced by ``kea``. If you have any |
| 143 | custom configuration files for these they will need to be adapted - refer to |
| 144 | the upstream documentation for ``dhcpcd`` and ``kea`` for further details. |
| 145 | |
| 146 | |
| 147 | .. _migration-3.2-packaging-changes: |
| 148 | |
| 149 | Packaging changes |
| 150 | ----------------- |
| 151 | |
| 152 | - ``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. |
| 153 | |
| 154 | - ``iptables``: ``iptables-apply`` and ``ip6tables-apply`` have been split out to their own package to avoid a bash dependency in the main ``iptables`` package |
| 155 | |
| 156 | |
| 157 | .. _migration-3.2-package-qa-checks: |
| 158 | |
| 159 | Package QA check changes |
| 160 | ------------------------ |
| 161 | |
| 162 | Previously, the following package QA checks triggered warnings, however they can |
| 163 | be indicators of genuine underlying problems and are therefore now treated as |
| 164 | errors: |
| 165 | |
| 166 | - :ref:`already-stripped <qa-check-already-stripped>` |
| 167 | - :ref:`compile-host-path <qa-check-compile-host-path>` |
| 168 | - :ref:`installed-vs-shipped <qa-check-installed-vs-shipped>` |
| 169 | - :ref:`ldflags <qa-check-ldflags>` |
| 170 | - :ref:`pn-overrides <qa-check-pn-overrides>` |
| 171 | - :ref:`rpaths <qa-check-rpaths>` |
| 172 | - :ref:`staticdev <qa-check-staticdev>` |
| 173 | - :ref:`unknown-configure-option <qa-check-unknown-configure-option>` |
| 174 | - :ref:`useless-rpaths <qa-check-useless-rpaths>` |
| 175 | |
| 176 | In addition, the following new checks were added and default to triggering an error: |
| 177 | |
| 178 | - :ref:`shebang-size <qa-check-shebang-size>`: Check for shebang (#!) lines longer than 128 characters, which can give an error at runtime depending on the operating system. |
| 179 | |
| 180 | - :ref:`unhandled-features-check <qa-check-unhandled-features-check>`: Check if any of the variables supported by the :ref:`features_check <ref-classes-features_check>` class is set while not inheriting the class itself. |
| 181 | |
| 182 | - :ref:`missing-update-alternatives <qa-check-missing-update-alternatives>`: Check if the recipe sets the :term:`ALTERNATIVE` variable for any of its packages, and does not inherit the :ref:`update-alternatives <ref-classes-update-alternatives>` class. |
| 183 | |
| 184 | - A trailing slash or duplicated slashes in the value of :term:`S` or :term:`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. |
| 185 | |
| 186 | |
| 187 | .. _migration-3.2-src-uri-file-globbing: |
| 188 | |
| 189 | Globbing no longer supported in ``file://`` entries in ``SRC_URI`` |
| 190 | ------------------------------------------------------------------ |
| 191 | |
| 192 | Globbing (``*`` and ``?`` wildcards) in ``file://`` URLs within :term:`SRC_URI` |
| 193 | did not properly support file checksums, thus changes to the source files |
| 194 | would not always change the do_fetch task checksum, and consequently would |
| 195 | not ensure that the changed files would be incorporated in subsequent builds. |
| 196 | |
| 197 | Unfortunately it is not practical to make globbing work generically here, so |
| 198 | the decision was taken to remove support for globs in ``file://`` URLs. |
| 199 | If you have any usage of these in your recipes, then you will now need to |
| 200 | either add each of the files that you expect to match explicitly, or |
| 201 | alternatively if you still need files to be pulled in dynamically, put the |
| 202 | files into a subdirectory and reference that instead. |
| 203 | |
| 204 | |
| 205 | .. _migration-3.2-deploydir-clean: |
| 206 | |
| 207 | deploy class now cleans ``DEPLOYDIR`` before ``do_deploy`` |
| 208 | ---------------------------------------------------------- |
| 209 | |
| 210 | ``do_deploy`` as implemented in the :ref:`deploy <ref-classes-deploy>` class now cleans up ${:term:`DEPLOYDIR`} before running, just as ``do_install`` cleans up ${:term:`D`} before running. This reduces the risk of ``DEPLOYDIR`` being accidentally contaminated by files from previous runs, possibly even with different config, in case of incremental builds. |
| 211 | |
| 212 | Most recipes and classes that inherit the ``deploy`` class or interact with ``do_deploy`` are unlikely to be affected by this unless they add ``prefuncs`` to ``do_deploy`` *which also* put files into ``${DEPLOYDIR}`` - these should be refactored to use ``do_deploy_prepend`` instead. |
| 213 | |
| 214 | |
| 215 | .. _migration-3.2-nativesdk-sdk-provides-dummy: |
| 216 | |
| 217 | Custom SDK / SDK-style recipes need to include ``nativesdk-sdk-provides-dummy`` |
| 218 | ------------------------------------------------------------------------------- |
| 219 | |
| 220 | 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 :term:`TOOLCHAIN_HOST_TASK`. |
| 221 | |
| 222 | |
| 223 | ``ld.so.conf`` now moved back to main ``glibc`` package |
| 224 | ------------------------------------------------------- |
| 225 | |
| 226 | There are cases where one doesn't want ``ldconfig`` on target (e.g. for |
| 227 | read-only root filesystems, it's rather pointless), yet one still |
| 228 | needs ``/etc/ld.so.conf`` to be present at image build time: |
| 229 | |
| 230 | When some recipe installs libraries to a non-standard location, and |
| 231 | therefore installs in a file in ``/etc/ld.so.conf.d/foo.conf``, we |
Andrew Geissler | c926e17 | 2021-05-07 16:11:35 -0500 | [diff] [blame] | 232 | need ``/etc/ld.so.conf`` containing:: |
Andrew Geissler | 6ce62a2 | 2020-11-30 19:58:47 -0600 | [diff] [blame] | 233 | |
| 234 | include /etc/ld.so.conf.d/*.conf |
| 235 | |
| 236 | in order to get those other locations picked up. |
| 237 | |
| 238 | Thus ``/etc/ld.so.conf`` is now in the main ``glibc`` package so that |
| 239 | there's always an ``ld.so.conf`` present when the build-time ``ldconfig`` |
| 240 | runs towards the end of image construction. |
| 241 | |
| 242 | The ``ld.so.conf`` and ``ld.so.conf.d/*.conf`` files do not take up |
| 243 | significant space (at least not compared to the ~700kB ``ldconfig`` binary), and they |
| 244 | might be needed in case ``ldconfig`` is installable, so they are left |
| 245 | in place after the image is built. Technically it would be possible to |
| 246 | remove them if desired, though it would not be trivial if you still |
| 247 | wanted the build-time ldconfig to function (:term:`ROOTFS_POSTPROCESS_COMMAND` |
| 248 | will not work as ``ldconfig`` is run after the functions referred to |
| 249 | by that variable). |
| 250 | |
| 251 | |
| 252 | .. _migration-3.2-virgl: |
| 253 | |
| 254 | Host DRI drivers now used for GL support within ``runqemu`` |
| 255 | ----------------------------------------------------------- |
| 256 | |
| 257 | ``runqemu`` now uses the mesa-native libraries everywhere virgl is used |
| 258 | (i.e. when ``gl``, ``gl-es`` or ``egl-headless`` options are specified), |
| 259 | but instructs them to load DRI drivers from the host. Unfortunately this |
| 260 | may not work well with proprietary graphics drivers such as those from |
| 261 | Nvidia; if you are using such drivers then you may need to switch to an |
| 262 | alternative (such as Nouveau in the case of Nvidia hardware) or avoid |
| 263 | using the GL options. |
| 264 | |
| 265 | |
| 266 | .. _migration-3.2-initramfs-suffix: |
| 267 | |
| 268 | initramfs images now use a blank suffix |
| 269 | --------------------------------------- |
| 270 | |
| 271 | The reference initramfs images (``core-image-minimal-initramfs``, |
| 272 | ``core-image-tiny-initramfs`` and ``core-image-testmaster-initramfs``) now |
| 273 | set an empty string for :term:`IMAGE_NAME_SUFFIX`, which otherwise defaults |
| 274 | to ``".rootfs"``. These images aren't root filesystems and thus the rootfs |
| 275 | label didn't make sense. If you are looking for the output files generated |
| 276 | by these image recipes directly then you will need to adapt to the new |
| 277 | naming without the ``.rootfs`` part. |
| 278 | |
| 279 | |
| 280 | .. _migration-3.2-image-artifact-names: |
| 281 | |
| 282 | Image artifact name variables now centralised in image-artifact-names class |
| 283 | --------------------------------------------------------------------------- |
| 284 | |
| 285 | The defaults for the following image artifact name variables have been moved |
| 286 | from bitbake.conf to a new ``image-artifact-names`` class: |
| 287 | |
| 288 | - :term:`IMAGE_BASENAME` |
| 289 | - :term:`IMAGE_LINK_NAME` |
| 290 | - :term:`IMAGE_NAME` |
| 291 | - :term:`IMAGE_NAME_SUFFIX` |
| 292 | - :term:`IMAGE_VERSION_SUFFIX` |
| 293 | |
| 294 | Image-related classes now inherit this class, and typically these variables |
| 295 | are only referenced within image recipes so those will be unaffected by this |
| 296 | change. However if you have references to these variables in either a recipe |
| 297 | that is not an image or a class that is enabled globally, then those will |
| 298 | now need to be changed to ``inherit image-artifact-names``. |
| 299 | |
| 300 | |
| 301 | .. _migration-3.2-misc: |
| 302 | |
| 303 | Miscellaneous changes |
| 304 | --------------------- |
| 305 | |
| 306 | - Support for the long-deprecated ``PACKAGE_GROUP`` variable has now been removed - replace any remaining instances with :term:`FEATURE_PACKAGES`. |
| 307 | - The ``FILESPATHPKG`` variable, having been previously deprecated, has now been removed. Replace any remaining references with appropriate use of :term:`FILESEXTRAPATHS`. |
| 308 | - Erroneous use of ``inherit +=`` (instead of ``INHERIT +=``) in a configuration file now triggers an error instead of silently being ignored. |
| 309 | - 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. |
| 310 | - ``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. |
Andrew Geissler | 3b8a17c | 2021-04-15 15:55:55 -0500 | [diff] [blame] | 311 | - 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. |
Andrew Geissler | 6ce62a2 | 2020-11-30 19:58:47 -0600 | [diff] [blame] | 312 | - 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. |
| 313 | - The ``tune-supersparc.inc`` tune file has been removed as it does not appear to be widely used and no longer works. |