xref: /OK3568_Linux_fs/buildroot/docs/manual/patch-policy.txt (revision 4882a59341e53eb6f0b4789bf948001014eff981)

// -*- mode:doc; -*-
// vim: set syntax=asciidoc:

[[patch-policy]]

== Patching a package

While integrating a new package or updating an existing one, it may be
necessary to patch the source of the software to get it cross-built within
Buildroot.

Buildroot offers an infrastructure to automatically handle this during
the builds. It supports three ways of applying patch sets: downloaded patches,
patches supplied within buildroot and patches located in a user-defined
global patch directory.

=== Providing patches

==== Downloaded

If it is necessary to apply a patch that is available for download, then add it
to the +<packagename>_PATCH+ variable. If an entry contains +://+,
then Buildroot will assume it is a full URL and download the patch
from this location. Otherwise, Buildroot will assume that the patch should be
downloaded from +<packagename>_SITE+. It can be a single patch,
or a tarball containing a patch series.

Like for all downloads, a hash should be added to the +<packagename>.hash+
file.

This method is typically used for packages from Debian.

==== Within Buildroot

Most patches are provided within Buildroot, in the package
directory; these typically aim to fix cross-compilation, libc support,
or other such issues.

These patch files should be named +<number>-<description>.patch+.

.Notes
- The patch files coming with Buildroot should not contain any package version
  reference in their filename.
- The field +<number>+ in the patch file name refers to the 'apply order',
  and shall start at 1; It is preferred to pad the number with zeros up to 4
  digits, like 'git-format-patch' does. E.g.: +0001-foobar-the-buz.patch+
- Previously, it was mandatory for patches to be prefixed with the name of
  the package, like +<package>-<number>-<description>.patch+, but that is
  no longer the case. Existing packages will be fixed as time passes. 'Do
  not prefix patches with the package name.'
- Previously, a +series+ file, as used by +quilt+, could also be added in
  the package directory. In that case, the +series+ file defines the patch
  application order. This is deprecated, and will be removed in the future.
  'Do not use a series file.'


==== Global patch directory

The +BR2_GLOBAL_PATCH_DIR+ configuration file option can be
used to specify a space separated list of one or more directories
containing global package patches. See xref:customize-patches[] for
details.

[[patch-apply-order]]
=== How patches are applied

. Run the +<packagename>_PRE_PATCH_HOOKS+ commands if defined;

. Cleanup the build directory, removing any existing +*.rej+ files;

. If +<packagename>_PATCH+ is defined, then patches from these
  tarballs are applied;

. If there are some +*.patch+ files in the package's Buildroot
  directory or in a package subdirectory named +<packageversion>+,
  then:
+
* If a +series+ file exists in the package directory, then patches are
  applied according to the +series+ file;
+
* Otherwise, patch files matching +*.patch+ are applied in alphabetical
  order.
  So, to ensure they are applied in the right order, it is highly
  recommended to name the patch files like this:
  +<number>-<description>.patch+, where +<number>+ refers to the
  'apply order'.

. If +BR2_GLOBAL_PATCH_DIR+ is defined, the directories will be
  enumerated in the order they are specified. The patches are applied
  as described in the previous step.

. Run the +<packagename>_POST_PATCH_HOOKS+ commands if defined.

If something goes wrong in the steps _3_ or _4_, then the build fails.

=== Format and licensing of the package patches

Patches are released under the same license as the software they apply
to (see xref:legal-info-buildroot[]).

A message explaining what the patch does, and why it is needed, should
be added in the header commentary of the patch.

You should add a +Signed-off-by+ statement in the header of the each
patch to help with keeping track of the changes and to certify that the
patch is released under the same license as the software that is modified.

If the software is under version control, it is recommended to use the
upstream SCM software to generate the patch set.

Otherwise, concatenate the header with the output of the
+diff -purN package-version.orig/ package-version/+ command.

If you update an existing patch (e.g. when bumping the package version),
make sure the existing From header and Signed-off-by tags are not
removed, but do update the rest of the patch comment when appropriate.

At the end, the patch should look like:

---------------
configure.ac: add C++ support test

Signed-off-by: John Doe <john.doe@noname.org>

--- configure.ac.orig
+++ configure.ac
@@ -40,2 +40,12 @@

AC_PROG_MAKE_SET
+
+AC_CACHE_CHECK([whether the C++ compiler works],
+               [rw_cv_prog_cxx_works],
+               [AC_LANG_PUSH([C++])
+                AC_LINK_IFELSE([AC_LANG_PROGRAM([], [])],
+                               [rw_cv_prog_cxx_works=yes],
+                               [rw_cv_prog_cxx_works=no])
+                AC_LANG_POP([C++])])
+
+AM_CONDITIONAL([CXX_WORKS], [test "x$rw_cv_prog_cxx_works" = "xyes"])
---------------

=== Integrating patches found on the Web

When integrating a patch of which you are not the author, you have to
add a few things in the header of the patch itself.

Depending on whether the patch has been obtained from the project
repository itself, or from somewhere on the web, add one of the
following tags:

---------------
Backported from: <some commit id>
---------------

or

---------------
Fetch from: <some url>
---------------

It is also sensible to add a few words about any changes to the patch
that may have been necessary.