AudioDB Meeting, Goldsmiths Dec 2011¶

MC, CR, CC, LF

Current state

• Library
  • Mainline code in trunk works OK
  • Library API is "quite good"
  • Error handling is a very weak point

• Indexing
  • LSH should be in a separate library, but it has no clear interface yet
  • There are contortions between e.g. query.c and query-indexed.c
  • There is no introspection, e.g. no way to say "is there an index that satisfies this query?" -- only to say "satisfy this query" and fail

• Tools
  • Currently we have a library, and a command-line binary which does not use the library API but instead uses some of the same .o files directly
  • Some functions are not properly library-ized, for example the command-line tool is the only way to create an index at the moment
  • Old fftExtract binary is still in use in some places with a new latent feature separation front-end

• Bindings
  • Lisp, Python: Quite good
  • pd: broken-ish
  • Some additional interfaces e.g. SOAP that predate the stable API and are pretty horrible now
  • Bindings are not built automatically

• Documentation
  • Essentially none
  • Public header has some rather cryptic stuff

• Licence
  • Currently all GPL

Actions

• Philosophy: Project should be a set of modules -- not so much a database, as libraries with tools that make use of them
• Discard broken / ancient interfaces e.g. SOAP
• Refactor API so that the (existing) "world object" concept is carried through in order to retain error condition state
• Document API
• Rebuild command-line utility in either C C++ or Python using the API only, and augment API if there are things it needs that don't exist there yet
• Document command-line utility
• (Keep bindings in same repo, don't spin them out)
• Try to build module API for LSH so that alternative implementations could be introduced
• Licence: LGPL for library, BSD for bindings