在构建系统中使用补丁

The build system integrates quilt for easy patch management. This document outlines some common patching tasks like adding a new patch or editing existing ones.

In order to let quilt create patches in the preferred format, a configuration file .quiltrc containing common diff and patch options must be created in the local home directory.

cat << EOF > ~/.quiltrc
QUILT_DIFF_ARGS="--no-timestamps --no-index -p ab --color=auto"
QUILT_REFRESH_ARGS="--no-timestamps --no-index -p ab"
QUILT_SERIES_ARGS="--color=auto"
QUILT_PATCH_OPTS="--unified"
QUILT_DIFF_OPTS="-p"
EDITOR="nano"
EOF

• EDITOR specifies the preferred editor for interactive patch editing
• The other variables control the patch format property like a/, b/ directory names and no timestamps
• FreeBSD does not support the --color=auto option and -pab must be written as -p ab
• -p ab is working with quilt 0.63 on Linux and is documented in man page.

To add a completely new patch to an existing package example start with preparing the source directory:

make package/example/{clean,prepare} V=s QUILT=1

For host-side packages, you may want to detail the make target:

make package/example/host/{clean,prepare} V=s QUILT=1

This unpacks the source tarball and prepares existing patches as quilt patch series (if any). The verbose output will show where the source got extracted.

Change to the prepared source directory.

cd build_dir/target-*/example-*

Note: It can happen that you need to go one level lower as the source is extracted in build_dir/target-*/BUILD_VARIANT/example-*. This happens when multiple build variants of a package are defined in the Makefile.

Apply all existing patches using quilt push.

quilt push -a

Create a new, empty patch file with the quilt new command:

quilt new 010-main_code_fix.patch

• The name should start with a number, followed by a hyphen and a very short description of what is changed
• The chosen number should be higher than any existing patch - use quilt series to see the list of patches
• The patch file name should be short but descriptive

After creating the empty patch, files to edit must be associated with it. The quilt add command can be used for that - once the file got added it can be edited as usual.
A shortcut for both adding a file and open it in an editor is the quilt edit command:

quilt edit src/main.c

• src/main.c gets added to 010-main_code_fix.patch
• The file is opened in the editor specified with EDITOR in .quiltrc

Repeat that for any file that needs to be edited.

After the changes are finished, they can be reviewed with the quilt diff command.

quilt diff

If the diff looks okay, proceed with quilt refresh to update the 010-main_code_fix.patch file with the changes made.

quilt refresh

Change back to the toplevel directory of the buildroot.

cd ../../../

To move the new patch file over to the buildroot, run update on the package:

make package/example/update V=s

Finally rebuild the package to test the changes:

make package/example/{clean,compile} package/index V=s

If problems occur, the patch needs to be edited again to solve the issues. Refer to the section below to learn how to edit existing patches.

Start with preparing the source directory:

make package/example/{clean,prepare} V=s QUILT=1

Change to the prepared source directory.

cd build_dir/target-*/example-*

List the patches available:

quilt series

Advance to the patch that needs to be edited:

quilt push 010-main_code_fix.patch

• When passing a valid patch filename to push, quilt will only apply the series until it reaches the specified patch
• If unsure, use quilt series to see existing patches and quilt top to see the current position
• If the current position is beyound the desired patch, use quilt pop to remove patches in the reverse order
• You can use the “force” push option (e.g. quilt push -f 010-main_code_fix.patch) to interactively apply a broken (i.e. has rejects) patch

Edit the patched files using the quilt edit command, repeat for every file that needs changes.

quilt edit src/main.c

Check which files are to be included in the patch:

quilt files

Review the changes with quilt diff.

quilt diff

If the diff looks okay, proceed with quilt refresh to update the current patch with the changes made.

quilt refresh

Change back to the toplevel diretory of the buildroot.

cd ../../../

To move the updated patch file over to the buildroot, run update on the package:

make package/example/update V=s

Finally rebuild the package to test the changes:

make package/example/{clean,compile} package/index V=s

To prepare the kernel tree, use:

make target/linux/{clean,prepare} V=s QUILT=1

The source tree is in the target-architecture subdirectory (potentially with a subarch):

cd build_dir/target-*/linux-*/linux-*

The process for modifying kernel patches is the same as for packages, only the make targets and directories differ.
For the kernel, an additional subdirectory for patches is used, generic/ contains patches common to all architectures and platform/ contains patches specific to the current target (the latter are found in the target/linux/<arch_name>/patches-* folder in the source).

So, you should first choose where the patch belongs, this is for patches in generic folder:

quilt new generic/010-main_code_fix.patch

patches in platform folder instead should be made with

quilt new platform/010-main_code_fix.patch

And in case you are editing files, it works like for packages but as you saw with the command to make the new patch you need to add the folder name before the name of the patch you are working on.

After you made your changes and saved your patch with quilt refresh, you can go back to top level directory

cd ../../../../

And then you can move the patches you created in these temporary directories back to main source tree:

make target/linux/update package/index V=s

You can also simply copy the patch files over from the local work folder of quilt here (target/linux/<arch_name>/<linux-version>/patches/generic or target/linux/<arch_name>/<linux-version>/patches) to the source folder (target/linux/generic/patches-* or target/linux/<arch_name>/patches-*)

the process might fail with an error that looks like this

bash: line 3: /run/media/alby/data_xeon_btrfs/source_code/my_LEDE_fork/source/staging_dir/host/bin/usign: No such file or directory

and you can safely ignore it, as it just means that you haven't run a compile yet so the “usign” tool used to sign packages does not yet exist. Since you are just working with source, it's irrelevant.

( Patches should be named with the correct prefix, platform/000-abc.patch or generic/000-abc.patch. If not the update may not work correctly.)

Afterwards, if we want to verify whether our patch is applied or not, we can go to the top level directory with

cd ../../../../

and preparing again the linux folder for some modification with

make target/linux/{clean,prepare} V=s QUILT=1

During this process all the applied patched will be shown, ours being among them, preceeded by generic or platform depending on what directory we placed the patch. Another way of retrieving the applied patches is through

quilt series

as explained on the previous sections, after having made make target/linux/{clean,prepare} ...

For example, gcc:

To prepare the tool tree, use:

make toolchain/gcc/{clean,prepare} V=99 QUILT=1

The source tree depends on chosen lib and gcc :

cd build_dir/toolchain-mips_r2_gcc-4.3.3+cs_uClibc-0.9.30.1/gcc-4.3.3

Refreshing the patches is done with:

make toolchain/gcc/update V=99

valid for target/linux/generic and <arch>:

The patches-* subdirectories contain the kernel patches applied for every
OpenWrt target. All patches should be named 'NNN-lowercase_shortname.patch'
and sorted into the following categories:

0xx - upstream backports
1xx - code awaiting upstream merge
2xx - kernel build / config / header patches
3xx - architecture specific patches
4xx - mtd related patches (subsystem and drivers)
5xx - filesystem related patches
6xx - generic network patches
7xx - network / phy driver patches
8xx - other drivers
9xx - uncategorized other patches

ALL patches must be in a way that they are potentially upstreamable, meaning:

- they must contain a proper subject
- they must contain a proper commit message explaining what they change
- they must contain a valid Signed-off-by line

from: PATCHES

When a patched package (or kernel) is updated to a newer version, existing patches might not apply cleanly anymore and patch will report fuzz when applying them. To rebase the whole patch series the refresh make target can be used:

make package/example/refresh V=s

For kernels, use:

make target/linux/refresh V=s

When implementing new changes, it is often required to edit patches multiple times. To speed up the process, it is possible to retain the prepared source tree between edit operations.

1. Initially prepare the source tree as documented above
2. Change to the prepared source directory
3. Advance to the patch needing changes
4. Edit the files and refresh the patch
5. Fully apply the remaining patches using quilt push -a (if any)
6. From the toplevel directory, run make package/example/{compile,install} or make target/linux/{compile,install} for kernels
7. Test the binary
8. If further changes are needed, repeat from step 2.
9. Finally run make package/example/update or make target/linux/update for kernels to copy the changes back to build system

• Official quilt man page
• How To Survive With Many Patches - Introduction to Quilt (PDF) (read online here)
• Applying patches newbie doubt