You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

368 lines
21 KiB

  1. # Python packages folder
  2. :warning: **Python 2 will soon be unsupported and removed from the feed - [see below](#python-2-end-of-life)** :warning:
  3. ## Table of contents
  4. 1. [Description](#description)
  5. 2. [Python 2 end-of-life](#python-2-end-of-life)
  6. 1. [Transition policy / schedule](#transition-policy--schedule)
  7. 3. [Introduction](#introduction)
  8. 4. [Using Python[3] in external/other package feeds](#using-python3-in-externalother-package-feeds)
  9. 5. [Build considerations](#build-considerations)
  10. 6. [General folder structure](#general-folder-structure)
  11. 7. [Building a Python[3] package](#building-a-python3-package)
  12. 1. [PKG_BUILD_DIR](#pkg_build_dir)
  13. 2. [PKG_UNPACK](#pkg_unpack)
  14. 3. [Include python[3]-package.mk](#include-python3-packagemk)
  15. 4. [Add Package/<PKG_NAME> OpenWrt definitions](#add-packagepkg_name-openwrt-definitions)
  16. 5. [Wrapping things up so that they build](#wrapping-things-up-so-that-they-build)
  17. 6. [Customizing things](#customizing-things)
  18. 7. [Host-side Python packages for build](#host-side-python-packages-for-build)
  19. ## Description
  20. This section describes specifics for the Python packages that are present in this repo, and how things are structured.
  21. In terms of license, contributing guide, etc, all of that information is described in the top [README.md](README.md) file, and it applies here as well. This document attempts to cover only technical aspects of Python/Python3 packages, and maybe some explanations about how things are (and why they are as they are).
  22. ## Python 2 end-of-life
  23. Python 2 will not be maintained past [1 January 2020](https://pythonclock.org/). As such, we will be transitioning Python 2 programs and libraries to Python 3, and Python 2 packages will be removed in early 2020.
  24. (Discussion for how to handle this transition can be found in [#8520](https://github.com/openwrt/packages/issues/8520).)
  25. ### Transition policy / schedule
  26. A mass removal event ("The Snap") will occur on 31 March 2020, or 2 weeks before the freeze for a 20.x release, whichever is sooner. The exact date will be confirmed when the 20.x release schedule is known, or by 15 March 2020.
  27. All Python 2 packages (the Python 2 interpreter, programs that depend on Python 2, and Python 2-only libraries) will be removed during this event.
  28. Leading up to "The Snap":
  29. * In general, new Python 2 packages are no longer accepted
  30. * Exceptions can be made on a case-by-case basis, given extraordinary circumstances or reasons, until 31 May 2019
  31. * From 31 May 2019 onward, absolutely no new Python 2 packages will be accepted
  32. * The Python 2 interpreter will remain in the feed until "The Snap"
  33. * The interpreter will continue to be updated, including the last release in January 2020 (if there is one)
  34. * Programs that depend on Python 2 will be transitioned to Python 3 (see [#8893](https://github.com/openwrt/packages/issues/8893))
  35. * If a program cannot be transitioned, a suitable replacement will be found
  36. * If a replacement cannot be found, the program will be removed during "The Snap"
  37. * Python 2 libraries will remain in the feed until "The Snap"
  38. * A Python 2-only library will be transitioned to Python 3 (or a suitable replacement found), if its Python 3 version is a dependency of another package in the feed
  39. * Python 2 libraries will receive normal updates until 31 October 2019
  40. * From 31 October 2019 onward:
  41. * Python 2-only libraries will receive security updates only
  42. * Python 2 libraries that share the same Makefile as their Python 3 version will continue to receive normal updates
  43. ## Introduction
  44. This sub-tree came to exist after a number of contributions (Python packages) were made to this repo, and the [lang](lang) subtree grew to a point where a decision was made to move all Python packages under [lang/python](lang/python).
  45. It contains the 2 Python interpreters (Python & Python3) and Python packages. Most of the Python packages are downloaded from [pypi.org](https://pypi.org/). Python packages from [pypi.org](https://pypi.org/) are typically preferred when adding new packages.
  46. If more packages (than the ones packaged here) are needed, they can be downloaded via [pip or pip3](https://pip.pypa.io). Note that the versions of `pip` & `setuptools` [available in this repo] are the ones that are packaged inside the Python & Python3 packages (yes, Python & Python3 come packaged with `pip` & `setuptools`).
  47. ## Using Python[3] in external/other package feeds
  48. In the feeds.conf (or feeds.conf.default file, whatever is preferred), the packages repo should be present.
  49. Example
  50. ```
  51. src-git packages https://git.openwrt.org/feed/packages.git
  52. src-git luci https://git.openwrt.org/project/luci.git
  53. src-git routing https://git.openwrt.org/feed/routing.git
  54. src-git telephony https://git.openwrt.org/feed/telephony.git
  55. #
  56. #
  57. src-git someotherfeed https://github.com/<github-user>/<some-other-package>
  58. ```
  59. Assuming that there are Python packages in the `<some-other-package>`, they should include `python[3]-package.mk` like this:
  60. ```
  61. include $(TOPDIR)/feeds/packages/lang/python/python-package.mk
  62. include $(TOPDIR)/feeds/packages/lang/python/python3-package.mk
  63. ```
  64. Same rules apply for `python[3]-package.mk` as the Python packages in this repo.
  65. And if only 1 of `python-package.mk` or `python3-package.mk` is needed, then only the needed mk file should be included (though it's not an issue if both are included).
  66. **One important consideration:**: if the local name is not `packages`, it's something else, like `openwrt-packages`. And in `feeds.conf[.default]` it's:
  67. ```
  68. src-git openwrt-packages https://git.openwrt.org/feed/packages.git
  69. ```
  70. Then, the inclusions also change:
  71. ```
  72. include $(TOPDIR)/feeds/openwrt-packages/lang/python/python-package.mk
  73. include $(TOPDIR)/feeds/openwrt-packages/lang/python/python3-package.mk
  74. ```
  75. Each maintainer[s] of external packages feeds is responsible for the local name, and relative inclusion path back to this feed (which is named `packages` by default).
  76. ## Build considerations
  77. In order to build the Python[3] interpreters, a host Python/Python3 interpreter needs to be built, in order to process some of the build for the target Python/Python3 build. The host Python[3] interpreters are also needed so that Python bytecodes are generated, so the host interpreters need to be the exact versions as on the target. And finally, the host Python[3] interpreters also provide pip & pip3, so that they may be used to install some Python[3] packages that are required to build other Python[3] packages.
  78. That's why you'll also see a Python/Python3 build & staging directories.
  79. As you're probably thinking, this sounds [and is] somewhat too much complication [just for packaging], but the status of things is-as-it-is, and it's probably much worse than what's currently visible on the surface [with respect to packaging Python[3] & packages].
  80. As mentioned earlier, Python[3] packages are shipped with bytecodes, and the reason for this is simply performance & size.
  81. The thought/discussion matrix derives a bit like this:
  82. * shipping both Python source-code & bytecodes takes too much space on some devices ; Python source code & byte-code take about similar disk-size
  83. * shipping only Python source code has a big performance penalty [on some lower end systems] ; something like 500 msecs (Python source-only) -> 70 msecs (Python byte-codes) time reduction for a simple "Hello World" script
  84. * shipping only Python byte-codes seems like a good trade-off, and this means that `python-src` & `python3-src` can be provided for people that want the source code
  85. By default, automatic Python[3] byte-code generation is disabled when running a Python script, in order to prevent a disk from accidentally filling up. Since some disks reside in RAM, this also means not filling up the RAM. If someone wants to convert Python source to byte-code then he/she is free to compile it [directly on the device] manually via the Python interpreter & library.
  86. ## General folder structure
  87. The basis of all these packages are:
  88. * [lang/python/python](lang/python/python) - The Python 2.7.y interpreter (supposedly, there won't ever by a 2.8.y)
  89. * [lang/python/python3](lang/python/python3) - The Python 3.x.y interpreter
  90. These 2 are normal OpenWrt packages, which will build the Python interpreters. They also provide `python[3]-pip` & `python[3]-setuptools`. Each Python or Python3 package is actually split into multiple sub-packages [e.g. python-email, python-sqlite3, etc]. This can be viewed inside [lang/python/python/files](lang/python/python/files) & [lang/python/python3/files](lang/python/python3/files).
  91. The reason for this splitting, is purely to offer a way for some people to package Python/Python3 in as-minimal-as-possible-and-still-runable way, and also to be somewhat maintainable when packaging. A standard Python[3] installation can take ~20-30 MBs of disk, which can be somewhat big for some people, so there are the `python[3]-base` packages which bring that down to ~5 MBs. This seems to be good enough (and interesting) for a number of people.
  92. The Python[3] interpreters are structured like this:
  93. * `python-base` (and `python3-base`), which is just the minimal package to startup Python[3] and run basic commands
  94. * `python` (and `python3`) are meta-packages, which install almost everything (python[3]-base [plus] Python[3] library [minus] some unit-tests & some windows-y things)
  95. * `python-light` (and `python3-light`) are `python` (and `python3`) [minus] packages that are in [lang/python/python/files](lang/python/python/files) or [lang/python/python3/files](lang/python/python3/files) ; the size of these 2 packages may be sensible (and interesting) to another group of people
  96. All other Python & Python3 packages (aside from the 2 intepreters) typically use these files:
  97. * **python[3]-host.mk** - this file contains paths and build rules for running the Python[3] interpreters on the host-side; they also provide paths to host interprete, host Python lib-dir & so on
  98. * **python[3]-package.mk**
  99. * includes **python[3]-host.mk**
  100. * contains all the default build rules for Python[3] packages; these will be detailed below in the [Building a Python[3] package](#Building a Python[3] package) section
  101. **Note** that Python/Python3 packages don't need to use these files (i.e. `python[3]-package.mk` & `python[3]-host.mk`), but they do provide some ease-of-use & reduction of duplicate code, especially when packaging for both Python & Python3. And they do contain some learned-lessons about packaging Python/Python3 packages, so it's a good idea to use them.
  102. ## Building a Python[3] package
  103. A Python package can be packaged for either Python or Python3 or both.
  104. This section will describe both, and then it can be inferred which is for which.
  105. Packaging for both Python & Python3 uses the `VARIANT` mechanism for packaging inside OpenWrt. (#### FIXME: find a link for this later if it exists)
  106. ### PKG_BUILD_DIR
  107. It's important when packaging for both Python & Python3 to override this variable, so that the build directory differs for each variant.
  108. Typically it's just something like:
  109. ```
  110. PKG_BUILD_DIR:=$(BUILD_DIR)/$(BUILD_VARIANT)-pyasn1-$(PKG_VERSION)
  111. ```
  112. Where `pyasn1` should be some other name, or maybe `PKG_NAME`
  113. This should be added before this include:
  114. ```
  115. include $(INCLUDE_DIR)/package.mk
  116. ```
  117. ### PKG_UNPACK
  118. In many cases, this needs to be overriden. This is usually because the way Python packages are archived, don't follow the convention of other `tar.gz` packages.
  119. So, something like:
  120. ```
  121. PKG_UNPACK=$(HOST_TAR) -C $(PKG_BUILD_DIR) --strip-components=1 -xzf $(DL_DIR)/$(PKG_SOURCE)
  122. ```
  123. should be added.
  124. It's not important whether this is after or before `include $(INCLUDE_DIR)/package.mk`
  125. ### Include python[3]-package.mk
  126. If packaging for Python, add this after `include $(INCLUDE_DIR)/package.mk`
  127. ```
  128. include ../python-package.mk
  129. ```
  130. If packaging for Python3, add this after `include $(INCLUDE_DIR)/package.mk`
  131. ```
  132. include ../python3-package.mk
  133. ```
  134. Order doesn't matter between `python-package.mk` & `python3-package.mk`.
  135. These will make sure that build rules for Python or Python3 can be specified and picked up for build.
  136. ### Add Package/<PKG_NAME> OpenWrt definitions
  137. This part is similar to default OpenWrt packages.
  138. It's usually recommended to have a `Package/<PKG_NAME>/Default` section that's common for both Python & Python3.
  139. Example:
  140. ```
  141. define Package/python-lxml/Default
  142. SECTION:=lang
  143. CATEGORY:=Languages
  144. SUBMENU:=Python
  145. URL:=https://lxml.de
  146. DEPENDS:=+libxml2 +libxslt +libexslt
  147. endef
  148. ```
  149. Then for each variant do something like:
  150. ```
  151. define Package/python-lxml
  152. $(call Package/python-lxml/Default)
  153. TITLE:=python-lxml
  154. DEPENDS+=+PACKAGE_python-lxml:python-light +PACKAGE_python-lxml:python-codecs
  155. VARIANT:=python
  156. endef
  157. define Package/python3-lxml
  158. $(call Package/python-lxml/Default)
  159. TITLE:=python3-lxml
  160. DEPENDS+=+PACKAGE_python3-lxml:python3-light
  161. VARIANT:=python3
  162. endef
  163. ```
  164. Some considerations here (based on the example above):
  165. * be sure to make sure that `DEPENDS` are correct for both variants; as seen in the example above, `python-codecs` is needed only for `python-lxml` (see **[note-encodings](#note-encodings)**)
  166. * consider adding conditional DEPENDS for each variant ; so for each Python[3] package add `+PACKAGE_python-lxml:<dep>` as seen in the above example ; the reason for this is build-time reduction ; if you want to build Python3 only packages, this won't build Python & Python packages + dependencies ; this is a known functionality of OpenWrt build deps
  167. * this should not happen anymore, but if adding `+PACKAGE_python-lxml` conditional deps creates circular dependencies, then open an issue so this can be resolved again.
  168. * `VARIANT=python` or `VARIANT=python3` must be added
  169. * typically each variant package is named `Package/python-<something>` & `Package/python3-<something>` ; this convention makes things easier to follow, though it could work without naming things this way
  170. * `TITLE` can be something a bit more verbose/neat ; typically the name is short as seen above
  171. <a name="note-encodings">**note-encodings**</a>: That's because some character encodings are needed, which are present in `python3-base` but not in `python-light` (but are present in `python-codecs`) ; this is because Python3 is designed to be more Unicode friendly than Python2 (it's one of the fundamental differences between the 2), and Python3 won't start without those encodings being present.
  172. Following these, 2 more definitions are required:
  173. ```
  174. define Package/python-lxml/description
  175. The lxml XML toolkit is a Pythonic binding
  176. for the C libraries libxml2 and libxslt.
  177. endef
  178. define Package/python3-lxml/description
  179. $(call Package/python-lxml/description)
  180. .
  181. (Variant for Python3)
  182. endef
  183. ```
  184. Typically, the description is the same for both, so just mentioning that one is a variant of the other is sufficient.
  185. ### Wrapping things up so that they build
  186. If all the above prerequisites have been met, all that's left is:
  187. ```
  188. $(eval $(call PyPackage,python-lxml))
  189. $(eval $(call BuildPackage,python-lxml))
  190. $(eval $(call Py3Package,python3-lxml))
  191. $(eval $(call BuildPackage,python3-lxml))
  192. ```
  193. The `$(eval $(call PyPackage,python-lxml))` part will instantiate all the default Python build rules so that the final Python package is packaged into an OpenWrt.
  194. And `$(eval $(call BuildPackage,python-lxml))` will bind all the rules generated with `$(eval $(call PyPackage,python-lxml))` into the OpenWrt build system.
  195. These packages will contain byte-codes and binaries (shared libs & other stuff).
  196. If a user wishes to ship source code, adding 2 more lines creates 2 more packages that ship Python source code:
  197. ```
  198. $(eval $(call PyPackage,python-lxml))
  199. $(eval $(call BuildPackage,python-lxml))
  200. $(eval $(call BuildPackage,python-lxml-src))
  201. $(eval $(call Py3Package,python3-lxml))
  202. $(eval $(call BuildPackage,python3-lxml))
  203. $(eval $(call BuildPackage,python3-lxml-src))
  204. ```
  205. The name `*-src` must be the Python package name; so for `python-lxml-src` a equivalent `python-lxml` name must exist.
  206. ### Customizing things
  207. Some packages need custom build rules (because they do).
  208. The default package build and install processes are defined in `python[3]-package.mk`.
  209. #### Building
  210. The default build process calls `setup.py install` inside the directory where the Python source package is extracted (`PKG_BUILD_DIR`). This "installs" the Python package to an intermediate location (`PKG_INSTALL_DIR`) where it is used by the default install process.
  211. There are several Makefile variables that can be used to customize this process (all optional):
  212. * `PYTHON_PKG_SETUP_DIR` / `PYTHON3_PKG_SETUP_DIR`: Path where `setup.py` can be found, relative to the package directory (`PKG_BUILD_DIR`).
  213. Default: empty value (`setup.py` is in the package directory)
  214. * `PYTHON_PKG_SETUP_VARS` / `PYTHON3_PKG_SETUP_VARS`: Additional environment variables to set for the call to `setup.py`. Should be in the form of `VARIABLE1=value VARIABLE2=value ...`.
  215. Default: empty value
  216. * `PYTHON_PKG_SETUP_GLOBAL_ARGS` / `PYTHON3_PKG_SETUP_GLOBAL_ARGS`: Additional command line arguments to pass to `setup.py`, before / in front of the `install` command.
  217. Default: empty value
  218. * `PYTHON_PKG_SETUP_ARGS` / `PYTHON3_PKG_SETUP_ARGS`: Additional command line arguments to pass to `setup.py`, after the `install` command.
  219. Default: `--single-version-externally-managed`
  220. Conceptually, these variables are used in this way (using a Python 2 package as an example):
  221. ```
  222. cd $(PKG_BUILD_DIR)/$(PYTHON_PKG_SETUP_DIR)
  223. $(PYTHON_PKG_SETUP_VARS) python setup.py $(PYTHON_PKG_SETUP_GLOBAL_ARGS) install $(PYTHON_PKG_SETUP_ARGS)
  224. ```
  225. The default build process can be completely overridden by defining custom `PyBuild/Compile` & `Py3Build/Compile` rules in the package Makefile.
  226. #### Installing
  227. The default install process copies some/all of the files from `PKG_INSTALL_DIR`, placed there by the build process, to a location passed to the install rule as the first argument (`$(1)`). The OpenWrt build system will then take those files and create the actual .ipk package archives.
  228. This default process uses 2 build rules:
  229. * `PyPackage/<package>/filespec` & `Py3Package/<package>/filespec` which are Python library files relative to `/usr/lib/pythonX.Y` ; by default this is `/usr/lib/python$(PYTHON[3]_VERSION)/site-packages` (`PYTHON[3]_PKG_DIR`) ; most Python[3] packages generate files that get installed in this sub-folder
  230. * `PyPackage/<package>/install` & `Py3Package/<package>/install` is similar to `Package/<package>/install` ; these allow binary (or other files) to be installed on the target
  231. Both the 2 above rules generate a `Package/<package>/install` build rule, which gets picked up by the build system. Both can be used together (they are not mutually exclusive), and provide a good enough flexibility for specifying Python[3] packages.
  232. The `PyPackage/<package>/filespec` & `Py3Package/<package>/filespec` rules contain one or more lines of the following format (whitespace added for clarity):
  233. ```
  234. <one of: +-=> | <file/directory path> | <file permissions>
  235. ```
  236. The initial character controls the action that will be taken:
  237. * `+`: Install the given path. If the path is a directory, all files and subdirectories inside are installed.
  238. * If file permissions is specified (optional), then the file or directory (and all files and subdirectories) are assigned the given permissions; if omitted, then the file or directory retains its original permissions.
  239. * `-`: Remove the given path. Useful when most of a directory should be installed except for a few files or subdirectories.
  240. * File permissions is not used / ignored in this case.
  241. * `=`: Assign the given file permissions to the given path. File permissions is required in this case.
  242. As mentioned, the default `PyPackage/<package>/filespec` & `Py3Package/<package>/filespec` install `PYTHON[3]_PKG_DIR`:
  243. ```
  244. define PyPackage/python-example/filespec
  245. +|$(PYTHON_PKG_DIR)
  246. endef
  247. ```
  248. If there is an `examples` directory and `test_*.py` files that can be omitted to save space, this can be specified as:
  249. ```
  250. define PyPackage/python-example/filespec
  251. +|$(PYTHON_PKG_DIR)
  252. -|$(PYTHON_PKG_DIR)/examples
  253. -|$(PYTHON_PKG_DIR)/test_*.py
  254. endef
  255. ```
  256. ### Host-side Python packages for build
  257. These can be installed via pip and ideally they should only be installed like this, because it's a bit simpler than running them through the OpenWrt build system. Build variants on the host-side build are more complicated (and nearly impossible to do sanely) in the current OpenWrt build system.
  258. Which is why [for example] if you need python cffi on the host build, it's easier to just add it via:
  259. ```
  260. HOST_PYTHON_PACKAGE_BUILD_DEPENDS:="cffi==$(PKG_VERSION)"
  261. HOST_PYTHON3_PACKAGE_BUILD_DEPENDS:="cffi==$(PKG_VERSION)"
  262. ```
  263. [cffi is one of those packages that needs a host-side package installed for both Python & Python3].
  264. This works reasonably well in the current OpenWrt build system, as binaries get built for this package and get installed in the staging-dir `$(STAGING_DIR)/usr/lib/pythonX.Y/site-packages`.