Preliminary RFR (L): 8001107: @Stable annotation for constant folding of lazily evaluated variables

[Vladimir Ivanov]

FYI, I've updated [1] with proposed changes.

Best regards,
Vladimir Ivanov

[1] http://cr.openjdk.java.net/~vlivanov/8001107/webrev.04

On 9/4/13 1:20 AM, Vladimir Kozlov wrote:
> On 9/3/13 2:04 PM, Vladimir Ivanov wrote:
>> John,
>>
>> I'm fine with both options (add an assert or use
>> element_value(index).as_char() in product).
>>
>> Vladimir K., what do you prefer?
>
> assert
>
> Vladimir K
>
>>
>> Best regards,
>> Vladimir Ivanov
>>
>> On 9/3/13 10:40 AM, John Rose wrote:
>>> On Aug 29, 2013, at 4:57 PM, Vladimir Kozlov <vladimir.kozlov at oracle.com
>>> <mailto:vladimir.kozlov at oracle.com>> wrote:
>>>
>>>>>> ciTypeArray.cpp: why we need these changes?
>>>>> I consider this as a cleanup. element_value(index).as_char() performs
>>>>> additional checks and it looks better (IMO, of course :-) ) than
>>>>> get_typeArrayOop()->char_at(index).
>>>>
>>>> But element_value() is much more expensive! I don't like this change.
>>>
>>> The old char_at function works fine, but I think it needs an assert that
>>> the array being referenced is truly a char[] array.  The more expensive
>>> element_value function performs this check, so it is safer.  It is also
>>> helpful to prove (at least via an assert) that element_value is a true
>>> extension of char_at.
>>>
>>> If we don't like making char_at slightly more expensive (it doesn't make
>>> much difference!) then I suggest putting the call to element_value into
>>> the #ifdef ASSERT, as follows:
>>>
>>>      assert(index >= 0 && index < length(), "out of range");
>>> !   jchar c = get_typeArrayOop()->char_at(index);
>>> ! #ifdef ASSERT
>>> !   jchar d = element_value(index).as_char();
>>> !   assert(c == d, "");
>>> ! #endif //ASSERT
>>> !   return c;
>>>    }
>>>
>>> — John