This issue has been automatically converted from the original issue lists and some formatting may not have been preserved.
Authors: Clive Feather, WG14
Date: 1993-12-03
Submitted against: C90
Status: Closed
Converted from: dr.htm, dr_074.html
Item 11 - alignment and structure padding
The existence of structure padding (subclause 6.5.2.1) can be detected by a
strictly conforming program by use of the sizeof operator and the offsetof
macro.
a. 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.
b. 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.
c. 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.
d. 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.)
e. 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?”
f. 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 6.3.3.4 (“the size is determined from the type of the
operand”) that sizeof a must equal sizeof (struct fred). Is this correct?
g. 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)?
Comment from WG14 on 1997-09-23:
Subclause 6.1.2.5 says, “... pointers to qualified or unqualified versions of compatible types shall have the same representation and alignment requirements.”
Subclause 6.5.2.1 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 subclause 6.1.2.5. These requirements may be further strengthened using the implementation-defined behavior made available in subclause 6.5.2.1. 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.