Prefer list over vector chromsweep's cache. Huge performance increases

when the database (-b) file has many more intervals than the query file.
Thanks to Neil Kindlon (Quinlan Lab)."

src/utils/chromsweep/chromsweep.cpp

@@ -32,8 +32,7 @@ ChromSweep::ChromSweep(BedFile *query, BedFile *db,
 , _reciprocal(reciprocal)
 {
     _hits.reserve(100000);
-    _cache.reserve(100000);
-    
+
     _query->Open();
     if (printHeader) _query->PrintHeader();
     _db->Open();
@@ -48,8 +47,7 @@ ChromSweep::ChromSweep(BedFile *query, BedFile *db,
 ChromSweep::ChromSweep(string &queryFile, string &dbFile) 
 {
     _hits.reserve(100000);
-    _cache.reserve(100000);
-    
+
     _query = new BedFile(queryFile);
     _db = new BedFile(dbFile);
     
@@ -69,7 +67,7 @@ ChromSweep::~ChromSweep(void) {
 
 
 void ChromSweep::ScanCache() {
-    vector<BED>::iterator c = _cache.begin();
+    list<BED>::iterator c = _cache.begin();
     while (c != _cache.end())
     {
 

src/utils/chromsweep/chromsweep.h

@@ -64,8 +64,12 @@ private:
     float _overlapFraction;
     // do we care about strandedness.
     bool _sameStrand, _diffStrand, _reciprocal;
-    // a cache of still active features from the database file
-    vector<BED> _cache;
+    // a cache of still active features from the database file\
+    // 2012-Oct-29: prefer LIST over VECTOR as, when _cache is large,
+    // the overhead of deleting from a vector is huge.  Deleting from
+    // the front of a LIST is cheap. Thanks to Neil Kindlon (Quinlan lab)
+    // for the very important fix.
+    list<BED> _cache;
     // the set of hits in the database for the current query
     vector<BED> _hits;
     // a queue from which we retrieve overlap results.  used by Next()