[foreign] RFR: do not infer layout from Java signatures

[Maurizio Cimadamore]

Pushed with your change. Thanks for testing!

Maurizio


On 18/05/18 13:57, Sundararajan Athijegannathan wrote:
> In test/jdk/java/nicl/System/UnixSystem.java test :)
>
> -Sundar
>
> On 18/05/18, 6:25 PM, Sundararajan Athijegannathan wrote:
>> +1.
>>
>> Except that you need to fix Mac specific layout
>>
>>     @NativeHeader
>>     static interface MacOSXSystem {
>>         @C(file="dummy", line=1, column=1, USR="c:@F at stat")
>> @NativeType(layout="(p:cp:[iSSQIIi[ll][ll][ll][ll]qqiIIi2q])i", 
>> ctype="dummy", size=1)
>>         @CallingConvention(value=1)
>>         public abstract int stat$INODE64(Pointer<Byte> path, 
>> Pointer<stat> buf);
>>
>> -Sundar
>>
>> On 18/05/18, 5:00 PM, Maurizio Cimadamore wrote:
>>> Hi,
>>> as part of the cleanup I pushed yesterday, I realized there are 
>>> quite few places where the binder infers layouts from the Java 
>>> signatures rather than sticking to what the annotations say. This 
>>> patch fixes such bad usages.
>>>
>>> http://cr.openjdk.java.net/~mcimadamore/panama/dont-infer-layout-v2/src/java.base/share/classes/jdk/internal/nicl/Util.java.udiff.html 
>>>
>>>
>>> The only place where we need to do something like this is with 
>>> variadic calls; so Util still has a function that goes from Class to 
>>> Layout, but its scope is much more limited, and the resulting layout 
>>> is also approximate (I've added comments in that direction), since 
>>> varargs are always passed in register-aligned slots, regardless of 
>>> the underlying real layout - so we don't need much 'inference' there.
>>>
>>> I fixed a test which was missing layout annotation in a Java callback.
>>>
>>> Maurizio
>>>
>>>