From: Lou Vanek <vanek@...>
Date: 2006-02-12T06:45:24+09:00
Subject: Re: ruby/dl question

Joel VanderWerf wrote:

> ara.t.howard@noaa.gov wrote:
> 
>>On Sun, 12 Feb 2006, Joel VanderWerf wrote:
> 
> ...
> 
>>>I don't understand. If the flock struct fields are not in a predictable
>>>order, how do you access them in C?
>>
>>that's what to compiler does - it's why one must say
>>
>>  sizeof(mystruct)
>>
>>vs.
>>
>>  sizeof(int) + sizeof(float)
> 
> 
> Ah I see. It isn't that the data structure itself is unpredictable (I
> thought maybe you had to read one of the fields to know how the others
> are laid out). It's rather that you don't know what the compiler is
> doing with it.
> 
> Since there are a variety of alignment and packing options in any
> compiler, I don't see how DL could guess what the correct ones are.
> Maybe it assumes longword alignment?
> 
> On further digging... in dl.h the extra space taken to align a short,
> for example, is computed by...
> 
> typedef struct { char c; short x; } s_short;
> #define ALIGN_SHORT  (sizeof(s_short) - sizeof(short))
> #define SHORT_ALIGN  ALIGN_SHORT
> 
> ...and this is used by the following macro, which advances the offset
> var to the next aligned position...
> 
> #define DLALIGN(ptr,offset,align) {\
>   while( (((unsigned long)((char *)ptr + offset)) % align) != 0 ) offset++;\
> }
> 
> ...to calculate the struct size in dlsizeof() in dl.c:
> 
>     case 'H':
>       DLALIGN(0,size,SHORT_ALIGN);
>     case 'h':
>       size += sizeof(short) * n;
>       break;
> 
> So I guess if you use 'H', you get aligned shorts (according to whatever
> compiler flags ruby and its extensions were compiled with), and if you
> use 'h' you get packed shorts.
> 
> 
> Don't get your shorts out of alignment!  ;)

And, for what it's worth, both gcc 3.4.4 and VC++ 13.10 pack the two shorts together
into the same word on x86.