Same seed for srand yields different results

First the background info... I've got a program that has several instances of an Agent class, and I wanted each Agent object to have a different seed for srand so that each class produces different random numbers. So in Agent.h I have:

Code:

---

class Agent {
  ...
private:
  ...
  static int seed;  // Random number generator seed
};

---

And in Agent.cp, outside of any method, I initialize seed at file scope:

Code:

---

int Agent::seed = time(NULL);

---

Finally, within the constructor of the Agent class, I increment seed so that each Agent object has a unique seed:

Code:

---

Agent::Agent() {
  ++seed;
  srand (seed);
  ...
}

---

The program then uses rand() to generate random numbers, which are then used to generate lots of different results, such as strings and other numbers. There's one big outer loop to my program that loops thru an array of Agent objects, calling various methods for each object. Every so many times through the loop I have it write out the results at that point of the program.

Now here's my problem... I'm trying to reproduce runs of my program to test other parameters, but I want the random numbers that are generated to be the same so that I know that the other parameters are truly what changed the results. So instead of setting seed to time(NULL) like above, I set it to a constant at file scope. When I run the program twice without changing anything, the initial results, which are generated by these random numbers, are the same. But after the program's been executing a while, the later results change. (The program's much too large to just print out all the random numbers, so I just examine the results (strings, etc), that are created from the random numbers.)

Since nothing changed between the two runs and the seed is the same, I would expect the same results. Do I just not understand how srand works? Or does C++ not always execute code in the exact same order everytime? For example, I loop through the Agent objects calling methods for each Agent, and there's lots of loops and function calls in the Agent methods -- maybe C++ is executing the commands in a different order to optimize it? That would explain my varying results.

I hope this makes sense... this program is a few thousand LOC and I didn't want to bore you with all the details, but if you need more info on my program please ask! Thank you!!

srand should only ever be called once, not everytime your constructor is called. For your testing purposes, you can just set the seed to the same value and call srand ONCE with that value. When you are ready to go beyond testing, change the seed value to be based off of the time function call. This probably means you don't need a seed member for the class anymore. You could alter your code so that your constructor will call srand(time(NULL)); if it is the first instance of the class and not during any subsequent instances.

Thanks for the help, hk, I will try that... but I'm still curious as to why I'm not getting the same results. But maybe it's one of those things that you just can't tell without tracing through lots of code!

Regardless of the multiple srand calls, you should be getting the same results.

A couple things come to mind as possibilities. Are there any variables that are used prior to being initialized? and are there more calls somewhere to time() or getenv(), or anything else which tries to ask the system for some piece of information?

Cheers