Thanks so much !!! you solved a big part of my problem !!
But now, all this is even stranger for me 
The shape=1 parameter has no effect for me (anyway if not provided, I guess that 1 is the default value(?)). So I guess this has neither effect for you…
The real effect comes from the np.asarray([ ]) inside the lopg call ! It’s not needed for the ‘A’ version (with a list of one-dimensional distribution components), but it is mandatory on the ‘B’ version (with a single multidimensional distribution list as components).
How did you find this formulation/solution ?
It so weird because , for me, A & B version only differs in the way they provide mixture distribution components, not it’s input, so why do we have to shape the input differently ?
Now, I’d like to be able to call the logp function with an array, not a single point (to avoid loop! again…)…
How can I do this ? (all my attemps failed 
)