On 24/04/2013 9:05 PM, Rickard Bäckman wrote:
David,
On Apr 24, 2013, at 12:11 PM, David Holmes wrote:
On 24/04/2013 7:09 PM, Rickard Bäckman wrote:
Nils,
no it doesn't matter. Rather intended. By initializing it to NULL we forced
implementors to use a pointer that would have to be initialized at some point.
Now it can be a class / struct
that is instead initialized by a default constructor.
So that addressed my question on the missing setter. But doesn't this also mean
that you are now prohibiting it from being a simple pointer-type as there is no
way to set it? Isn't maintaining the setter more flexible as it can be used in
either case (direct assignment or copy constructor). Though lack of
initialization in the current code still looks wrong.
Yes it makes it harder for the type to be a simple pointer, though that could
be worked around by putting the pointer inside a struct. Not a great solution
perhaps.
The other solution would be to have a setter, the question is what to
initialize it with? Should we add another hook?
I don't see a truly satisfactory solution, but I think Parfait will flag
that you are returning an uninitialized variable here.
So yes perhaps you need to define TRACE_DATA_INIT ?
David
/R
David
/R
On Apr 24, 2013, at 10:45 AM, Nils Loodin wrote:
Does it matter that the pointer gets initialized to NULL before, but not now?
There isn't any null checks anywhere that depends on that?
Regards,
Nils
On 04/24/2013 09:51 AM, Rickard Bäckman wrote:
Hi all,
can I have a couple of reviews for this small change. The short story is that
the current way the thread-local _trace_buffer is somewhat inflexible.
By changing the type of the getter this structure gets more flexible for
different implementations. I also think that the name is misused. Just naming it
to _trace_data is more generic and less implementation-specific.
The webrev: http://cr.openjdk.java.net/~rbackman/8013117/
Thanks
/R
