1Migration notes for 3.4 (honister)
2----------------------------------
3
4This section provides migration information for moving to the Yocto
5Project 3.4 Release (codename "honister") from the prior release.
6
7Override syntax changes
8~~~~~~~~~~~~~~~~~~~~~~~
9
10In this release, the ``:`` character replaces the use of ``_`` to
11refer to an override, most commonly when making a conditional assignment
12of a variable. This means that an entry like::
13
14   SRC_URI_qemux86 = "file://somefile"
15
16now becomes::
17
18   SRC_URI:qemux86 = "file://somefile"
19
20since ``qemux86`` is an override. This applies to any use of override
21syntax, so the following::
22
23   SRC_URI_append = " file://somefile"
24   SRC_URI_append_qemux86 = " file://somefile2"
25   SRC_URI_remove_qemux86-64 = " file://somefile3"
26   SRC_URI_prepend_qemuarm = "file://somefile4 "
27   FILES_${PN}-ptest = "${bindir}/xyz"
28   IMAGE_CMD_tar = "tar"
29   BASE_LIB_tune-cortexa76 = "lib"
30   SRCREV_pn-bash = "abc"
31   BB_TASK_NICE_LEVEL_task-testimage = '0'
32
33would now become::
34
35   SRC_URI:append = " file://somefile"
36   SRC_URI:append:qemux86 = " file://somefile2"
37   SRC_URI:remove:qemux86-64 = " file://somefile3"
38   SRC_URI:prepend:qemuarm = "file://somefile4 "
39   FILES:${PN}-ptest = "${bindir}/xyz"
40   IMAGE_CMD:tar = "tar"
41   BASE_LIB:tune-cortexa76 = "lib"
42   SRCREV:pn-bash = "abc"
43   BB_TASK_NICE_LEVEL:task-testimage = '0'
44
45This also applies to
46:ref:`variable queries to the datastore <bitbake:bitbake-user-manual/bitbake-user-manual-metadata:functions for accessing datastore variables>`,
47for example using ``getVar`` and similar so ``d.getVar("RDEPENDS_${PN}")``
48becomes ``d.getVar("RDEPENDS:${PN}")``.
49
50Whilst some of these are fairly obvious such as :term:`MACHINE` and :term:`DISTRO`
51overrides, some are less obvious, for example the packaging variables such as
52:term:`RDEPENDS`, :term:`FILES` and so on taking package names (e.g. ``${PN}``,
53``${PN}-ptest``) as overrides. These overrides are not always in
54:term:`OVERRIDES` but applied conditionally in specific contexts
55such as packaging. ``task-<taskname>`` is another context specific override, the
56context being specific tasks in that case. Tune overrides are another special
57case where some code does use them as overrides but some does not. We plan to try
58and make the tune code use overrides more consistently in the future.
59
60There are some variables which do not use override syntax which include the
61suffix to variables in ``layer.conf`` files such as :term:`BBFILE_PATTERN`,
62:term:`SRCREV`\ ``_xxx`` where ``xxx`` is a name from :term:`SRC_URI` and
63:term:`PREFERRED_VERSION`\ ``_xxx``. In particular, ``layer.conf`` suffixes
64may be the same as a :term:`DISTRO` override causing some confusion. We do
65plan to try and improve consistency as these issues are identified.
66
67To help with migration of layers, a script has been provided in OE-Core.
68Once configured with the overrides used by a layer, this can be run as::
69
70   <oe-core>/scripts/contrib/convert-overrides.py <layerdir>
71
72.. note::
73
74   Please read the notes in the script as it isn't entirely automatic and it isn't
75   expected to handle every case. In particular, it needs to be told which overrides
76   the layer uses (usually machine and distro names/overrides) and the result should
77   be carefully checked since it can be a little enthusiastic and will convert
78   references to ``_append``, ``_remove`` and ``_prepend`` in function and variable
79   names.
80
81For reference, this conversion is important as it allows BitBake to more reliably
82determine what is an override and what is not, as underscores are also used in
83variable names without intending to be overrides. This should allow us to proceed
84with other syntax improvements and simplifications for usability. It also means
85BitBake no longer has to guess and maintain large lookup lists just in case
86e.g. ``functionname`` in ``my_functionname`` is an override, and thus should improve
87efficiency.
88
89New host dependencies
90~~~~~~~~~~~~~~~~~~~~~
91
92The ``lz4c``, ``pzstd`` and ``zstd`` commands are now required to be
93installed on the build host to support LZ4 and Zstandard compression
94functionality. These are typically provided by ``lz4`` and ``zstd``
95packages in most Linux distributions. Alternatively they are available
96as part of ``buildtools-tarball`` if your distribution does not provide
97them. For more information see
98:ref:`ref-manual/system-requirements:required packages for the build host`.
99
100Removed recipes
101~~~~~~~~~~~~~~~
102
103The following recipes have been removed in this release:
104
105- ``assimp``: problematic from a licensing perspective and no longer
106  needed by anything else
107- ``clutter-1.0``: legacy component moved to meta-gnome
108- ``clutter-gst-3.0``: legacy component moved to meta-gnome
109- ``clutter-gtk-1.0``: legacy component moved to meta-gnome
110- ``cogl-1.0``: legacy component moved to meta-gnome
111- ``core-image-clutter``: removed along with clutter
112- ``linux-yocto``: removed version 5.4 recipes (5.14 and 5.10 still
113  provided)
114- ``mklibs-native``: not actively tested and upstream mklibs still
115  requires Python 2
116- ``mx-1.0``: obsolete (last release 2012) and isn't used by anything in
117  any known layer
118- ``packagegroup-core-clutter``: removed along with clutter
119
120Removed classes
121~~~~~~~~~~~~~~~
122
123- ``clutter``: moved to meta-gnome along with clutter itself
124- ``image-mklibs``: not actively tested and upstream mklibs still
125  requires Python 2
126- ``meta``: no longer useful. Recipes that need to skip installing
127  packages should inherit ``nopackages`` instead.
128
129Prelinking disabled by default
130~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
131
132Recent tests have shown that prelinking works only when PIE is not
133enabled (see `here <https://rlbl.me/prelink-1>`__ and `here <https://rlbl.me/prelink-2>`__),
134and as PIE is both a desirable security feature, and the only
135configuration provided and tested by the Yocto Project, there is
136simply no sense in continuing to enable prelink.
137
138There's also a concern that no one is maintaining the code, and there
139are open bugs (including :yocto_bugs:`this serious one </show_bug.cgi?id=14429>`).
140Given that prelink does intricate address arithmetic and rewriting
141of binaries the best option is to disable the feature. It is recommended
142that you consider disabling this feature in your own configuration if
143it is currently enabled.
144
145Virtual runtime provides
146~~~~~~~~~~~~~~~~~~~~~~~~
147
148Recipes shouldn't use the ``virtual/`` string in :term:`RPROVIDES` and
149:term:`RDEPENDS` --- it is confusing because ``virtual/`` has no special
150meaning in :term:`RPROVIDES` and :term:`RDEPENDS` (unlike in the
151corresponding build-time :term:`PROVIDES` and :term:`DEPENDS`).
152
153Tune files moved to architecture-specific directories
154~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
155
156The tune files found in ``conf/machine/include`` have now been moved
157into their respective architecture name directories under that same
158location; e.g. x86 tune files have moved into an ``x86`` subdirectory,
159MIPS tune files have moved into a ``mips`` subdirectory, etc.
160The ARM tunes have an extra level (``armv8a``, ``armv8m``, etc.) and
161some have been renamed to make them uniform with the rest of the tunes.
162See :yocto_git:`this commit </poky/commit/?id=1d381f21f5f13aa0c4e1a45683ed656ebeedd37d>`
163for reference.
164
165If you have any references to tune files (e.g. in custom machine
166configuration files) they will need to be updated.
167
168Extensible SDK host extension
169~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
170
171For a normal SDK, some layers append to :term:`TOOLCHAIN_HOST_TASK`
172unconditionally which is fine, until the eSDK tries to override the
173variable to its own values. Instead of installing packages specified
174in this variable it uses native recipes instead --- a very different
175approach. This has led to confusing errors when binaries are added
176to the SDK but not relocated.
177
178To avoid these issues, a new :term:`TOOLCHAIN_HOST_TASK_ESDK` variable has
179been created. If you wish to extend what is installed in the host
180portion of the eSDK then you will now need to set this variable.
181
182Package/recipe splitting
183~~~~~~~~~~~~~~~~~~~~~~~~
184
185- ``perl-cross`` has been split out from the main ``perl`` recipe to
186  its own ``perlcross`` recipe for maintenance reasons. If you have
187  bbappends for the perl recipe then these may need extending.
188
189- The ``wayland`` recipe now packages its binaries in a
190  ``wayland-tools`` package rather than putting them into
191  ``wayland-dev``.
192
193- Xwayland has been split out of the xserver-xorg tree and thus is now
194  in its own ``xwayland`` recipe. If you need Xwayland in your image
195  then you may now need to add it explicitly.
196
197- The ``rpm`` package no longer has ``rpm-build`` in its :term:`RRECOMMENDS`;
198  if by chance  you still need rpm package building functionality in
199  your image and you have not already done so then you should add
200  ``rpm-build`` to your image explicitly.
201
202- The Python ``statistics`` standard module is now packaged in its own
203  ``python3-statistics`` package instead of ``python3-misc`` as
204  previously.
205
206Image / SDK generation changes
207~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
208
209- Recursive dependencies on the ``do_build`` task are now disabled when
210  building SDKs. These are generally not needed; in the unlikely event
211  that you do encounter problems then it will probably be as a result of
212  missing explicit dependencies that need to be added.
213
214- Errors during "complementary" package installation (e.g. for ``*-dbg``
215  and ``*-dev`` packages) during image construction are no longer
216  ignored. Historically some of these packages had installation problems,
217  that is no longer the case. In the unlikely event that you see errors
218  as a result, you will need to fix the installation/packaging issues.
219
220- When building an image, only packages that will be used in building
221  the image (i.e. the first entry in :term:`PACKAGE_CLASSES`) will be
222  produced if multiple package types are enabled (which is not a typical
223  configuration). If in your CI system you need to have the original
224  behaviour, use ``bitbake --runall build <target>``.
225
226- The ``-lic`` package is no longer automatically added to
227  :term:`RRECOMMENDS` for every other package when
228  :term:`LICENSE_CREATE_PACKAGE` is set to "1". If you wish all license
229  packages to be installed corresponding to packages in your image, then
230  you should instead add the new ``lic-pkgs`` feature to
231  :term:`IMAGE_FEATURES`.
232
233Miscellaneous
234~~~~~~~~~~~~~
235
236- Certificates are now properly checked when BitBake fetches sources
237  over HTTPS. If you receive errors as a result for your custom recipes,
238  you will need to use a mirror or address the issue with the operators
239  of the server in question.
240
241- ``avahi`` has had its GTK+ support disabled by default. If you wish to
242  re-enable it, set ``AVAHI_GTK = "gtk3"`` in a bbappend for the
243  ``avahi`` recipe or in your custom distro configuration file.
244
245- Setting the ``BUILD_REPRODUCIBLE_BINARIES`` variable to "0" no longer
246  uses a strangely old fallback date of April 2011, it instead disables
247  building reproducible binaries as you would logically expect.
248
249- Setting noexec/nostamp/fakeroot varflags to any value besides "1" will
250  now trigger a warning. These should be either set to "1" to enable, or
251  not set at all to disable.
252
253- The previously deprecated ``COMPRESS_CMD`` and
254  ``CVE_CHECK_CVE_WHITELIST`` variables have been removed. Use
255  :term:`CONVERSION_CMD` and ``CVE_CHECK_WHITELIST`` (replaced by
256  :term:`CVE_CHECK_IGNORE` in version 3.5) respectively
257  instead.
258
259- The obsolete ``oe_machinstall`` function previously provided in the
260  :ref:`utils <ref-classes-utils>` class has been removed. For
261  machine-specific installation it is recommended that you use the
262  built-in override support in the fetcher or overrides in general
263  instead.
264
265- The ``-P`` (``--clear-password``) option can no longer be used with
266  ``useradd`` and ``usermod`` entries in :term:`EXTRA_USERS_PARAMS`.
267  It was being implemented using a custom patch to the ``shadow`` recipe
268  which clashed with a ``-P`` option that was added upstream in
269  ``shadow`` version 4.9, and in any case is fundamentally insecure.
270  Hardcoded passwords are still supported but they need to be hashed, see
271  examples in :term:`EXTRA_USERS_PARAMS`.
272
273
274