Re: [Sbcl-devel] [Sbcl-commits] master: tests/hash.impure.lisp: Improve non-thread behavior for weak-hash-table tests

[Stas B. <sta...@gm...>]

I used it to fix sb-introspect/allocation-information.6.

On Mon, Jan 15, 2018 at 9:32 PM Alastair Bridgewater <
ala...@gm...> wrote:

> Good call, and more reliable as well. I'd somehow gotten the impression
> that we'd deleted the lisp side of that at some point. Commit forthcoming.
>
> On Mon, Jan 15, 2018 at 11:13 AM, Stas Boukarev <sta...@gm...>
> wrote:
>
>> Would SB-SYS:SCRUB-CONTROL-STACK do the same job?
>>
>> On Mon, Jan 15, 2018 at 6:11 PM Alastair Bridgewater via Sbcl-commits <
>> sbc...@li...> wrote:
>>
>>> The branch "master" has been updated in SBCL:
>>>        via  c2f1dc83e5e9d72a0ab06238d562d1e0194933ef (commit)
>>>       from  5fecdf33a2dfc9c19f498ef351691c5f290d6b73 (commit)
>>>
>>> - Log -----------------------------------------------------------------
>>> commit c2f1dc83e5e9d72a0ab06238d562d1e0194933ef
>>> Author: Alastair Bridgewater <ala...@gm...>
>>> Date:   Mon Jan 15 08:48:23 2018 -0500
>>>
>>>     tests/hash.impure.lisp: Improve non-thread behavior for
>>> weak-hash-table tests
>>>
>>>       * Actually arrange to erase parts of the control stack, so that
>>>     even "precise-GC" systems can't accidentally hold a root due to
>>>     poor stack slot management.
>>>
>>>       * Also fix a clear thinko in a NOTINLINE declaimation.
>>>
>>>       * This fixes (:HASH-TABLE :WEAKNESS :REMOVAL) on -sb-thread PPC.
>>> ---
>>>  tests/hash.impure.lisp | 26 +++++++++++++++++++++-----
>>>  1 file changed, 21 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/tests/hash.impure.lisp b/tests/hash.impure.lisp
>>> index 2cc9847..61767a2 100644
>>> --- a/tests/hash.impure.lisp
>>> +++ b/tests/hash.impure.lisp
>>> @@ -267,7 +267,7 @@
>>>
>>>  (defvar *cons-here*)
>>>
>>> -(declaim (notinline args))
>>> +(declaim (notinline take))
>>>  (defun take (&rest args)
>>>    (declare (ignore args)))
>>>
>>> @@ -275,10 +275,26 @@
>>>    "Execute BODY and try to reduce the chance of leaking a conservative
>>> root."
>>>    #-sb-thread
>>>    `(multiple-value-prog1
>>> -       (progn ,@body)
>>> -     (loop repeat 20000 do (setq *cons-here* (cons nil nil)))
>>> -     ;; KLUDGE: Clean the argument passing regs.
>>> -     (apply #'take (loop repeat 36 collect #'cons)))
>>> +       (let ((*s* (make-array 25)))
>>> +         (declare (special *s*)
>>> +                  (dynamic-extent *s*))
>>> +         ;; We can't "just" arrange to run the allocation on a separate
>>> +         ;; thread stack, then deallocate the thread before doing
>>> +         ;; whatever, so start by giving ourselves some headroom via a
>>> +         ;; DX allocation around the actual allocation...
>>> +         (multiple-value-prog1
>>> +             (progn ,@body)
>>> +           (loop repeat 20000 do (setq *cons-here* (cons nil nil)))
>>> +           ;; KLUDGE: Clean the argument passing regs.
>>> +           (apply #'take (loop repeat 36 collect #'cons))))
>>> +     (let ((*s* (make-array 300)))
>>> +       (declare (special *s*)
>>> +                (dynamic-extent *s*))
>>> +       ;; ... and use a larger DX allocation to try to clear the stack
>>> +       ;; (thus, roots, conservative or otherwise) where we executed
>>> +       ;; the BODY (this definitely clears stack space, the
>>> +       ;; uncertainty is if it covers all of the space used by BODY).
>>> +       (take *s*)))
>>>    #+sb-thread
>>>    (let ((values (gensym))
>>>          (sem (gensym)))
>>>
>>> -----------------------------------------------------------------------
>>>
>>>
>>> hooks/post-receive
>>> --
>>> SBCL
>>>
>>>
>>> ------------------------------------------------------------------------------
>>> Check out the vibrant tech community on one of the world's most
>>> engaging tech sites, Slashdot.org! http://sdm.link/slashdot
>>> _______________________________________________
>>> Sbcl-commits mailing list
>>> Sbc...@li...
>>> https://lists.sourceforge.net/lists/listinfo/sbcl-commits
>>>
>>
>>
>> ------------------------------------------------------------------------------
>> Check out the vibrant tech community on one of the world's most
>> engaging tech sites, Slashdot.org! http://sdm.link/slashdot
>> _______________________________________________
>>
> Sbcl-devel mailing list
>> Sbc...@li...
>> https://lists.sourceforge.net/lists/listinfo/sbcl-devel
>>
>>
>