RyanJuckett.com

I recently wrote a text parser for Wavefront OBJ files and once I got it all up and running, I was surprised by the performance. I tend to be somewhat performance conscious when writing code so after realizing I had somehow created the slowest OBJ parser known to man, I was perplexed. It was taking 20 seconds to load the Stanford Bunny (4.83MB as an OBJ file with exported normals).

When parsing a 3d mesh from an OBJ file, it is optimal to collapse equal vertices into an indexed list. This is one of the more complicated steps so my suspicion was that something went wrong there. I was using a hash table to do the comparisons so it should have been fast. I disabled that section of code, and timed the load again. It barely affected the result.

Another common pitfall when parsing files is getting stalls from seeking through the file itself. I had already taken that into consideration and just loaded the whole thing into memory for processing. I was out of ideas and decided to profile the load and see what I had done wrong. I learned that almost all of my time was spent in strlen which was a big red flag considering I never even called strlen during my entire load.