On 02/27/2012 11:43 AM, Tobias Brandt wrote:

>>> If you wrote to the file with a C++ program, then I guess the
>>> compiler aligned the data so that the whole struct is 128 bytes
>>> in size. Technically, the C++ compiler is allowed to do
>>> anything short of changing the order of the struct fields.
>
>> That is correct for non-POD types. The C++ compiler must treat
>> POD structs essentially as if they are C structs.
>
> Correct me if I'm wrong. But as far a I know the C standard also
> allows arbitrary alignment.

You were correct. I somehow misread "short of changing the order" as meaning "even changing the order". But even then I wasn't entirely correct.

Just  found this thread:

  http://stackoverflow.com/q/281045

C guarantees that the members are not reordered, but C++ allows reordering by "The order of allocation of nonstatic data members separated by an access-specifier is unspecified (11.1)."

Ali

On Monday, 27 February 2012 at 19:28:07 UTC, Tobias Brandt wrote:
>> So, something like this should work:
>> [...]
>
> It really depends on how you wrote the file originally. If you
> know that it is packed, i.e. 10+32+32+32=106 bytes per record,
> then yes.
>
> If you wrote to the file with a C++ program, then I guess the
> compiler aligned the data so that the whole struct is 128 bytes
> in size. Technically, the C++ compiler is allowed to do
> anything short of changing the order of the struct fields.
>
> You could just let your C++ program print out sizeof(TaqIdx) or
> manually divide the file size by the number of records (if you
> know it) to make sure.

Just looked at my old C++ code.  And the struct looks like this:

struct TaqIdx {
  char symbol[10];
  int tdate;
  int begrec;
  int endrec;
}__attribute__((packed));

So I am guessing I want to use the align(1) as Justin suggested. Correct?

TJB

> Just looked at my old C++ code.  And the struct looks like this:
>
>
> struct TaqIdx {
>  char symbol[10];
>  int tdate;
>  int begrec;
>  int endrec;
> }__attribute__((packed));
>
> So I am guessing I want to use the align(1) as Justin suggested. Correct?

Yes.