Hello Kobayashi-san,
I understand your concern.
Lets begin with an equation based definition of DNL:
DNL = [(Vn+1- Vn)/Videal] - 1
Most modern DACs are guaranteed to be monotonic and this is expressed explicitly in the datasheet. This expression is also coherent with the specifications in the electrical characteristics table because the DNL is expressed as: -1 LSB =< DNL =< +1 LSB. A DNL of less than -1 would be required for an increase in code to result in an decrease in the analog output.
The no-missing codes feature is a little different because we need to have DNL expressed as: -1 LSB < DNL < +1 LSB to explicitly state no missing codes. Generally speaking, though, when the maximum DNL specification is bound to +/- 1 LSB the converter is said to have no missing codes because the maximum bound is statistically very unlikely. This bound does not mean that a DNL of +1 LSB or -1 LSB was observed in characterization. This figure is a 3-sigma value; a statistical value arrived at based on the population studied during characterization.
So, what I can say is that the DAC8564 is probably not going to have any missing codes. There is a very very very small chance that eventually you would have a device that has a missing code, but there is no chance that you will ever get a device that is non-monotonic.
DNL bound to +/-1 LSB maximum is a pretty standard industry wide specification a modern DAC. Are you comparing the DAC8564 to another DAC that has a tighter DNL specification and/or that claims to have no-missing codes explicitly in the datasheet?