GC problems in tests

[Stas Boukarev]

I added (defvar *gc* (sb-thread:make-thread (lambda () (loop (sleep 0.0001) (gc :full t))))) to pure-runner, impure-runner. One of the interesting results — deprecation.impure fails with some random garbage:

        "0 fell through ETYPECASE expression.
         Wanted one of (SIMPLE-STRING SB-FORMAT::FORMAT-DIRECTIVE)."

        "SB-X86-64-ASM::MOV fell through ETYPECASE expression.
         Wanted one of (SIMPLE-STRING SB-FORMAT::FORMAT-DIRECTIVE)."

        "#<unknown immediate object, lowtag=#b1001, widetag=#x39 {501015B300002039}> fell through ETYPECASE expression.
         Wanted one of (SIMPLE-STRING SB-FORMAT::FORMAT-DIRECTIVE)."

arm64 fails with similar results even more often.

[Stas Boukarev (stassats)]

stand-alone for arm64:

(sb-thread:make-thread (lambda () (loop (sleep 0.0001) (gc :full t))))

(loop repeat 8
      do
      (sb-thread:make-thread
       (lambda ()
         (let* ((*standard-output* (make-broadcast-stream))
                (*error-output* *standard-output*))
           (loop (compile nil `(lambda () (defun a () (cons)))))))))

* fatal error encountered in SBCL pid 90454 pthread 0x16b4f3000:
no transport function for object 0x7005ea0b0f (widetag 0xc3)

[Stas Boukarev (stassats)]

Seems that sb-xref-for-internals is necessary.

[Stas Boukarev (stassats)]

I fixed the darwin-arm64 issue above in aafddc813f3c9618ffc9eee9dab64b19a20e8372, which is probably separate from the overall deprecation.impure thing.