Ada bootstrap failure due to loop optimization and builtin_stack_alloc

[Richard Kenner]

    The debugging information is perfectly capable of reflecting this, at
    least for DWARF2.  There may be some hoops required when outputing the
    debug information to preserve the original type and at the necessary
    DW_OP_indirect.

Indeed the latter is the issue.  The debugging information could certainly
represent the type of the data, but it won't be the type that the user
expects and the question is whether we can encode that it must remove
one level of indirection before presenting the data to the programmer.

I was planning to send a reminder on this issue today anyway, but your
message served that purpose: we should decide how to handle this.