[illumos-Developer] [dtrace-discuss] dynamic size argument to tracemem()

[Bryan Cantrill]

On Tue, Jul 12, 2011 at 4:54 PM, Adam Leventhal <ahl at delphix.com> wrote:
> Hey Bryan,
>
> Did you consider having the behavior be that if the third argument was
> greater than the second (unsigned) that it would generate an error?

Yeah, I did consider that and we had a conversation about that here.
The consensus was that implicitly clamping it was the cleaner behavior
-- that the closest analogue here was the clamping behavior of strsize
(where you don't see an error, just a silent clamping) and that you
don't want to see the console flooded with error messages in the case
where it's consistently larger than the static size.  What are your
thoughts?

> Regarding tst.dynsize.d, could the output fail to match if the tick
> probe switched CPUs? Is that sufficiently unlikely, or would it be
> worth using a profile probe with a cpu predicate (caching the first
> value we see)?

Yeah, that's true -- but unlikely, as you say.  FWIW, that's true in
many places in the test suite.  In fact there was a port of DTrace out
there that changes the CPU on which the tick probe fires on every
firing, and the test suite fell over in a zillion places.  So if this
is a problem that we want to tackle, I think we might want to do it
somewhat holistically...

>  { "tracemem", DT_IDENT_ACTFUNC, 0, DT_ACT_TRACEMEM,
>        DT_ATTR_STABCMN, DT_VERS_1_0,
> -       &dt_idops_func, "void(@, size_t)" },
> +       &dt_idops_func, "void(@, size_t, ...)" },
>
> I think that should be: "void(@, size_t, [size_t])" like copyinstr() or index().

I could have done in that way, but I'm actually doing the check
manually because I want to propagate my own error tag and test on it.

Thanks for taking a look at this -- and I only have two more wads to
come!  (They're both tiny.)

        - Bryan