The following code is reading a big object collection (95G of compressed objects that are uncompressed via the WriteObject streamer) from disk and prints their content as strings.
object.cxx:
std::vector<char> ObjectHandler::GetObject(const std::string& path)
{
TFile *file = new TFile(path.c_str());
// If file was not found or empty
if (file->IsZombie()) {
cout << "The object was not found at " << path << endl;
}
// Get the AliCDBEntry from the root file
AliCDBEntry *entry = (AliCDBEntry*)file->Get("AliCDBEntry");
// Create an outcoming buffer
TBufferFile *buffer = new TBufferFile(TBuffer::kWrite);
// Stream and serialize the AliCDBEntry object to the buffer
buffer->WriteObject((const TObject*)entry);
// Obtain a pointer to the buffer
char *pointer = buffer->Buffer();
// Store the object to the referenced vector
std::vector<char> vector(pointer, pointer + buffer->Length());
// Release the open file
delete file;
delete buffer;
return vector;
}
main.cxx:
ObjectHandler objHandler;
boost::filesystem::path dataPath("/tmp");
boost::filesystem::recursive_directory_iterator endIterator;
if (boost::filesystem::exists(dataPath) && boost::filesystem::is_directory(dataPath)) {
for (static boost::filesystem::recursive_directory_iterator directoryIterator(dataPath); directoryIterator != endIterator;
++directoryIterator) {
if (boost::filesystem::is_regular_file(directoryIterator->status())) {
cout << directoryIterator->path().string() << endl;
std::vector<char> vector = objHandler.GetObject(directoryIterator->path().string());
cout << vector << endl;
}
}
}
1) Is calling by value the correct way to implement this method? Am i doing additional copies that could be avoided if calling by reference?
2) This code is leaking and i am suspecting that either the char *pointer is to blame, or the actual std::vector that is returned by the ObjectHandler::GetObject() method. I've tested the implementation with the following code:
struct sysinfo sys_info;
sysinfo (&sys_info);
cout << "Total: " << sys_info.totalram *(unsigned long long)sys_info.mem_unit / 1024 << " Free: " << sys_info.freeram *(unsigned long long)sys_info.mem_unit/ 1024 << endl;
and the free ram is continuously reduced, until it reaches 0 and the program is killed.
"Memory Leak" is a term that can encompass a few things; depending on who you talk to. One is a new without matching delete. The other, often looked over, is memory that's still referenced and in scope, but just not used or needed.
If you don't use a profiller, then you can't be sure which you have, but since you've a large vector being passed around, and we don't know what you do with it, you could be doing the 2nd and no one would ever see.