A single repository with an example file for every mimetype.
Because examples are useful and file extensions are unreliable.
Mimetype associations often rely on "magic" values at the beginning of files. Testing to see if these are working correctly requires an actual valid example file, not just an arbitrary file with the correct file extension.
For example, this is the way that single-page and multipage DJVU are distinguished.
Even when mimetype associations are working correctly, applications themselves sometimes have trouble with a certain file format. Having a public-domain example to reference is helpful when submitting a bug report.
For example, older versions of Inkscape would sometimes show the wrong color for imported Adobe Illustrator files.
In order of priority:
- Public domain or Creative Commons licensing.
- Correctly represents the mimetype.
- Stable URIs (permalinks are best).
- Relatively small file size (to save disk space and bandwidth).
The IANA lists about 1800 registered mimetypes, some of which are deprecated or duplicates. There are many more which are unofficial.
It's a big job, but not impossible. See the coverage report for details.
Suppose you want an example of a VRML file.
Wikipedia discusses VRML and links to examples, but finding an actual VRML file via web search is non-trivial.
A Google filetype search for .vrml files? Not helpful.
https://google.com/search?q=filetype%3Avrml
Oops, that should be .wrl files.
https://google.com/search?q=filetype%3Awrl
But most of the results still aren't relevant.
Even if you find a good link, you either have to keep track of the file or bookmark the link, which could give a 404 the next time you open it.
With a mimetype menagerie, it's as simple as a link to a directory:
And finding files from scratch is a simple search with a file manager or on the commandline like this:
$ cd media-types/ $ find . -iname '*VRML*' ./model/vrml $ ls model/vrml/ HelloWorld.wrl
Some mimetypes may not have any public domain examples, or may be patent-encumbered. These can be linked to, but will have to be omitted from the actual git repository.
The fuzzing program American Fuzzy Lop (afl) has a number of testcases, such as archives and image formats:
https://github.com/loverszhaokai/AFL/tree/master/testcases
The FFMPEG project has a compendium of video files, though it's unclear how many are public domain.
Alex Schiller's file format commons is similar, though is not organized by mimetype.
The test documents for Apache's Tika parsers are not organized by mimetype, but are impressively comprehensive.
Ubuntu includes the example-content package.