[foreign] RFR: refresh layout grammar

[Maurizio Cimadamore]

On 23/05/18 20:26, Henry Jen wrote:
> On May 23, 2018, at 11:40 AM, Maurizio Cimadamore <maurizio.cimadamore at oracle.com> wrote:
>>
>>
>> On 23/05/18 19:10, Henry Jen wrote:
>>>> On May 23, 2018, at 9:46 AM, Maurizio Cimadamore <maurizio.cimadamore at oracle.com> wrote:
>>>>
>>>> Hi,
>>>> I've put together a change which tweaks the grammar of layout descriptions used in annotations to correspond to what's described here [1].
>>>>
>>>> The big changes are in DescriptorParser, which now has to understand the new grammar, and also in TestDescriptorGrammar, which has to combinatorially generate sentences of the new grammar.
>>>>
>>>> All the other tedious changes are test changes, required to go from the old to the new grammar; I've adhered to the following rules:
>>>>
>>>> 'b' -> ‘u8'
>>> b is a bit, so should it be u1?
>> uhm - booleans don't exist in C, so typically a byte will be allocated holding a 1 or a 0. You can think of it as u8[u1(value)u7(padding)] if you want, but the results are the same?
> b is for bit-field, so that definitely a u1.
My bad then

I meant B -> 'u8' :-)

Maurizio
> B is for boolean, that is _Bool in C, the x64 ABI state is 1 byte in size and alignment, so u8 is fine.
>
> Cheers,
> Henry
>
>>> o is octet, which is the byte and should be u8.
>> ok, I haven't seen octets around in existing tests
>>>> 'c' -> ‘i8'
>>> char is either signed or unsigned depends on platform. Since we are targeting x64 for now, unsigned is correct.
>> ok
>>
>> Maurizio
>>>> 's' -> 'i16'
>>>> 'i' -> 'i32'
>>>> 'l' -> 'i64'
>>>> 'q' -> 'i64'
>>>>
>>>> Also note that in the new grammar capital letter denote big-endiannes, while in the old one they meant unsigned-ness.
>>>>
>>>> There's a trivia in the DoubleUpcall test, which was using 'unsigned double' ('D') - not sure why…
>>> Hmm, not sure.
>>>
>>> Cheers,
>>> Henry
>>>
>>>> Webrev here:
>>>>
>>>> http://cr.openjdk.java.net/~mcimadamore/panama/layout-grammar-refresh/
>>>>
>>>> P.S.
>>>> Might be useful to give it a spin on Mac and see if UnixSystemTest works (the layout conversion was not very trivial w/o jextract support :-))
>>>>
>>>> Cheers
>>>>
>>>> Maurizio