position question, cylinder vs circle, what do you do?

Cylinder is always better but some time the geometry is not allowing it. The length of the cylinder has to be bigger than the DIA of the cylinder or at least somehow close to have good results. Fit to datums makes sense when you have material modifier in datums and it means that pcdmis will shift/rotate the axis to the allowed bonus in order to improve the results. Having this on, will often result in having 0 as measurements. I generally never use it as the engineering team wants to always see the deviation in axis between features and datums. Axis worse means that you cover the worst case of results because a cylinder has more than 1 point to compare the deviation from datums as opposed to a circle (that only has the center). Once again it is considered better to use it as "worst"... Having said that I don't use worst and go with middle or whatever is called as worst point is changing between measurements and does not give me repeatability. 

If you are not reporting the worst end of the axis, you could risk accepting bad parts because position applies over the entire length of the feature.  That is why you should be measuring cylinders rather than circles - your measurements should always cover as much of the surface as possible and extend over as much of the length as possible.  Circles are really only truly valid for thin, sheet-metal parts or when measuring using a 2D camera system.  However, a lot of CMM programmers choose to use circles because they save time.  It really comes down to how much risk you are willing to take sacrificing accuracy in order to achieve faster throughput.

As for "fit to datums", it tells PC-DMIS to use the datums you have selected in your datum reference frame and to optimise the result.  This is required in order to comply with the standard (ASME or ISO).  By over-riding those settings you are deviating from that requirement and again, run a risk, this time, of possibly failing good parts because you are not allowing the result to fully optimise.

 constadin makes a valid point about repeatability and the need to have usable information to feed back to production.  However, rather than simply "turn everything off", it is often better to have two sets of results - one with all the correct settings active, material modifiers included etc. that can be used for a pass/fail assessment and then another that is used purely for providing adjustment information.  This could be an XactMeasure dimension with the modifiers removed and fit to datums turned off or a legacy dimension relative to the active alignment.

It really comes down to what you are looking for here as long as you understand what each option is doing. As I said in my case, we validate our programs in terms of repeatability and accuracy before program release and a random factor as which cylinder's end will be reported can really render the validation impossible. Having said that we understand that the worst case is not taken into account although we have set pretty strict tolerances to compensate for this.

Same goes for FIT TO DATUMs. Even if it is the correct setting to have in terms of standard guidelines, a good part in our case will report 0 deviation each time due to datums shift, effectively rendering our validation procedure obsolete.

It really comes down to what you are looking for here as long as you understand what each option is doing. As I said in my case, we validate our programs in terms of repeatability and accuracy before program release and a random factor as which cylinder's end will be reported can really render the validation impossible. Having said that we understand that the worst case is not taken into account although we have set pretty strict tolerances to compensate for this.

Same goes for FIT TO DATUMs. Even if it is the correct setting to have in terms of standard guidelines, a good part in our case will report 0 deviation each time due to datums shift, effectively rendering our validation procedure obsolete.