BTW, did you capture the fact that BOOT_ADAFLAGS are no longer passed with the new configure; make bootstrap ? FWIW, I can't really do more testing or give more feedback on the new approach until this is fixed.
I just verified that it works with --disable-bootstrap; make bootstrap. As for bootstraping libada, as you said we already build the object needed by the compiler. Your argument that "compiling more unneeded stuff will help find bugs" is not really an argument, since it would equally apply to basically all the tests under the "tests" directory. Also you should keep in mind that building more stuff as part of the bootstrap is also painful on non x86* machines which are typically not that fast. And requiring to have two builds (one with --disable-bootstrap and one without) is certainly not realistic, at least not in the world I live in. Arno