diff query.cpp @ 405:ef4792df8f93 api-inversion

invert audioDB::insert / audiodb_insert(). Start off by removing audioDB::insertDatum, and essentially reusing it as audiodb_insert. We now ignore the fact that the command-line parsing code has "helpfully" opened a std::ifstream for the times file and an fd for the power file, and simply go ahead and do our own dirty work. We can delete audioDB::insertDatum entirely, but unfortunately we can't delete audioDB::insertPowerData and audioDB::insertTimestamps, because the index and query code respectively use them. Instead, move the two methods closer to their single uses. audiodb_insert() is perhaps not as short and simple as it might have been hoped given the existence of audiodb_insert_datum(); some of that is C and its terribly way of making you pay every time you use dynamic memory; some of it is the fact that the three different files (feature, times, power) each requires slightly different treatment. Hey ho. We can implement audiodb_batchinsert() in terms of audiodb_insert(); the function is pleasingly small. We can't quite use it for audioDB::batchinsert yet, as we have to deal with the O2_FLAG_LARGE_ADB case (which codepath is untested in libtests/). This means that we can delete whole swathes of hideous code from audioDB.cpp, including not just the versions of audiodb_insert() and audiodb_batchinsert() but also an entire audioDB constructor. Yay. (audioDB::unitNormAndInsertL2 has also died a deserved death).
author mas01cr
date Fri, 05 Dec 2008 22:32:49 +0000
parents 2d5c3f8e8c22
children a7d61291fbda
line wrap: on
line diff
--- a/query.cpp	Fri Dec 05 22:32:43 2008 +0000
+++ b/query.cpp	Fri Dec 05 22:32:49 2008 +0000
@@ -242,6 +242,42 @@
   CHECKED_READ(trkfid, *data_buffer_p, trackTable[track] * sizeof(double) * dbH->dim);
 }
 
+void audioDB::insertTimeStamps(unsigned numVectors, std::ifstream *timesFile, double *timesdata) {
+  assert(usingTimes);
+
+  unsigned numtimes = 0;
+
+  if(!timesFile->is_open()) {
+    error("problem opening times file on timestamped database", timesFileName);
+  }
+
+  double timepoint, next;
+  *timesFile >> timepoint;
+  if (timesFile->eof()) {
+    error("no entries in times file", timesFileName);
+  }
+  numtimes++;
+  do {
+    *timesFile >> next;
+    if (timesFile->eof()) {
+      break;
+    }
+    numtimes++;
+    timesdata[0] = timepoint;
+    timepoint = (timesdata[1] = next);
+    timesdata += 2;
+  } while (numtimes < numVectors + 1);
+
+  if (numtimes < numVectors + 1) {
+    error("too few timepoints in times file", timesFileName);
+  }
+
+  *timesFile >> next;
+  if (!timesFile->eof()) {
+    error("too many timepoints in times file", timesFileName);
+  }
+}
+
 // These names deserve some unpicking.  The names starting with a "q"
 // are pointers to the query, norm and power vectors; the names
 // starting with "v" are things that will end up pointing to the