TRACE :ENCAPSULATE NIL broken on darwin

[Nikodemus Siivola]

Commented-out test-case in debug.impure.lisp:

;;; This is not a WITH-TEST :FAILS-ON PPC DARWIN since there are
;;; suspicions that the breakpoint trace might corrupt the whole image
;;; on that platform.
#-(and (or ppc x86 x86-64) darwin)
(with-test (:name (trace :encapsulate nil)
            :fails-on '(or :ppc :sparc :mips))
  (let ((out (with-output-to-string (*trace-output*)
               (trace trace-this :encapsulate nil)
               (assert (eq 'ok (trace-this)))
               (untrace))))
    (assert (search "TRACE-THIS" out))
    (assert (search "returned OK" out))))

[Nikodemus Siivola (nikodemus)]

As long as the suspicion remains, attaching big runtime warning to doing a breakpoint trace on Darwin seems like a good idea. (Once that is done the importance can be downgraded, imo.)

[Nikodemus Siivola (nikodemus)]

Breakpoints are currently broken on x86[-64]/darwin as we use an illegal instruction instead of INT3 to trap, but the runtime installs INT3 as a trap.

This arrangement is due to historical flakiness: it used to be possible to lose a SIGTRAP on darwin. I'm not sure if this is still the case. If not, we can just go back to INT3, if not we should teach the runtime to use illegal instruction traps instead.

[Alastair Bridgewater (alastair-bridgewater)]

As of 1.0.36.9, the runtime has been taught to use illegal instruction traps when appropriate. I don't have a mac, so I can't proceed further.

[Nikodemus Siivola (nikodemus)]

Not quite there on Darwin yet, unfortunately.

http://thread.gmane.org/gmane.lisp.steel-bank.cvs/13959/focus=14571

[Alastair Bridgewater (alastair-bridgewater)]

Knocking this down to Medium importance because TRACE :ENCAPSULATE NIL is broken on all non-x86oid targets, and because there has been no activity on this for nearly three months, and no activity for about eight months before that.

[Stas Boukarev (stassats)]

In 2c62dd2b0a1c85b8178820307ef424245c835675