Quote:Another issue is whether you are timing the code correctly, or using too small a dataset such that external factors can sway the measurement significantly. I don't suppose you can show us the source code for these tests, and the actual results you get on your machine?As I said, it was a loose measurement, taken with breakpoints and counting seconds.
I think I may have found the problem, though. It seems that the allocation itself was fast enough, but the surrounding code that reads through the file is not. My mistake for not catching that straight off.
The code is part of an object loader (which loads a .obj and associated .mtl if there is one) into a set of data for rendering. The C++ code reads the file once to determine how large it need be, the memory is allocated, and then the file is read again to load the data into it. The Python code reads the file once. This may be where the speed difference is.
The C++ sections where the file is read look something like:
string line;ifstream obj_file(filename.c_str());while (!obj_file.eof()) { getline(obj_file,line); //Process line, either counting it, or adding its data, depending on //what it's doing to the file at this time.}file = open(os.path.join(*filename),"r")for line in file.readlines(): #process line, and add it to the proper listBecause the C++ code is opening the file several times and reading presumably one line at a time, it's going to take longer.
New question: basically, the problem is that the file contains content of an unknown size. What is the fastest way to create a data structure that contains it, in RAM--including the time spent reading from the file?
Thanks,
-G
