comparison audioDB-internals.h @ 465:1030664df98c api-inversion

No more audioDB::index_allocate and audioDB::index_init_query

No more SERVER_LSH_INDEX_SINGLETON, either; instead each adb_t contains
a single cache of the last used in-core index.  At the moment, this
cache is unused by the server (and the previous cache code has been
replaced by a comment), but I think that this way everyone can be
allowed to benefit without anyone having to explicitly manage indexes
themselves.

I'm not going to say how long I wandered in a maze of valgrind before
giving up and keeping the hacky workaround for loading the lsh tables
[see the FIXME comment in audiodb_index_init_query()]; let's just say
that it was long enough to find the extra bonus crashy close(lshfid) in
audioDB::index_index_db.

Also, delete the abstraction-inverting LSH stuff from query.cpp where we
are making our reporters; the fix for that, which is presumably when
creating small indexes for large datasets, is to implement
space-efficient reporters.  (The accumulator code, which is my second
attempt, is more space-efficient than the reporters; inspiration may
wish to be drawn...)

   adb_header_t *header;
   std::vector<std::string> *keys;
   std::map<std::string,uint32_t> *keymap;
   std::vector<uint32_t> *track_lengths;
   std::vector<off_t> *track_offsets;
+  LSH *cached_lsh;
 };
 
 typedef struct {
   bool operator() (const adb_result_t &r1, const adb_result_t &r2) {
     return strcmp(r1.key, r2.key) < 0;