Many check_debug regressions

[Jonathan Wakely]

On 13 January 2014 10:31, Paolo Carlini wrote:
> Great. Note however, that the above backward logic means that, at some
> point, somebody, committed a debug-mode change which caused many
> regressions. And we went ahead, reported no regression in debug-mode for
> additional unrelated changes... Worrisome...

I *think* the tests that are failing are new ones I added when adding
normal-mode allocator changes, they've never passed in debug mode.

So it was just that my nomal-mode changes didn't also cover debug mode
and I didn't test debug mode, it's not that we made debug mode changes
and didn't test debug mode.