Defect Report #074

Submission Date: 03 Dec 93
Submittor: WG14
Source: Clive Feather
Item 11 - alignment and structure padding
The existence of structure padding (subclause can be detected by a strictly conforming program by use of the sizeof operator and the offsetof macro.
  1. If a structure has a field of type t, can the alignment requirements of the field be different from the alignment requirements of objects of the same type that are not members of structures?

    If the answer to (a) is ``yes,'' then where applicable the remaining questions should be assumed to have been asked for both objects within structures and objects outside structures.

  2. If an array has a component type of t, can the alignment requirements of the elements of the array be different from those of independent variables of type t?

    The alignment requirement of a type is that addresses of objects of that type must be multiples of some constant (subclause 3.1); for some type t, this is written A(t) in this Defect Report.

  3. For any type t, can the expression sizeof (t) % A(t) be non-zero (in other words, can A(t) be a value other than 1, sizeof (t), or a factor of sizeof (t))? It would appear not, because otherwise adjacent elements of an array of objects of type t would either not be correctly aligned, or else would not be contiguously allocated.
  4. Can A(struct foo) be greater than the least common multiple of A(type_1), A(type_2), ..., A(type_n), where type_1 to type_n are the types of the elements of struct foo? In particular, if a structure holds exactly one element, can A(structure type) be different from A(element type)? (In each case, if the answer to (a) is ``yes,'' A(type) should be interpreted appropriately.)
  5. If, at any point in a structure or union (obviously excluding the start), there is more than one size of padding that can satisfy all alignment requirements, can any size be used, or must the smallest (possibly zero) padding be used because that is all that is ``necessary to achieve the appropriate alignment?''
  6. If a structure type has trailing padding to ensure that its use as an array element would be correctly aligned, must objects of that type which are not array elements also have the padding? If not, what is the effect of using memcpy to copy the value of one such object to another thus?

    struct fred a, b;
    ... /*
    memcpy(&a, &b, sizeof (struct fred));

    It appears from subclause (``the size is determined from the type of the operand'') that sizeof a must equal sizeof (struct fred). Is this correct?

  7. When an element of a structure is in turn a structure, can trailing padding of the inner structure be reused to hold other elements of the enclosing structure? For example, in:

    struct outer
    struct inner { long a; char b; } inner;
    char c;

    is it permitted for offsetof (struct outer, c) to be less than sizeof (struct inner)?
Subclause says, ``... pointers to qualified or unqualified versions of compatible types shall have the same representation and alignment requirements.''
Subclause says, ``Each non-bit-field member of a structure or union object is aligned in an implementation-defined manner appropriate to its type.'' And later, ``There may therefore be unnamed padding within a structure object, ... as necessary to achieve the appropriate alignment.''
a) It is possible for an implementation to state generalized requirements to satisfy sublause These requirements may be further strengthened using the implementation-defined behavior made available in subclause Yes, the alignment requirements can be different.
b) In several places the C Standard states that a single object may be treated as an array of one element. Nowhere does it give permission for array element types to have different alignment requirements from isolated object types.
c) sizeof (t) must indeed be a multiple of A(t).
d) Yes. A structure object can have an alignment that is greater than the least common multiple of the alignments of its members.
e) The phrase ``necessary to achieve the appropriate alignment'' is not considered to mean the use of the minimum padding possible. The Committee does not see any advantage to changing this phrase.
f) Yes. See answer to question (b). sizeof (struct fred) must equal sizeof a.
g) Such sharing of storage by objects would cause the requirements of subclause 6.3 to be violated and is not allowed.
Previous Defect Report < - > Next Defect Report