Skip to content

jwboyer/fedkernel

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

41 Commits
 
 
 
 
 
 
 
 

Repository files navigation

These scripts are used to generate an exploded source git tree from the
Fedora kernel pkg-git tree.  Specifically, they use the build information
found in koji for each successful build to generate a source tree from
the SRPM and tag it with the corresponding package Name-Version-Release
tag.  The exploded tree is hosted on kernel.org at:

https://git.kernel.org/cgit/linux/kernel/git/jwboyer/fedora.git/

Below are some details on how it works, some of the assumptions made, and
a big list of things that need to be improved.

## Quick Start

The script depends on fedpkg, fedora-messaging, python3-koji, and
python3-GitPython. To run the message consumer, run:

$ fedora-messaging --conf fedkernel.toml.example consume --callback-file kernel-git.py:callback

The script supports providing a Koji build ID to manually create the tree:

$ ./kernel-git.py 1349425

## How it works.

Fedora uses git for its package SCM.  This means that for every build in
koji, we have an immutable corresponding SCM commit.  The scripts
leverage this fact to take the build information from koji and construct
the exploded tree from exactly the set of sources that were sent to be
built.  Essentially the scripts:

1. determine the pkg-git sha1sum for the build
2. perform a checkout of that exact commit
3. prep the pkg-git tree for that commit
4. determine the upstream base release, tag or git revision
5. resets an upstream linux.git tree the determined sha1sum
6. extract the patches Fedora applies
7. applies said patches to the linux.git tree
8. tags the linux.git tree with the corresponding package build NVR

The script gets the build information either from a user specified build
NVR on the command line, or by listening to the fedmsg bus for creation
automatically*.  This sounds pretty simple, and in operation it is but
there are a number of gotchas and assumptions the scripts rely upon.

## Assumptions

- There are pkg-git and linux.git trees already checked out in the
configured locations.
- The build being used has patches applied in the spec file using git am
- The branches for the various releases are already created in the
linux.git trees.

The first item means you have essentially throw-away trees for both
pkg-git and linux.git somewhere.  They require you to have all branches
created.  The linux.git tree requires you to have both Linus' tree and
the upstream stable tree configured as remotes.

The second item above is somewhat the crutch these scripts really rely
upon.  Fedora kernels in the F22 and older branches use an ApplyPatch
shell function to apply the patches on top of the base.  This works fine
and is generally very fast when creating the tree in the %prep section.
However, because that function is using raw patch(1), the generated
source has no commit history and the patches can be in any format that
patch(1) accepts.  That has the drawback that the patches themselves can
be raw diffs with no changelogs, etc.

Starting with Fedora 23, the kernel spec file uses git-am to apply
patches on top of the base tarball/patches.  This requires that all
patches be acceptable to git-am, which means the patches themselves have
quite a bit more context and information in them.  While it does slow
down the %prep section, the additional context provided for each change
makes it a good tradeoff.  It also means we can easily get the exact
patches back out of the prepped sources.  That is key to automating the
creation of an exploded source tree.

(It is certainly possible to have the contextual information in the patch
 with patch(1), but in practice the human element crept in and laziness
 or indifference meant we carried a number of sub-par diffs as patches.
 Using git-am provides the enforcement and then we get the benefits
 later.)

## TODOs

There are quite a few things that are lacking in this code.  They are, in
no specific order:

- Error checking
- Poor fedmsg handling
- Lack of ability to create new branches automatically (relying on assumptions)
- Lack of graceful failure and cleanup
- No kernel configuration information in the generated tree
- No pushing of exploded tree

Arguably, the last one in this list is probably smart for now.  We
wouldn't want the exloded tree pushed to a public git server if it was
created incorrectly, so manual pushing is somewhat of a safeguard.
However it also means if someone forgets to push the tree, it won't be
updated.

### Developer info

Hosted at: https://github.com/jwboyer/fedkernel

The code for the koji connection is borrowed from koji itself.  The spec
parsing code is inspired by the rpmdev-bumpspec tool.  The remainder of
the code is primarily written by:

- Josh Boyer <[email protected]>

Changes and/or pull requests can be sent to either the primary author or
the Fedora kernel list: [email protected]

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages