The accuracy you need to control a servo depends a lot on the servo in my
experience. A $10 servo does not need that much accuracy, it will be
overwhelmed by only 100 steps of input, each resolving to 1.8 degrees on
the control yoke, and this is a happy case, because the potentiometer in
the servo should be about 0.5% accurate to allow this to happen, and it
isn't.

If you can live with an accuracy of a few tens of usec out of the 1msec
variable delay you will still have 50 or more steps and you will be able
to write an ISR based on a timer (TMR0) that will implement a nice and
constant and stable output PWM, and you can even write it in C if you use
a faster crystal (10MHz or more).

The code itself is very simple. In the ISR you would have something like:

  if(Pwmflag) { // while we do PWM
    if(!Pwmreg.Pwmbit) { // the O/P bit is 0, start
      Pwmcounter = Pwmset; // load the pwm counter from setpoint register
      Pwmreg.Pwmbit = 1; // turn on output
    } else { // the O/P is 1, we count period
      if(!--Pwmcounter) // decrement by 1
        Pwmflag = Pwmreg.Pwmbit = 0; // turn off output and flag 'done'
    }
  }

Now you arrange the ISR to be called at a constant say 20usec and in your
main code you set Pwmflag and then wait for it to be reset. The ISR does
everything else. A little more work is required to generate repeating
pulses of PWM as required by the servo (hint: use another counter in the
ISR that sets Pwmflag every X timer interrupts).

There are other ways to generate PWM for servos, with resolution down to
1T cycle, but using assembly, and no interrupts.

hope this helps,

Peter

--
http://www.piclist.com hint: The PICList is archived three different
ways.  See http://www.piclist.com/#archives for details.