I am solving a 3 fields coupled problem with Firedrake. I would like to use a multiplicative block-preconditioner but PETSc complains about a local size not being correct.

[0] Arguments are incompatible
[0] Local size 9 not compatible with block size 2

I have noticed that the use of a MatAij solves the problem (see comment in line 132) but I am afraid this is just a side effect.

Am I missing something? 

Thanks in advance,

Nicolas