Sonic Annotator

At last.fm we are using csv output of Vamp summaries. When transforms are defined via n3 files, sonic annotator usually reads them in the order they appear in the RDF file (which is not guaranteed in the first place, I imagine). However, this is not the main problem.

The main problem arises in FeatureExtractionManager.cpp, when the desired outputs are checked for being extracted by the same (and equally parameterized) plugin, and subsequently stored in the map m_plugins: the key in that map are pointers to plugin, and hence have no particular order. Since during output sonic annotator loops over this map, a human-readable plugin order is not guaranteed (sorted by pointer), and in practice it happens that sometimes the outputs of a plugin appear in a different place in the csv-stdout output.

This can be avoided, since the transforms do have a unique order which is "inherited" by the plugins. The plugins can actually be ordered by their first output (or indeed any one of their outputs), and I used this to re-order them prior to the actual output loop. Changes start at line 770 in the attached file.

This solves the problem for me. I admit that this is quite a special problem, but it seems the ordered version is the more elegant solution. Elegant is not how I did it though, so if someone with a better sense of what things should look like in Sonic Annotator could make it more elegant, that would be super!