Hi Craig,

I've not seen the 3-fold time differences you report - can you post an example that demonstrates this? The example I posted did 5,000,000 assignments with the m. on the left and 5,000,000 without the m. in loops of 100,000.

I've updated the analysis to chech paired results - much better indication:

* analysis

select avg(nWith) as AvgWith, avg(nWithout) as AvgWithout, max(nWith) as MaxWith, max(nWithout) as MaxWithout, ;
	min(nWith) as MinWith, min(nWithout) as MinWithout, Max(nWith - nWithout) as MaxDelta, Min(nWith-nWithout) as MinDelta, ;
	avg(nWith - nWithout) as AvgDelta ;
	from results

An the results show that the difference always exists but, on my machine at least, the differences are pretty small:

Avgwith  Avgwithout      
0.328    0.300
Maxwith  Maxwithout     
0.350    0.311
Minwith  Minwithout     
0.320    0.290
Maxdelta Mindelta  Avgdelta        
0.059    0.009     0.028

The maximum difference in 100,000 assignments is 6/100ths of a second - certainly not the 300% difference that you get. I'm not sure there's much more blood to squeeze out of this particular optimisation stone :)

Cheers,

Andrew

---

>
>Andrew,
>
>I've tested this in every version since 2.5 and continually get the same results. Lines with the m. take 3x longer to execute. This can show outside loops. If you have several assignment statements, you'll get a definate speed degradation.