Since GLEP 67 was approved, bug assignment became easier. However, there were still many metadata.xml files which made this suboptimal. Today, I have fixed most of them and I would like to provide this short guide on how to write good metadata.xml files.
The bug assignment procedure
To understand the points that I am going to make, let’s take a look at how bug assignment happens these days. Assuming a typical case of bug related to a specific package (or multiple packages), the procedure for assigning the bug involves, for each package:
- reading all <maintainer/> elements from the package’s metadata.xml file, in order;
- filtering the maintainers based on restrict="" attributes (if any);
- filtering and reordering the maintainers based on <description/>s;
- assigning the bug to the first maintainer left, and CC-ing the remaining ones.
I think the procedure is quite clear. Since we no longer have <herd/> elements with special meaning applied to them, the assignment is mostly influenced by maintainer occurrence order. Restrictions can be used to limit maintenance to specific versions of a package, and descriptions to apply special rules and conditions.
Now, for semi-automatic bug assignment, only the first or the first two of the above steps can be clearly automated. Applying restrictions correctly requires understanding whether the bug can be correctly isolated to a specific version range, as some bugs (e.g. invalid metadata) may require being fixed in multiple versions of the package. Descriptions, in turn, are written for humans and require a human to interpret them.
What belongs in a good description
Now, many of existing metadata.xml files had either useless or even problematic maintainer descriptions. This is a problem since it increases time needed for bug assignment, and makes automation harder. Common examples of bad maintainer descriptions include:
Assign bugs to him;
CC him on bugs— this is either redundant or contradictory. Ensure that maintainers are listed in correct order, and bugs will be assigned correctly. Those descriptions only force a human to read them and possibly change the automatically determined order.
proxied maintainer— this is some information but it does not change anything. If the maintainer comes first, he’s obviously the primary one. If the maintainer has non-Gentoo e-mail and there are proxies listed, he’s obviously proxied. And even if we did not know that, does it change anything? Again, we are forced to read information we do not need.
Good maintainer descriptions include:
Upstream; CC on bugs concerning upstream,
Only CC on bugs that involve USE=”d3d9″— useful information that influences bug assignment;
Feel free to fix/update,
All modifications to this package must be approved by the wxwidgets herd.— important information for other developers.
So, before adding another description, please answer two questions: will the information benefit anyone? Can’t it be expressed in machine-readable form?
Since a lot of the affected packages are maintained by proxied maintainers, I’d like to explicitly point out how proxy-maintained packages are to be described. This overlaps with the current Proxy maintainers team policy.
For proxy-maintained packages, the maintainers should be listed in the following order:
- actual package maintainers, in appropriate order — including developers maintaining or co-maintaining the package, proxied maintainers and Gentoo projects;
- developer proxies, preferably described as such — i.e. developers who do not actively maintain the package but only proxy for the maintainers;
- Proxy-maintainers project — serving as the generic fallback proxy.
I would like to put more emphasis on the key point here — the maintainers should be listed in an order making it clearly possible to distinguish packages that are maintained only by a proxied maintainer (with developers acting as proxies) from packages that are maintained by Gentoo developers and co-maintained by a proxied maintainer.
Third-party repositories (overlays)
As a last point, I would like to point out the special case of unofficial Gentoo repositories. Unlike the core repositories, metadata.xml files can not be fully trusted there. The reason for this is quite simple — many users copy (fork) packages from Gentoo along with metadata.xml files. If we were to trust those files — we would be assigning overlay bugs to Gentoo developers maintaining the original package!
For this reason, all bugs on unofficial repository packages are assigned to the repository owners.