Generated parameter names for -parameters (JEP118)

[Eric McCorkle]

The problem seems to be javac not generating the second entry.  Adding a
main to the class file and attempting to run it confirms this (hotspot
will throw a ClassFormatError if there is an incorrect number of
parameters).

Jesper, do you want to file the bug about this, since you found it?

On 05/29/13 18:28, Alex Buckley wrote:
> I agree with your assessment below. Either javac is not correctly
> generating the second parameters array entry, or javap is not correctly
> decoding it.
> 
> Note that when an entry appears in the parameters array for a parameter
> which was _not_ explicitly declared in source, you can "rely" on its
> ACC_MANDATED or ACC_SYNTHETIC flags - every Java compiler is compelled
> by the JLS to set them correctly. However, you cannot rely on the name
> or ACC_FINAL flag for such a parameter, for they are not specified and
> hence are compiler-specific.
> 
> Alex
> 
> On 5/29/2013 3:09 PM, Jesper Steen Møller wrote:
>> Jonathan Gibbons wrote:
>>> I think you should be more explicit in what you think is wrong,
>>> instead of just posting a javap dump and expecting others to
>>> read through it.
>>
>> I'm sorry -- you are quite right. I should have explained the problem
>> properly.
>>
>> The problem is with the MethodParameters attribute for the anonymous
>> inner class' constructor.
>>
>>> With respect to your question,
>>> Why do you think an entry is missing from the table?
>>
>> The constructor code has three arguments: The this reference itself,
>> the enclosing class' this reference, and the value of the captured
>> final variable of the enclosing scope.
>> The descriptor shows the two "non-this" arguments:
>> (LParameterNames;Ljava/lang/String;)V
>>
>> As I understand, the MethodParameters attribute should describe the
>> latter two parameters here, matching the method descriptor, but it
>> doesn't quite:
>> The hex dump of the MethodParameters table is this:
>> MethodParameters: length = 0x5
>> 02 00 0A 80 10
>> ^^
>> Length byte
>>
>>  From what I've read in '4.7.22 The MethodParameters Attribute', the
>> length should be 9, since there's one byte for the length, and four
>> bytes for each parameter.
>>
>> u1 parameters_count = 0x02
>> parameters [2] = {
>> parameter {
>>    u2 name_index == 0x000A //  'this$0' in the constant pool
>>    u2 access_flags == 0x8010 // meaning MANDATED+FINAL
>>   },
>> parameter {
>>      // MISSING!
>>    }
>> }
>>
>> Also, I'm also suspecting that the entry is in error since javap
>> didn't properly decode it and show the actual parameter names, as it
>> can for other classes.
>>
>>> What entry do you think is missing from the table?
>>
>> The second one, containing the name of the captured local variable
>> from the enclosing context. I'm sure this doesn't really have a name,
>> since no code ac
>> I'm expecting it to be synthetic and final 0x1010, but I could be
>> mistaken.
>>
>> FWIW, this is from build 1.8.0-ea-b90.
>>
>> TIA, Jesper
>>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: eric_mccorkle.vcf
Type: text/x-vcard
Size: 303 bytes
Desc: not available
Url : http://mail.openjdk.java.net/pipermail/compiler-dev/attachments/20130530/c96bdea7/eric_mccorkle.vcf