Andrew Geissler | f034379 | 2020-11-18 10:42:21 -0600 | [diff] [blame] | 1 | .. SPDX-License-Identifier: CC-BY-SA-2.0-UK |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 2 | |
| 3 | *** |
| 4 | FAQ |
| 5 | *** |
| 6 | |
Andrew Geissler | 4c19ea1 | 2020-10-27 13:52:24 -0500 | [diff] [blame] | 7 | **Q:** How does Poky differ from :oe_home:`OpenEmbedded <>`? |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 8 | |
| 9 | **A:** The term ``Poky`` refers to the specific reference build |
| 10 | system that the Yocto Project provides. Poky is based on |
| 11 | :term:`OpenEmbedded-Core (OE-Core)` and :term:`BitBake`. Thus, the |
| 12 | generic term used here for the build system is the "OpenEmbedded build |
| 13 | system." Development in the Yocto Project using Poky is closely tied to |
| 14 | OpenEmbedded, with changes always being merged to OE-Core or BitBake |
| 15 | first before being pulled back into Poky. This practice benefits both |
| 16 | projects immediately. |
| 17 | |
| 18 | **Q:** My development system does not meet the required Git, tar, and |
Andrew Geissler | 3b8a17c | 2021-04-15 15:55:55 -0500 | [diff] [blame] | 19 | Python versions. In particular, I do not have Python &MIN_PYTHON_VERSION; or greater. |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 20 | Can I still use the Yocto Project? |
| 21 | |
| 22 | **A:** You can get the required tools on your host development system a |
| 23 | couple different ways (i.e. building a tarball or downloading a |
Andrew Geissler | 4c19ea1 | 2020-10-27 13:52:24 -0500 | [diff] [blame] | 24 | tarball). See the |
Andrew Geissler | 09209ee | 2020-12-13 08:44:15 -0600 | [diff] [blame] | 25 | ":ref:`ref-manual/system-requirements:required git, tar, python and gcc versions`" |
Andrew Geissler | 4c19ea1 | 2020-10-27 13:52:24 -0500 | [diff] [blame] | 26 | section for steps on how to update your build tools. |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 27 | |
| 28 | **Q:** How can you claim Poky / OpenEmbedded-Core is stable? |
| 29 | |
| 30 | **A:** There are three areas that help with stability; |
| 31 | |
| 32 | - The Yocto Project team keeps :term:`OpenEmbedded-Core (OE-Core)` small and |
| 33 | focused, containing around 830 recipes as opposed to the thousands |
| 34 | available in other OpenEmbedded community layers. Keeping it small |
| 35 | makes it easy to test and maintain. |
| 36 | |
| 37 | - The Yocto Project team runs manual and automated tests using a small, |
| 38 | fixed set of reference hardware as well as emulated targets. |
| 39 | |
| 40 | - The Yocto Project uses an autobuilder, which provides continuous |
| 41 | build and integration tests. |
| 42 | |
| 43 | **Q:** How do I get support for my board added to the Yocto Project? |
| 44 | |
| 45 | **A:** Support for an additional board is added by creating a Board |
| 46 | Support Package (BSP) layer for it. For more information on how to |
| 47 | create a BSP layer, see the |
Andrew Geissler | 09209ee | 2020-12-13 08:44:15 -0600 | [diff] [blame] | 48 | ":ref:`dev-manual/common-tasks:understanding and creating layers`" |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 49 | section in the Yocto Project Development Tasks Manual and the |
Andrew Geissler | 09209ee | 2020-12-13 08:44:15 -0600 | [diff] [blame] | 50 | :doc:`/bsp-guide/index`. |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 51 | |
| 52 | Usually, if the board is not completely exotic, adding support in the |
| 53 | Yocto Project is fairly straightforward. |
| 54 | |
| 55 | **Q:** Are there any products built using the OpenEmbedded build system? |
| 56 | |
| 57 | **A:** The software running on the `Vernier |
Andrew Geissler | d1e8949 | 2021-02-12 15:35:20 -0600 | [diff] [blame] | 58 | LabQuest <https://vernier.com/labquest/>`__ is built using the |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 59 | OpenEmbedded build system. See the `Vernier |
Andrew Geissler | d1e8949 | 2021-02-12 15:35:20 -0600 | [diff] [blame] | 60 | LabQuest <https://www.vernier.com/products/interfaces/labq/>`__ website |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 61 | for more information. There are a number of pre-production devices using |
| 62 | the OpenEmbedded build system and the Yocto Project team announces them |
| 63 | as soon as they are released. |
| 64 | |
| 65 | **Q:** What does the OpenEmbedded build system produce as output? |
| 66 | |
| 67 | **A:** Because you can use the same set of recipes to create output of |
| 68 | various formats, the output of an OpenEmbedded build depends on how you |
| 69 | start it. Usually, the output is a flashable image ready for the target |
| 70 | device. |
| 71 | |
| 72 | **Q:** How do I add my package to the Yocto Project? |
| 73 | |
| 74 | **A:** To add a package, you need to create a BitBake recipe. For |
| 75 | information on how to create a BitBake recipe, see the |
Andrew Geissler | 09209ee | 2020-12-13 08:44:15 -0600 | [diff] [blame] | 76 | ":ref:`dev-manual/common-tasks:writing a new recipe`" |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 77 | section in the Yocto Project Development Tasks Manual. |
| 78 | |
| 79 | **Q:** Do I have to reflash my entire board with a new Yocto Project |
| 80 | image when recompiling a package? |
| 81 | |
| 82 | **A:** The OpenEmbedded build system can build packages in various |
| 83 | formats such as IPK for OPKG, Debian package (``.deb``), or RPM. You can |
| 84 | then upgrade the packages using the package tools on the device, much |
| 85 | like on a desktop distribution such as Ubuntu or Fedora. However, |
| 86 | package management on the target is entirely optional. |
| 87 | |
| 88 | **Q:** I see the error |
| 89 | '``chmod: XXXXX new permissions are r-xrwxrwx, not r-xr-xr-x``'. What is |
| 90 | wrong? |
| 91 | |
| 92 | **A:** You are probably running the build on an NTFS filesystem. Use |
| 93 | ``ext2``, ``ext3``, or ``ext4`` instead. |
| 94 | |
| 95 | **Q:** I see lots of 404 responses for files when the OpenEmbedded build |
| 96 | system is trying to download sources. Is something wrong? |
| 97 | |
| 98 | **A:** Nothing is wrong. The OpenEmbedded build system checks any |
| 99 | configured source mirrors before downloading from the upstream sources. |
| 100 | The build system does this searching for both source archives and |
| 101 | pre-checked out versions of SCM-managed software. These checks help in |
| 102 | large installations because it can reduce load on the SCM servers |
| 103 | themselves. The address above is one of the default mirrors configured |
| 104 | into the build system. Consequently, if an upstream source disappears, |
| 105 | the team can place sources there so builds continue to work. |
| 106 | |
| 107 | **Q:** I have machine-specific data in a package for one machine only |
| 108 | but the package is being marked as machine-specific in all cases, how do |
| 109 | I prevent this? |
| 110 | |
| 111 | **A:** Set ``SRC_URI_OVERRIDES_PACKAGE_ARCH`` = "0" in the ``.bb`` file |
| 112 | but make sure the package is manually marked as machine-specific for the |
| 113 | case that needs it. The code that handles |
| 114 | ``SRC_URI_OVERRIDES_PACKAGE_ARCH`` is in the |
| 115 | ``meta/classes/base.bbclass`` file. |
| 116 | |
| 117 | **Q:** I'm behind a firewall and need to use a proxy server. How do I do |
| 118 | that? |
| 119 | |
| 120 | **A:** Most source fetching by the OpenEmbedded build system is done by |
| 121 | ``wget`` and you therefore need to specify the proxy settings in a |
| 122 | ``.wgetrc`` file, which can be in your home directory if you are a |
| 123 | single user or can be in ``/usr/local/etc/wgetrc`` as a global user |
| 124 | file. |
| 125 | |
| 126 | Following is the applicable code for setting various proxy types in the |
| 127 | ``.wgetrc`` file. By default, these settings are disabled with comments. |
Andrew Geissler | c926e17 | 2021-05-07 16:11:35 -0500 | [diff] [blame] | 128 | To use them, remove the comments:: |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 129 | |
| 130 | # You can set the default proxies for Wget to use for http, https, and ftp. |
| 131 | # They will override the value in the environment. |
| 132 | #https_proxy = http://proxy.yoyodyne.com:18023/ |
| 133 | #http_proxy = http://proxy.yoyodyne.com:18023/ |
| 134 | #ftp_proxy = http://proxy.yoyodyne.com:18023/ |
| 135 | |
| 136 | # If you do not want to use proxy at all, set this to off. |
| 137 | #use_proxy = on |
| 138 | |
| 139 | The Yocto Project also includes a |
| 140 | ``meta-poky/conf/site.conf.sample`` file that shows how to configure CVS |
| 141 | and Git proxy servers if needed. For more information on setting up |
| 142 | various proxy types and configuring proxy servers, see the |
Andrew Geissler | 09209ee | 2020-12-13 08:44:15 -0600 | [diff] [blame] | 143 | ":yocto_wiki:`Working Behind a Network Proxy </Working_Behind_a_Network_Proxy>`" |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 144 | Wiki page. |
| 145 | |
| 146 | **Q:** What's the difference between target and target\ ``-native``? |
| 147 | |
| 148 | **A:** The ``*-native`` targets are designed to run on the system being |
| 149 | used for the build. These are usually tools that are needed to assist |
| 150 | the build in some way such as ``quilt-native``, which is used to apply |
| 151 | patches. The non-native version is the one that runs on the target |
| 152 | device. |
| 153 | |
| 154 | **Q:** I'm seeing random build failures. Help?! |
| 155 | |
| 156 | **A:** If the same build is failing in totally different and random |
| 157 | ways, the most likely explanation is: |
| 158 | |
| 159 | - The hardware you are running the build on has some problem. |
| 160 | |
| 161 | - You are running the build under virtualization, in which case the |
| 162 | virtualization probably has bugs. |
| 163 | |
| 164 | The OpenEmbedded build system processes a massive amount of data that |
| 165 | causes lots of network, disk and CPU activity and is sensitive to even |
| 166 | single-bit failures in any of these areas. True random failures have |
| 167 | always been traced back to hardware or virtualization issues. |
| 168 | |
| 169 | **Q:** When I try to build a native recipe, the build fails with |
| 170 | ``iconv.h`` problems. |
| 171 | |
| 172 | **A:** If you get an error message that indicates GNU ``libiconv`` is |
| 173 | not in use but ``iconv.h`` has been included from ``libiconv``, you need |
| 174 | to check to see if you have a previously installed version of the header |
| 175 | file in ``/usr/local/include``. |
| 176 | :: |
| 177 | |
| 178 | #error GNU libiconv not in use but included iconv.h is from libiconv |
| 179 | |
| 180 | If you find a previously installed |
| 181 | file, you should either uninstall it or temporarily rename it and try |
| 182 | the build again. |
| 183 | |
| 184 | This issue is just a single manifestation of "system leakage" issues |
| 185 | caused when the OpenEmbedded build system finds and uses previously |
| 186 | installed files during a native build. This type of issue might not be |
| 187 | limited to ``iconv.h``. Be sure that leakage cannot occur from |
| 188 | ``/usr/local/include`` and ``/opt`` locations. |
| 189 | |
| 190 | **Q:** What do we need to ship for license compliance? |
| 191 | |
| 192 | **A:** This is a difficult question and you need to consult your lawyer |
| 193 | for the answer for your specific case. It is worth bearing in mind that |
| 194 | for GPL compliance, there needs to be enough information shipped to |
| 195 | allow someone else to rebuild and produce the same end result you are |
| 196 | shipping. This means sharing the source code, any patches applied to it, |
| 197 | and also any configuration information about how that package was |
| 198 | configured and built. |
| 199 | |
| 200 | You can find more information on licensing in the |
Andrew Geissler | 09209ee | 2020-12-13 08:44:15 -0600 | [diff] [blame] | 201 | ":ref:`overview-manual/development-environment:licensing`" |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 202 | section in the Yocto |
| 203 | Project Overview and Concepts Manual and also in the |
Andrew Geissler | 09209ee | 2020-12-13 08:44:15 -0600 | [diff] [blame] | 204 | ":ref:`dev-manual/common-tasks:maintaining open source license compliance during your product's lifecycle`" |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 205 | section in the Yocto Project Development Tasks Manual. |
| 206 | |
| 207 | **Q:** How do I disable the cursor on my touchscreen device? |
| 208 | |
| 209 | **A:** You need to create a form factor file as described in the |
Andrew Geissler | 6ce62a2 | 2020-11-30 19:58:47 -0600 | [diff] [blame] | 210 | ":ref:`bsp-guide/bsp:miscellaneous bsp-specific recipe files`" section in |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 211 | the Yocto Project Board Support Packages (BSP) Developer's Guide. Set |
Andrew Geissler | c926e17 | 2021-05-07 16:11:35 -0500 | [diff] [blame] | 212 | the ``HAVE_TOUCHSCREEN`` variable equal to one as follows:: |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 213 | |
| 214 | HAVE_TOUCHSCREEN=1 |
| 215 | |
| 216 | **Q:** How do I make sure connected network interfaces are brought up by |
| 217 | default? |
| 218 | |
| 219 | **A:** The default interfaces file provided by the netbase recipe does |
| 220 | not automatically bring up network interfaces. Therefore, you will need |
| 221 | to add a BSP-specific netbase that includes an interfaces file. See the |
Andrew Geissler | 6ce62a2 | 2020-11-30 19:58:47 -0600 | [diff] [blame] | 222 | ":ref:`bsp-guide/bsp:miscellaneous bsp-specific recipe files`" section in |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 223 | the Yocto Project Board Support Packages (BSP) Developer's Guide for |
| 224 | information on creating these types of miscellaneous recipe files. |
| 225 | |
Andrew Geissler | c926e17 | 2021-05-07 16:11:35 -0500 | [diff] [blame] | 226 | For example, add the following files to your layer:: |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 227 | |
| 228 | meta-MACHINE/recipes-bsp/netbase/netbase/MACHINE/interfaces |
| 229 | meta-MACHINE/recipes-bsp/netbase/netbase_5.0.bbappend |
| 230 | |
| 231 | **Q:** How do I create images with more free space? |
| 232 | |
| 233 | **A:** By default, the OpenEmbedded build system creates images that are |
| 234 | 1.3 times the size of the populated root filesystem. To affect the image |
| 235 | size, you need to set various configurations: |
| 236 | |
| 237 | - *Image Size:* The OpenEmbedded build system uses the |
| 238 | :term:`IMAGE_ROOTFS_SIZE` variable to define |
| 239 | the size of the image in Kbytes. The build system determines the size |
| 240 | by taking into account the initial root filesystem size before any |
| 241 | modifications such as requested size for the image and any requested |
| 242 | additional free disk space to be added to the image. |
| 243 | |
| 244 | - *Overhead:* Use the |
| 245 | :term:`IMAGE_OVERHEAD_FACTOR` variable |
| 246 | to define the multiplier that the build system applies to the initial |
| 247 | image size, which is 1.3 by default. |
| 248 | |
| 249 | - *Additional Free Space:* Use the |
| 250 | :term:`IMAGE_ROOTFS_EXTRA_SPACE` |
| 251 | variable to add additional free space to the image. The build system |
| 252 | adds this space to the image after it determines its |
| 253 | ``IMAGE_ROOTFS_SIZE``. |
| 254 | |
| 255 | **Q:** Why don't you support directories with spaces in the pathnames? |
| 256 | |
| 257 | **A:** The Yocto Project team has tried to do this before but too many |
| 258 | of the tools the OpenEmbedded build system depends on, such as |
| 259 | ``autoconf``, break when they find spaces in pathnames. Until that |
| 260 | situation changes, the team will not support spaces in pathnames. |
| 261 | |
| 262 | **Q:** How do I use an external toolchain? |
| 263 | |
| 264 | **A:** The toolchain configuration is very flexible and customizable. It |
| 265 | is primarily controlled with the ``TCMODE`` variable. This variable |
| 266 | controls which ``tcmode-*.inc`` file to include from the |
| 267 | ``meta/conf/distro/include`` directory within the :term:`Source Directory`. |
| 268 | |
| 269 | The default value of ``TCMODE`` is "default", which tells the |
| 270 | OpenEmbedded build system to use its internally built toolchain (i.e. |
| 271 | ``tcmode-default.inc``). However, other patterns are accepted. In |
| 272 | particular, "external-\*" refers to external toolchains. One example is |
| 273 | the Sourcery G++ Toolchain. The support for this toolchain resides in |
| 274 | the separate ``meta-sourcery`` layer at |
Andrew Geissler | d1e8949 | 2021-02-12 15:35:20 -0600 | [diff] [blame] | 275 | https://github.com/MentorEmbedded/meta-sourcery/. |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 276 | |
| 277 | In addition to the toolchain configuration, you also need a |
| 278 | corresponding toolchain recipe file. This recipe file needs to package |
| 279 | up any pre-built objects in the toolchain such as ``libgcc``, |
| 280 | ``libstdcc++``, any locales, and ``libc``. |
| 281 | |
| 282 | **Q:** How does the OpenEmbedded build system obtain source code and |
| 283 | will it work behind my firewall or proxy server? |
| 284 | |
| 285 | **A:** The way the build system obtains source code is highly |
| 286 | configurable. You can setup the build system to get source code in most |
| 287 | environments if HTTP transport is available. |
| 288 | |
| 289 | When the build system searches for source code, it first tries the local |
| 290 | download directory. If that location fails, Poky tries |
| 291 | :term:`PREMIRRORS`, the upstream source, and then |
| 292 | :term:`MIRRORS` in that order. |
| 293 | |
| 294 | Assuming your distribution is "poky", the OpenEmbedded build system uses |
| 295 | the Yocto Project source ``PREMIRRORS`` by default for SCM-based |
| 296 | sources, upstreams for normal tarballs, and then falls back to a number |
| 297 | of other mirrors including the Yocto Project source mirror if those |
| 298 | fail. |
| 299 | |
| 300 | As an example, you could add a specific server for the build system to |
| 301 | attempt before any others by adding something like the following to the |
Andrew Geissler | c926e17 | 2021-05-07 16:11:35 -0500 | [diff] [blame] | 302 | ``local.conf`` configuration file:: |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 303 | |
| 304 | PREMIRRORS_prepend = "\ |
| 305 | git://.*/.* http://www.yoctoproject.org/sources/ \n \ |
| 306 | ftp://.*/.* http://www.yoctoproject.org/sources/ \n \ |
| 307 | http://.*/.* http://www.yoctoproject.org/sources/ \n \ |
| 308 | https://.*/.* http://www.yoctoproject.org/sources/ \n" |
| 309 | |
| 310 | These changes cause the build system to intercept Git, FTP, HTTP, and |
| 311 | HTTPS requests and direct them to the ``http://`` sources mirror. You |
| 312 | can use ``file://`` URLs to point to local directories or network shares |
| 313 | as well. |
| 314 | |
Andrew Geissler | c926e17 | 2021-05-07 16:11:35 -0500 | [diff] [blame] | 315 | Aside from the previous technique, these options also exist:: |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 316 | |
| 317 | BB_NO_NETWORK = "1" |
| 318 | |
| 319 | This statement tells BitBake to issue an error |
| 320 | instead of trying to access the Internet. This technique is useful if |
| 321 | you want to ensure code builds only from local sources. |
| 322 | |
Andrew Geissler | c926e17 | 2021-05-07 16:11:35 -0500 | [diff] [blame] | 323 | Here is another technique:: |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 324 | |
| 325 | BB_FETCH_PREMIRRORONLY = "1" |
| 326 | |
| 327 | This statement |
| 328 | limits the build system to pulling source from the ``PREMIRRORS`` only. |
| 329 | Again, this technique is useful for reproducing builds. |
| 330 | |
Andrew Geissler | c926e17 | 2021-05-07 16:11:35 -0500 | [diff] [blame] | 331 | Here is another technique:: |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 332 | |
| 333 | BB_GENERATE_MIRROR_TARBALLS = "1" |
| 334 | |
| 335 | This |
| 336 | statement tells the build system to generate mirror tarballs. This |
| 337 | technique is useful if you want to create a mirror server. If not, |
| 338 | however, the technique can simply waste time during the build. |
| 339 | |
| 340 | Finally, consider an example where you are behind an HTTP-only firewall. |
| 341 | You could make the following changes to the ``local.conf`` configuration |
Andrew Geissler | c926e17 | 2021-05-07 16:11:35 -0500 | [diff] [blame] | 342 | file as long as the ``PREMIRRORS`` server is current:: |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 343 | |
| 344 | PREMIRRORS_prepend = "\ |
| 345 | ftp://.*/.* http://www.yoctoproject.org/sources/ \n \ |
| 346 | http://.*/.* http://www.yoctoproject.org/sources/ \n \ |
| 347 | https://.*/.* http://www.yoctoproject.org/sources/ \n" |
| 348 | BB_FETCH_PREMIRRORONLY = "1" |
| 349 | |
| 350 | These changes would cause the build system to successfully fetch source |
| 351 | over HTTP and any network accesses to anything other than the |
| 352 | ``PREMIRRORS`` would fail. |
| 353 | |
| 354 | The build system also honors the standard shell environment variables |
| 355 | ``http_proxy``, ``ftp_proxy``, ``https_proxy``, and ``all_proxy`` to |
| 356 | redirect requests through proxy servers. |
| 357 | |
| 358 | .. note:: |
| 359 | |
| 360 | You can find more information on the |
Andrew Geissler | 09209ee | 2020-12-13 08:44:15 -0600 | [diff] [blame] | 361 | ":yocto_wiki:`Working Behind a Network Proxy </Working_Behind_a_Network_Proxy>`" |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 362 | Wiki page. |
| 363 | |
| 364 | **Q:** Can I get rid of build output so I can start over? |
| 365 | |
| 366 | **A:** Yes - you can easily do this. When you use BitBake to build an |
| 367 | image, all the build output goes into the directory created when you run |
| 368 | the build environment setup script (i.e. |
Andrew Geissler | 4c19ea1 | 2020-10-27 13:52:24 -0500 | [diff] [blame] | 369 | :ref:`structure-core-script`). By default, this :term:`Build Directory` |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 370 | is named ``build`` but can be named |
| 371 | anything you want. |
| 372 | |
| 373 | Within the Build Directory, is the ``tmp`` directory. To remove all the |
| 374 | build output yet preserve any source code or downloaded files from |
| 375 | previous builds, simply remove the ``tmp`` directory. |
| 376 | |
| 377 | **Q:** Why do ``${bindir}`` and ``${libdir}`` have strange values for |
| 378 | ``-native`` recipes? |
| 379 | |
| 380 | **A:** Executables and libraries might need to be used from a directory |
| 381 | other than the directory into which they were initially installed. |
| 382 | Complicating this situation is the fact that sometimes these executables |
| 383 | and libraries are compiled with the expectation of being run from that |
| 384 | initial installation target directory. If this is the case, moving them |
| 385 | causes problems. |
| 386 | |
| 387 | This scenario is a fundamental problem for package maintainers of |
| 388 | mainstream Linux distributions as well as for the OpenEmbedded build |
| 389 | system. As such, a well-established solution exists. Makefiles, |
| 390 | Autotools configuration scripts, and other build systems are expected to |
| 391 | respect environment variables such as ``bindir``, ``libdir``, and |
| 392 | ``sysconfdir`` that indicate where executables, libraries, and data |
| 393 | reside when a program is actually run. They are also expected to respect |
| 394 | a ``DESTDIR`` environment variable, which is prepended to all the other |
| 395 | variables when the build system actually installs the files. It is |
| 396 | understood that the program does not actually run from within |
| 397 | ``DESTDIR``. |
| 398 | |
| 399 | When the OpenEmbedded build system uses a recipe to build a |
| 400 | target-architecture program (i.e. one that is intended for inclusion on |
| 401 | the image being built), that program eventually runs from the root file |
| 402 | system of that image. Thus, the build system provides a value of |
| 403 | "/usr/bin" for ``bindir``, a value of "/usr/lib" for ``libdir``, and so |
| 404 | forth. |
| 405 | |
| 406 | Meanwhile, ``DESTDIR`` is a path within the :term:`Build Directory`. |
| 407 | However, when the recipe builds a |
| 408 | native program (i.e. one that is intended to run on the build machine), |
| 409 | that program is never installed directly to the build machine's root |
| 410 | file system. Consequently, the build system uses paths within the Build |
| 411 | Directory for ``DESTDIR``, ``bindir`` and related variables. To better |
| 412 | understand this, consider the following two paths where the first is |
Andrew Geissler | 4c19ea1 | 2020-10-27 13:52:24 -0500 | [diff] [blame] | 413 | relatively normal and the second is not: |
| 414 | |
| 415 | .. note:: |
| 416 | |
| 417 | Due to these lengthy examples, the paths are artificially broken |
| 418 | across lines for readability. |
| 419 | |
| 420 | :: |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 421 | |
| 422 | /home/maxtothemax/poky-bootchart2/build/tmp/work/i586-poky-linux/zlib/ |
| 423 | 1.2.8-r0/sysroot-destdir/usr/bin |
| 424 | |
| 425 | /home/maxtothemax/poky-bootchart2/build/tmp/work/x86_64-linux/ |
| 426 | zlib-native/1.2.8-r0/sysroot-destdir/home/maxtothemax/poky-bootchart2/ |
| 427 | build/tmp/sysroots/x86_64-linux/usr/bin |
| 428 | |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 429 | Even if the paths look unusual, |
| 430 | they both are correct - the first for a target and the second for a |
| 431 | native recipe. These paths are a consequence of the ``DESTDIR`` |
| 432 | mechanism and while they appear strange, they are correct and in |
| 433 | practice very effective. |
| 434 | |
| 435 | **Q:** The files provided by my ``*-native`` recipe do not appear to be |
| 436 | available to other recipes. Files are missing from the native sysroot, |
| 437 | my recipe is installing to the wrong place, or I am getting permissions |
| 438 | errors during the do_install task in my recipe! What is wrong? |
| 439 | |
| 440 | **A:** This situation results when a build system does not recognize the |
| 441 | environment variables supplied to it by :term:`BitBake`. The |
| 442 | incident that prompted this FAQ entry involved a Makefile that used an |
| 443 | environment variable named ``BINDIR`` instead of the more standard |
| 444 | variable ``bindir``. The makefile's hardcoded default value of |
| 445 | "/usr/bin" worked most of the time, but not for the recipe's ``-native`` |
| 446 | variant. For another example, permissions errors might be caused by a |
| 447 | Makefile that ignores ``DESTDIR`` or uses a different name for that |
Andrew Geissler | 3b8a17c | 2021-04-15 15:55:55 -0500 | [diff] [blame] | 448 | environment variable. Check the build system to see if these kinds |
Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame] | 449 | of issues exist. |
Andrew Geissler | 6ce62a2 | 2020-11-30 19:58:47 -0600 | [diff] [blame] | 450 | |
| 451 | **Q:** I'm adding a binary in a recipe but it's different in the image, what is |
| 452 | changing it? |
| 453 | |
| 454 | **A:** The first most obvious change is the system stripping debug symbols from |
| 455 | it. Setting :term:`INHIBIT_PACKAGE_STRIP` to stop debug symbols being stripped and/or |
| 456 | :term:`INHIBIT_PACKAGE_DEBUG_SPLIT` to stop debug symbols being split into a separate |
| 457 | file will ensure the binary is unchanged. The other less obvious thing that can |
| 458 | happen is prelinking of the image. This is set by default in local.conf via |
| 459 | :term:`USER_CLASSES` which can contain 'image-prelink'. If you remove that, the |
| 460 | image will not be prelinked meaning the binaries would be unchanged. |