From @salortiz

In latest nom failing inside BUILD can't be detected anymore​:

<sortiz> star-m​: class Foo { submethod BUILD { fail "noway" } } ; with Foo.new { say "good" } else { .fail }
<camelia> star-m 2016.04​: OUTPUT«noway␤ in submethod BUILD at <tmp> line 1␤ in block <unit> at <tmp> line 1␤␤Actually thrown at​:␤ in block <unit> at <tmp> line 1␤␤»
<sortiz> m​: class Foo { submethod BUILD { fail "noway" } } ; with Foo.new { say "good" } else { .fail }
<camelia> rakudo-moar 5d4949​: OUTPUT«good␤»

Bisectable points to commit d5e09d9 "Streamline Mu.BUILDALL"​:

<sortiz> bisect​: class Foo { submethod BUILD { fail "noway" } } ; Foo.new ~~ Failure;
<bisectable> sortiz​: exit code on a “good” revision is 1 (which is bad), bisecting with inverted logic
<bisectable> sortiz​: (2016-06-11) rakudo/rakudo@d5e09d9

In my real use case, the failure is detected and reported much later, at DESTROY time​:

WARNING​: unhandled Failure detected in DESTROY​:
Env open 'NoSuChDiR'​: No such file or directory
  in submethod BUILD at /home/sog/work/MyGIT/p6-LMDB/lib/LMDB.pm6 (LMDB) line 240

From @salortiz

Solved by lizmat++ in rakudo/rakudo@b351d61006
with tests in Raku/roast@f2105b8c40

@salortiz - Status changed from 'new' to 'resolved'