xref: /openbmc/docs/meta-layer-guidelines.md (revision 0ee8da09)
1# OpenBMC meta layer guidelines
2
3While most of these could be called "rules", in specific scenarios might be
4actively against the intended goals.  This is why each guideline has a very
5large "Why" section, to ensure that the intent is being met, and that if
6exceptions to the rules exist, then can be understood and managed by the
7project.  In general, if there's a question, and the community agrees, these
8guidelines can be overridden on a case by case basis.
9
10## Meta layers should not patch projects that exist within the openBMC tree
11
12**Why?**
13
14In general, keeping the codebase building in the long term is difficult.
15Opening the possibility that patches exist that repo maintainers aren't aware
16of makes it much more likely that a single machine breaks, or we have behavior
17differences between two repos.
18
19Also, in general, the maintainer is there to ensure that the greater community,
20features, and codebase are prioritized over any one patch, and generally does so
21in code review.  If patches are checked into meta layers, generally the
22maintainer is not a reviewer, thereby defeating most of the purpose of the role
23of the maintainer.
24
25**What should I do instead?**
26
27Discuss with the project maintainers and the community about whether or not the
28feature you're building needs to be configurable, or if it can be applied to all
29projects.  If it can be applied to all, simply check it into the master branch
30through a gerrit review, and follow the processes outlined for the repository.
31If it needs to be per-project or per-machine configurable, check it in under a
32compile time option, at the suggestion of the maintainer, and add a
33PACKAGECONFIG entry that can be set to enable it.
34
35## Meta layers should not patch Yocto recipes and projects
36
37**Why?**
38
39Yocto itself is an open source project that accepts contributions.  The more
40changes that OpenBMC stacks against Yocto recipes, the more unmaintainable it
41becomes, and the longer it takes to rebase to new Yocto versions.  In general,
42the Yocto community is as responsive (sometimes much faster) than the OpenBMC
43community in regards to pull requests.
44
45**What should I do instead?**
46
47Submit any changes needed to the Yocto upstream repositories, using their
48process.  If the Yocto process has gone several weeks without responses,
49cherry-pick the commit into the OpenBMC tree, with a pointer to the review in
50the commit message, and documentation.  The patch commit message should follow
51the Yocto requirements for patch comments, including a statement of the patch
52upstream status.
53[yocto patch requirements](https://wiki.yoctoproject.org/wiki/Best_Known_Methods_%28BKMs%29_for_Package_Updating#Patch_Comments)
54
55## Meta layers should avoid using EXTRA_OEMAKE and EXTRA_OEMESON
56
57**Why?**
58
59There are some OpenBMC projects that are utilized outside of OpenBMC.  As such,
60there are configuration items that are not intended to be used in OpenBMC, or
61configuration items that would pose a security risk.  Also, as options change
62and are deprecated, the project needs a single place to update the available
63config items and dependencies.
64
65In addition, subprojects might change their build tooling, for example from
66autotools to meson, in pursuit of other goals.  Having tool-specific
67configurations makes that change far more difficult to do.
68
69**What should I do instead?**
70
71In the root recipe, add a PACKAGECONFIG entry for the feature in question, then
72use that to enable said feature in your meta layer.
73
74## Meta layers should not have recipes that point to proprietary licensed code
75
76**Why?**
77
78OpenBMC is an open source project, and is intended to be built from source, with
79appropriate distribution licenses such that it can be reused.  Pointing to
80commercially licensed repositories actively opposes that goal.
81
82**What should I do instead?**
83
84Find an equivalent open source project that meets the needs, or request that the
85project owner relicenses their project.
86
87## Meta layer recipes should only point to well maintained open source projects
88
89**Why?**
90
91Without this guideline, a loophole is present that allows OpenBMC developers to
92bypass code review by pointing the upstream recipe to a public repository that
93they control, but which OpenBMC has no input on the content of.  This splits the
94discussion forums in unproductive ways, and prevents all the other good
95processes within OpenBMC like bug tracking and continuous integration from
96having an effect.
97
98**What should I do instead?**
99
100The advice tends to be on a case by case basis, but if the code is only intended
101for use on OpenBMC, then push a design doc, and push the code to openbmc gerrit
102under the openbmc/openbmc repository where it can be reviewed, along with an
103OWNERS file, signaling your willingness to maintain this project.  Then, once
104the community has looked through your design, a repo will be created for code to
105be pushed to.  If you're pulling in code from a dead project, inquire to the
106community through the mailing list or discord whether or not the OpenBMC
107community would be willing to adopt support and maintenance of said project.
108
109## Meta layers should not point to OpenBMC specific repositories outside of
110   https://github.com/openbmc
111
112**What defines an OpenBMC specific repository?**
113
114Generally an OpenBMC specific repository is something that does any of the
115following:
116* Relies on or implements on OpenBMC defined Dbus interfaces.
117* Is not designed with configurability to be used outside of OpenBMC
118* Relies on OpenBMC specific layouts, configuration information to accomplish
119  its primary function.
120
121**Why?**
122
123OpenBMC as a codebase needs to remain buildable in the long term.  Companies,
124maintainers, and contributors come and go over the course of the project.
125Keeping all long-term-supported code in the OpenBMC github ensures that the
126project will remain useful.  In addition, having significant portions of
127OpenBMC-specific code outside of github limits the ability of the project to
128do refactoring that might be necessary to keep up with the latest best
129practices, Yocto updates, and package revisions, as there isn't always a clear
130path to getting patches accepted when repositories are not within the OpenBMC
131github project
132
133**What should I do instead?**
134
135Discuss the code you'd like to write with the community using the mailing list
136and/or discord.  If the design would require a new repository, request one using
137the OpenBMC technical oversight committee process.
138
139## Don't use SRCREV="${AUTOREV}" in a recipe
140
141**Why?**
142
143Repository branches can change at any time.  Pointing to an autorev revision
144increases the likelihood that builds break, and makes builds far less
145reproducible.
146
147In addition, having an accounting of exactly what is in your build prevents
148errors when a repo is quietly updated while working, and suddenly changes
149significantly.
150
151**What should I do instead?**
152
153Point SRCREV to a specific commit of the repository, and increase the revision
154either via the autobump script in CI, which can be requested on the mailing
155list, or manually as new revisions exist.
156