Upstream MEtadata GAthered with YAml (UMEGAYA)
Introduction
I am starting an experimental effort to collect meta-information about upstream in a file called debian/upstream in the source packages maintained by the DebianMed project. Since these source packages are stored in a subversion repository on Alioth, the information can be updated without uploading the source packages to the Debian archive.
A draft collector system is being implemented on upstream-metadata.debian.net, and its source is available on git.debian.org. The plan is to use it to prepare tables that can be fed to the UltimateDebianDatabase.
Proof of principle (in progress)
First attempt through a YAML intermediate
To make the DebianMed web sentinels use the UDD, fed from the debian/upstream via upstream-metadata.debian.net, to display bibliographic information about which academic article to cite when using our packages. This is currently done by collecting the information in the central file used to create the med-* metapackages.
The bibliographic data is refreshed daily at http://upstream-metadata.debian.net/for_UDD/biblio.yaml via a local cron job. As specified in config-org.yaml, it is retreived by the script fetch_bibref.sh and loaded in the UDD as triples (package, key, value) with the bibref_gatherer.
Second attempt through a pool of files
The Umegaya instance running at http://upstream-metadata.debian.net is collecting and organising debian/upstream and debian/copyright files as pools. Currently they are pushed manually in the QA team's Subversion repository's directory packages-metadata. An UDD importer is in development.
Syntax
The debian/upstream file is in YAML format. In its simplest form, it looks much like the paragraph format used in Debian control files. Nevertheless, there may be some times unexpected behaviours, for instance field contents that have a colon inside have to be quoted in some cases. In doubt, there are validaters available, like this Online YAML Parser. With the libyaml-libyaml-perl package installed, the following command can also validate YAML files:
perl -MYAML::XS -e '$/=""; Load(<STDIN>)' < upstream
Be careful not to use the plain Perl YAML module as it accepts files with invalid syntax (661700).
Only a subset of YAML is used: sequences are only expected to contain scalars and mappings are only expected to contain a scalar or a mapping, but with only one level of imbrication.
In addition, two conventions that are not part of the YAML format are used:
- Field names are case-insensitive.
- Nested mappings are shortcuts for longer field names composed of both mapping field names separated by a dash. The following two examples are equivalent:
Foo: Bar: baz
Foo-Bar: baz
Fields
In alphabetic order. Let's try to use the same vocabulary as in DOAP as much as possible. Fields that are the same as in DOAP are followed by an asterisk.
- Archive
- When the upstream work is part of a large archive, like CPAN.
- Bug-Database
- A URL to the list of known bugs for the project.
- Bug-Submit
- A URL that is the place where new bug reports should be sent.
- Contact
- Which person, mailing list, forum,… to send messages in the first place.
- DOI
- Deprecated; see Reference-DOI instead.
- Donation
- An URL to a donation form (or instructions).
- FAQ
- An URL to the online FAQ.
- Gallery
- An URL to a gallery of pictures made with the program (not screenshots).
- Name *
- Upstream name of the packaged work.
- Homepage *
- The packaged work's homepage.
- PMID
- Deprecated; see Reference-PMID instead.
- PubMed
- Deprecated; see Reference-PMID instead.
- Reference
The following fields are used to document the academic publication describing the packaged work, and are usually pasted from ?BibTex references. There is a big issue to solve: what if the Debian package contains more than one work, published in different articles.?
- Reference-Author
- Author list.
- Reference-DOI
- This is the digital object identifier of the academic publication describing the packaged work.
- Reference-Eprint
- Hyperlink to the PDF file of the article.
- Reference-Journal
- Abbreviated journal name.
- Reference-Number
- Issue number.
- Reference-Pages
- Article page number(s).
- Reference-PMID
Same as the DOI, but with the ID number in the PubMed database.
- Reference-Title
- Article title.
- Reference-URL
- Hyperlink to the electronic version of the article.
- Reference-Volume
- Journal volume.
- Reference-Year
- Year of publication
- References
- An URL to a upstream page containing more references.
- Registration
- An URL to a registration form (or instructions).
- Repository
- URL to a repository containing the upstream sources.
- Repository-Browse
- An URL to browse the repository containing the upstream sources.
- Screenshots
URL to an upstream page containing screenshots (not screenshots.debian.net).
- Watch
Currently it contains the main line of debian/watch. It is therefore assumed to be in format version 3. For surveying multiple locations, it could contain a YAML sequence.
- Webservice
- URL to an web page where the packaged program can also be used.
Reserved fields
The following fields are used internally and must not be present in debian/upstream.
- YAML-ALL
- Used to dump the loaded record.
- YAML-URL
- Used to override the repository's URL provided by debcheckout.
- YAML-REFRESH-DATE
Used to deduce how long umegaya will ignore calls to refresh (to avoid hammering Alioth).
TODO: ignore them safely.
Discussion
Let's discuss here, on a mailing list (debian-med or debian-qa), or a discussion page, if available.
The data is not really Debian-specific, lets put it outside Debian and use the ?PackageMap to map between Debian package names and the data:
http://lists.debian.org/debian-mentors/2009/11/msg00450.html
To do: formalise the above using ?http://search.cpan.org/~ddumont/Config-Model/lib/Config/Model/Backend/Yaml.pm, and generate docs as explained in http://ddumont.wordpress.com/2011/04/08/configuration-doc-generation-with-configmodel/ .
* In addition to ?DOAP, other Semantic Web ontologies/namespaces/schemas should be reused in order to not reinvent the wheel, and enable such metadata to participate to the ?Semantic Web (see also Open Linked Data matters). As such, SPDX would be an interesting standard to link to, as well as ADMS.F/OSS, for packages description, IMHO. Syntactically, any form of RDF would be interesting to explicitely convey the prefixes in the field names... and I'm not sure it can be done in ?YAML -- OlivierBerger
Notes
debian/upstream-metadata.yaml was formerly used instead of debian/upstream.