JLS string concatenation clarification

[Vitaly Davidovich]

Much more efficient after jit compiles it as well? At least C2 has optos
specifically covering scenarios where a StringBuilder is used for
intermediate string manipulation.  I'd think something as simple as this
case would be handled well, although I haven't checked.

Sent from my phone
On Jan 21, 2014 8:10 PM, "Martin Buchholz" <martinrb at google.com> wrote:

> So, the context is that some of us have noticed that compiling
> x + y
> where x and y are both Strings, to
> (x == null ? "null" : x).concat(y)
> is much more efficient than creating a StringBuilder, and the vast
> majority of our users would thank us for implementing that optimization in
> javac.  The fact that Strings even have an object identity is essentially a
> bug.
>
>
>
> On Tue, Jan 21, 2014 at 4:45 PM, Alex Buckley <alex.buckley at oracle.com>wrote:
>
>> No, we can't. There are three levels to consider:
>>
>> - "Real" constant expressions like 3+"". They can avoid creating a new
>> String object per https://bugs.openjdk.java.net/browse/JDK-4836025.
>>
>> - "Semi" constant expressions like s+"". Concatenating the empty string
>> literal "" with a String expression is mentioned in
>> https://bugs.openjdk.java.net/browse/JDK-4036535. The expectation (per
>> JLS1) that a new String object is always created in this case is misplaced
>> (per the JDK's actual behavior). But, the JLS never clarified this.
>>
>> - "Not" constant expressions: s+empty. Not detectable in general. Always
>> creates a new String object.
>>
>> Alex
>>
>>
>> On 1/21/2014 10:41 AM, Martin Buchholz wrote:
>>
>>> Can we agree that when
>>> String empty = "";
>>> then
>>> s + empty
>>> should be permitted to be the same object as s.
>>>
>>>
>>> On Tue, Jan 21, 2014 at 9:59 AM, Louis Wasserman <lowasser at google.com
>>> <mailto:lowasser at google.com>> wrote:
>>>
>>>     JLS 15.18.1 states:
>>>
>>>         The |String| object is newly created (§12.5
>>>         <http://docs.oracle.com/javase/specs/jls/se7/html/jls-
>>> 12.html#jls-12.5>)
>>>
>>>         unless the expression is a compile-time constant expression
>>>         (§15.28
>>>         <http://docs.oracle.com/javase/specs/jls/se7/html/jls-
>>> 15.html#jls-15.28>).
>>>
>>>
>>>     JLS 12.5 states:
>>>
>>>         Execution of a string concatenation operator (§15.18.1
>>>         <http://docs.oracle.com/javase/specs/jls/se7/html/jls-
>>> 15.html#jls-15.18.1>)
>>>
>>>         that is not part of a constant expression (§15.28
>>>         <http://docs.oracle.com/javase/specs/jls/se7/html/jls-
>>> 15.html#jls-15.28>)
>>>
>>>         sometimes creates a new |String| object to represent the result.
>>>
>>>
>>>     It's not clear to me, from reading this, whether string
>>>     concatenation with + /always/ creates a new String object or
>>>     /sometimes /creates a new String object.  15.18.1 seems to imply
>>>
>>>     always, but 12.5 says "sometimes."
>>>
>>>     My understanding is that String.concat used to be used for string
>>>     concatenation, and String.concat currently returns its receiver if
>>>     its argument is the empty string.  This seems to be evidence for the
>>>     "sometimes" interpretation.
>>>
>>>     --
>>>     Louis Wasserman
>>>
>>>
>>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/compiler-dev/attachments/20140121/b0a6d6f4/attachment-0001.html