llvm dev - Jan 2013 - [LLVMdev] Clang's approach to anonymous struct pointer parameters

For the following code:

    struct XBeePacket;
    
    typedef void (*CompletionProc)(XBeePacket* inPacket, void* inParam2);
    
    struct
    XBeePacket
    {
        bool            mField1;
        CompletionProc  mCompletionProc;
    };

Why does clang emit this IR?

    %struct.XBeePacket = type { i8, {}* }
    
    define void
    @MyCompletionProc(%struct.XBeePacket* %inPacket, i8* %inParam2)
        nounwind uwtable ssp
    {
    entry:
      ...
    }

It makes the struct field for the completion proc an an anonymous aggregate
pointer. Is that better than making the first parameter to MyCompletionProc
void* or something similar?

I guess there's no way in LLVM IR to forward-declare in a way that solves
this, huh?

-- 
Rick

Hi Rick,

this is a bug in Clang's LLVM-IR code generator:
http://llvm.org/bugs/show_bug.cgi?id=14920

Best,
Florian

On Sunday, January 20, 2013 01:57:37 Rick Mann wrote:
> For the following code:
> 
>     struct XBeePacket;
>     
>     typedef void (*CompletionProc)(XBeePacket* inPacket, void* inParam2);
>     
>     struct
>     XBeePacket
>     {
>         bool            mField1;
>         CompletionProc  mCompletionProc;
>     };
> 
> Why does clang emit this IR?
> 
>     %struct.XBeePacket = type { i8, {}* }
>     
>     define void
>     @MyCompletionProc(%struct.XBeePacket* %inPacket, i8* %inParam2)
>         nounwind uwtable ssp
>     {
>     entry:
>       ...
>     }
> 
> It makes the struct field for the completion proc an an anonymous aggregate
pointer. Is that better than making the first parameter to MyCompletionProc
void* or something similar?
> 
> I guess there's no way in LLVM IR to forward-declare in a way that
solves this, huh?
> 
> 
-- 
Florian Brandner
Embedded Systems Engineering Group
Department of Applied Mathematics and Computer Science
Technical University of Denmark
Richard Petersens Plads
Building 322, room 206
2800 Lyngby
Denmark

phone: +45 45255223
web: http://www.imm.dtu.dk/~flbr/
email: flbr at imm.dtu.dk

On Jan 20, 2013, at 5:38, Florian Brandner <flbr at imm.dtu.dk> wrote:
> Hi Rick,
> 
> this is a bug in Clang's LLVM-IR code generator:
> http://llvm.org/bugs/show_bug.cgi?id=14920
Oh, interesting, thanks. 

How do I create a recursive definition like this?
> 
> Best,
> Florian
> 
> On Sunday, January 20, 2013 01:57:37 Rick Mann wrote:
>> For the following code:
>> 
>>    struct XBeePacket;
>> 
>>    typedef void (*CompletionProc)(XBeePacket* inPacket, void*
inParam2);
>> 
>>    struct
>>    XBeePacket
>>    {
>>        bool            mField1;
>>        CompletionProc  mCompletionProc;
>>    };
>> 
>> Why does clang emit this IR?
>> 
>>    %struct.XBeePacket = type { i8, {}* }
>> 
>>    define void
>>    @MyCompletionProc(%struct.XBeePacket* %inPacket, i8* %inParam2)
>>        nounwind uwtable ssp
>>    {
>>    entry:
>>      ...
>>    }
>> 
>> It makes the struct field for the completion proc an an anonymous
aggregate pointer. Is that better than making the first parameter to
MyCompletionProc void* or something similar?
>> 
>> I guess there's no way in LLVM IR to forward-declare in a way that
solves this, huh?
> -- 
> Florian Brandner
> Embedded Systems Engineering Group
> Department of Applied Mathematics and Computer Science
> Technical University of Denmark
> Richard Petersens Plads
> Building 322, room 206
> 2800 Lyngby
> Denmark
> 
> phone: +45 45255223
> web: http://www.imm.dtu.dk/~flbr/
> email: flbr at imm.dtu.dk
> 
> 
>