13.3 Operational and Representation Attributes

In general, the meaning of a given attribute should not depend on whether the attribute was specified via an attribute_definition_clause, or chosen by default by the implementation. 

Ramification: For example, the Size attribute is of type universal_integer. Therefore, the expected type for Y in “for X'Size use Y;” is universal_integer, which means that Y can be of any integer type. 

Discussion: For attributes that denote subprograms, the required profile is indicated separately for the individual attributes. 

Ramification: For an attribute_definition_clause with a name, the name need not statically denote the entity it denotes. For example, the following kinds of things are allowed: 

Discussion: For each specifiable attribute, we generally say something like, “The ... attribute may be specified for ... via an attribute_definition_clause.”

The above wording allows for T'Class'Alignment, T'Class'Size, T'Class'Input, and T'Class'Output to be specifiable.

A specifiable attribute is not necessarily specifiable for all entities for which it is defined. For example, one is allowed to ask T'Component_Size for an array subtype T, but “for T'Component_Size use ...” is only allowed if T is a first subtype, because Component_Size is a type-related aspect. 

Ramification: This implies, for example, that if one writes: 

R has to be a procedure with two parameters with the appropriate subtypes and modes as shown in 13.13.2.

To be honest: {type-related (attribute_definition_clause) [partial]} {subtype-specific (attribute_definition_clause) [partial]} An attribute_definition_clause is type-related or subtype-specific if the attribute_designator denotes a type-related or subtype-specific attribute, respectively. 

Discussion: A storage element is not intended to be a single bit, unless the machine can efficiently address individual bits. 

Ramification: For example, on a machine with 8-bit storage elements, if there exist 32-bit integer registers, with a full set of arithmetic and logical instructions to manipulate those registers, a word ought to be 4 storage elements — that is, 32 bits. 

Discussion: The “given the implementation's run-time model” part is intended to imply that, for example, on an 80386 running MS-DOS, the word might be 16 bits, even though the hardware can support 32 bits.

A word is what ACID refers to as a “natural hardware boundary”.

Storage elements may, but need not be, independently addressable (see 9.10, “Shared Variables”). Words are expected to be independently addressable.

Implementation defined: The set of machine scalars.

Ramification: Here, the “first of the storage elements” is intended to mean the one with the lowest address; the endianness of the machine doesn't matter.

Ramification: Address is not allowed for enumeration literals, predefined operators, derived task types, or derived protected types, since they are not program units.

The validity of a given address depends on the run-time model; thus, in order to use Address clauses correctly, one needs intimate knowledge of the run-time model.

If the Address of an object is specified, any explicit or implicit initialization takes place as usual, unless a pragma Import is also specified for the object (in which case any necessary initialization is presumably done in the foreign language).

Any compilation unit containing an attribute_reference of a given type depends semantically on the declaration of the package in which the type is declared, even if not mentioned in an applicable with_clause — see 10.1.1. In this case, it means that if a compilation unit contains X'Address, then it depends on the declaration of System. Otherwise, the fact that the value of Address is of a type in System wouldn't make sense; it would violate the “legality determinable via semantic dependences” Language Design Principle.

AI83-00305 — If X is a task type, then within the body of X, X denotes the current task object; thus, X'Address denotes the object's address.

Interrupt entries and their addresses are described in J.7.1, “Interrupt Entries”.

If X is not allocated on a storage element boundary, X'Address points at the first of the storage elements that contains any part of X. This is important for the definition of the Position attribute to be sensible. 

Implementation Advice: For an array X, X'Address should point at the first component of the array rather than the array bounds.

Ramification: On the other hand, we have no advice to offer about discriminants and tag fields; whether or not the address points at them is not specified by the language. If discriminants are stored separately, then the Position of a discriminant might be negative, or might raise an exception. 

Reason: Aliased objects are the ones for which the Unchecked_Access attribute is allowed; hence, these have to be allocated on an addressable boundary anyway. Similar considerations apply to objects of a by-reference type.

An implementation need not go to any trouble to make Address work in other cases. For example, if an object X is not aliased and not of a by-reference type, and the implementation chooses to store it in a register, X'Address might return System.Null_Address (assuming registers are not addressable). For a subprogram whose calling convention is Intrinsic, or for a package, the implementation need not generate an out-of-line piece of code for it. 

Implementation Advice: The recommended level of support for the Address attribute should be followed.

The intended meaning of the various attributes, and their attribute_definition_clauses, is more explicit.

The address_clause has been renamed to at_clause and moved to Annex J, “Obsolescent Features”. One can use an Address clause (“for T'Address use ...;”) instead.

The attributes defined in RM83-13.7.3 are moved to Annex G, A.5.3, and A.5.4. 

By default, the Alignment of a subtype should reflect the “natural” alignment for objects of the subtype on the machine. The Alignment, whether specified or default, should be known at compile time, even though Addresses are generally not known at compile time. (The generated code should never need to check at run time the number of zero bits at the end of an address to determine an alignment).

There are two symmetric purposes of Alignment clauses, depending on whether or not the implementation has control over object allocation. If the implementation allocates an object, the implementation should ensure that the Address and Alignment are consistent with each other. If something outside the implementation allocates an object, the implementation should be allowed to assume that the Address and Alignment are consistent, but should not assume stricter alignments than that.

Ramification: The Alignment is passed by an allocator to the Allocate operation; the implementation has to choose a value such that if the address returned by Allocate is aligned as requested, the generated code can correctly access the object.

The above mention of “modulo” is referring to the "mod" operator declared in System.Storage_Elements; if X mod N = 0, then X is by definition aligned on an N-storage-element boundary. 

Ramification: The user has to either give an Alignment clause also, or else know what Alignment the implementation will choose by default. 

Ramification: {AI95-00291-02} There is no recommendation to support any nonconfirming Alignment clauses for types not mentioned above. Remember that 13.1 requires support for confirming Alignment clauses for all types. 

Implementation Advice: The recommended level of support for the Alignment attribute should be followed.

Discussion: Most objects are allocated by the implementation; for these, the implementation obeys the Alignment. The implementation is of course allowed to make an object more aligned than its Alignment requires — an object whose Alignment is 4 might just happen to land at an address that's a multiple of 4096. For formal parameters, the implementation might want to force an Alignment stricter than the parameter's subtype. For example, on some systems, it is customary to always align parameters to 4 storage elements.

Hence, one might initially assume that the implementation could evilly make all Alignments 1 by default, even though integers, say, are normally aligned on a 4-storage-element boundary. However, the implementation cannot get away with that — if the Alignment is 1, the generated code cannot assume an Alignment of 4, at least not for objects allocated outside the control of the implementation.

Of course implementations can assume anything they can prove, but typically an implementation will be unable to prove much about the alignment of, say, an imported object. Furthermore, the information about where an address “came from” can be lost to the compiler due to separate compilation.

{AI95-00114-01} The Alignment of an object that is a component of a packed composite object will usually be 0, to indicate that the component is not necessarily aligned on a storage element boundary. For a subtype, an Alignment of 0 means that objects of the subtype are not normally aligned on a storage element boundary at all. For example, an implementation might choose to make Component_Size be 1 for an array of Booleans, even when pragma Pack has not been specified for the array. In this case, Boolean'Alignment would be 0. (In the presence of tasking, this would in general be feasible only on a machine that had atomic test-bit and set-bit instructions.)

If the machine has no particular natural alignments, then all subtype Alignments will probably be 1 by default.

Specifying an Alignment of 0 in an attribute_definition_clause does not require the implementation to do anything (except return 0 when the Alignment is queried). However, it might be taken as advice on some implementations.

It is an error for an Address clause to disobey the object's Alignment. The error cannot be detected at compile time, in general, because the Address is not necessarily known at compile time (and is almost certainly not static). We do not require a run-time check, since efficiency seems paramount here, and Address clauses are treading on thin ice anyway. Hence, this misuse of Address clauses is just like any other misuse of Address clauses — it's erroneous.

A type extension can have a stricter Alignment than its parent. This can happen, for example, if the Alignment of the parent is 4, but the extension contains a component with Alignment 8. The Alignment of a class-wide type or object will have to be the maximum possible Alignment of any extension.

The recommended level of support for the Alignment attribute is intended to reflect a minimum useful set of capabilities. An implementation can assume that all Alignments are multiples of each other — 1, 2, 4, and 8 might be the only supported Alignments for subtypes. An Alignment of 3 or 6 is unlikely to be useful. For objects that can be allocated statically, we recommend that the implementation support larger alignments, such as 4096. We do not recommend such large alignments for subtypes, because the maximum subtype alignment will also have to be used as the alignment of stack frames, heap objects, and class-wide objects. Similarly, we do not recommend such large alignments for stack-allocated objects.

If the maximum default Alignment is 8 (say, Long_Float'Alignment = 8), then the implementation can refuse to accept stricter alignments for subtypes. This simplifies the generated code, since the compiler can align the stack and class-wide types to this maximum without a substantial waste of space (or time).

Note that the recommended level of support takes into account interactions between Size and Alignment. For example, on a 32-bit machine with 8-bit storage elements, where load and store instructions have to be aligned according to the size of the thing being loaded or stored, the implementation might accept an Alignment of 1 if the Size is 8, but might reject an Alignment of 1 if the Size is 32. On a machine where unaligned loads and stores are merely inefficient (as opposed to causing hardware traps), we would expect an Alignment of 1 to be supported for any Size. 

The nonnegative part is missing from RM83 (for mod_clauses, nee alignment_clauses, which are an obsolete version of Alignment clauses). 

Ramification: Note that Size is in bits even if Machine_Radix is 10. Each decimal digit (and the sign) is presumably represented as some number of bits. 

Implementation Advice: The Size of an array object should not include its bounds.

Implementation defined: The meaning of Size for indefinite subtypes.

Reason: The effects of specifying the Size of a subtype are: 

A component_clause or a Component_Size clause can cause an object to be smaller than its subtype's specified size. A pragma Pack cannot; if a component subtype's size is specified, this limits how tightly the composite object can be packed.

The Size of a class-wide (tagged) subtype is unspecified, because it's not clear what it should mean; it should certainly not depend on all of the descendants that happen to exist in a given program. Note that this cannot be detected at compile time, because in a generic unit, it is not necessarily known whether a given subtype is class-wide. It might raise an exception on some implementations. 

Ramification: A Size clause for a numeric subtype need not affect the underlying numeric type. For example, if I say: 

I am not guaranteed that S'Base'Last >= 2**63–1, nor that intermediate results will be represented in 64 bits. 

Reason: There is no need to complicate implementations for this sort of thing, because the right way to affect the base range of a type is to use the normal way of declaring the base range: 

Ramification: The Size of a large unconstrained subtype (e.g. String'Size) is likely to raise Constraint_Error, since it is a nonstatic expression of type universal_integer that might overflow the largest signed integer type. There is no requirement that the largest integer type be able to represent the size in bits of the largest possible object. 

Implementation Advice: If the Size of a subtype allows for efficient independent addressability, then the Size of most objects of the subtype should equal the Size of the subtype.

Ramification: Thus, on a typical 32-bit machine, “for S'Size use 32;” will guarantee that aliased objects of subtype S, and components whose subtype is S, will have Size = 32 (assuming the implementation chooses to obey this Implementation Advice). On the other hand, if one writes, “for S2'Size use 5;” then stand-alone objects of subtype S2 will typically have their Size rounded up to ensure independent addressability.

Note that “for S'Size use 32;” does not cause things like formal parameters to have Size = 32 — the implementation is allowed to make all parameters be at least 64 bits, for example.

Note that “for S2'Size use 5;” requires record components whose subtype is S2 to be exactly 5 bits if the record type is packed. The same is not true of array components; their Size may be rounded up to the nearest factor of the word size.

Implementation Note: {AI95-00291-02} {gaps} On most machines, arrays don't contain gaps between elementary components; if the Component_Size is greater than the Size of the component subtype, the extra bits are generally considered part of each component, rather than gaps between components. On the other hand, a record might contain gaps between elementary components, depending on what sorts of loads, stores, and masking operations are generally done by the generated code.

{AI95-00291-02} For an array, any extra bits stored for each elementary component will generally be part of the component — the whole point of storing extra bits is to make loads and stores more efficient by avoiding the need to mask out extra bits. The PDP-10 is one counter-example; since the hardware supports byte strings with a gap at the end of each word, one would want to pack in that manner. 

Implementation Advice: A Size clause on a composite subtype should not affect the internal layout of components.

Reason: That's what Pack pragmas, record_representation_clauses, and Component_Size clauses are for. 

Implementation Note: This applies to static enumeration subtypes, using the internal codes used to represent the values.

For a two's-complement machine, this implies that for a static signed integer subtype S, if all values of S are in the range 0 .. 2ⁿ–1, or all values of S are in the range –2^n–1 .. 2^n–1–1, for some n less than or equal to the word size, then S'Size should be <= the smallest such n. For a one's-complement machine, it is the same except that in the second range, the lower bound “–2^n–1” is replaced by “–2^n–1+1”.

If an integer subtype (whether signed or unsigned) contains no negative values, the Size should not include space for a sign bit.

Typically, the implementation will choose to make the Size of a subtype be exactly the smallest such n. However, it might, for example, choose a biased representation, in which case it could choose a smaller value.

On most machines, it is in general not a good idea to pack (parts of) multiple stand-alone objects into the same storage element, because (1) it usually doesn't save much space, and (2) it requires locking to prevent tasks from interfering with each other, since separate stand-alone objects are independently addressable. Therefore, if S'Size = 2 on a machine with 8-bit storage elements, the size of a stand-alone object of subtype S will probably not be 2. It might, for example, be 8, 16 or 32, depending on the availability and efficiency of various machine instructions. The same applies to components of composite types, unless packing, Component_Size, or record layout is specified.

For an unconstrained discriminated object, if the implementation allocates the maximum possible size, then the Size attribute should return that maximum possible size. 

Ramification: The Size of an object X is not usually the same as that of its subtype S. If X is a stand-alone object or a parameter, for example, most implementations will round X'Size up to a storage element boundary, or more, so X'Size might be greater than S'Size. On the other hand, X'Size cannot be less than S'Size, even if the implementation can prove, for example, that the range of values actually taken on by X during execution is smaller than the range of S.

For example, if S is a first integer subtype whose range is 0..3, S'Size will be probably be 2 bits, and components of packed composite types of this subtype will be 2 bits (assuming Storage_Unit is a multiple of 2), but stand-alone objects and parameters will probably not have a size of 2 bits; they might be rounded up to 32 bits, for example. On the other hand, Unchecked_Conversion will use the 2-bit size, even when converting a stand-alone object, as one would expect.

Another reason for making the Size of an object bigger than its subtype's Size is to support the run-time detection of uninitialized variables. {uninitialized variables [partial]} The implementation might add an extra value to a discrete subtype that represents the uninitialized state, and check for this value on use. In some cases, the extra value will require an extra bit in the representation of the object. Such detection is not required by the language. If it is provided, the implementation has to be able to turn it off. For example, if the programmer gives a record_representation_clause or Component_Size clause that makes a component too small to allow the extra bit, then the implementation will not be able to perform the checking (not using this method, anyway).

The fact that the size of an object is not necessarily the same as its subtype can be confusing: 

The programmer might think that My_Device'Size is 8, and that My_Device'Address points at an 8-bit location. However, this is not true. In Ada 83 (and in Ada 95), My_Device'Size might well be 32, and My_Device'Address might well point at the high-order 8 bits of the 32-bit object, which are always all zero bits. If My_Device'Address is passed to an assembly language subprogram, based on the programmer's assumption, the program will not work properly. 

Reason: It is not reasonable to require that an implementation allocate exactly 8 bits to all objects of subtype Device_Register. For example, in many run-time models, stand-alone objects and parameters are always aligned to a word boundary. Such run-time models are generally based on hardware considerations that are beyond the control of the implementer. (It is reasonable to require that an implementation allocate exactly 8 bits to all components of subtype Device_Register, if packed.) 

Ramification: The correct way to write the above code is like this: 

If the implementation cannot accept 8-bit stand-alone objects, then this will be illegal. However, on a machine where an 8-bit device register exists, the implementation will probably be able to accept 8-bit stand-alone objects. Therefore, My_Device'Size will be 8, and My_Device'Address will point at those 8 bits, as desired.

If an object of subtype Device_Register is passed to a foreign language subprogram, it will be passed according to that subprogram's conventions. Most foreign language implementations have similar run-time model restrictions. For example, when passing to a C function, where the argument is of the C type char* (that is, pointer to char), the C compiler will generally expect a full word value, either on the stack, or in a register. It will not expect a single byte. Thus, Size clauses for subtypes really have nothing to do with passing parameters to foreign language subprograms. 

Ramification: For example, if a task object is represented as a pointer to some information (including a task stack), then the size of the object should be the size of the pointer. The Storage_Size, on the other hand, should include the size of the stack.

Discussion: {AI95-00051-02} Note that the “corresponding limitation” for a record or array type implies that an implementation may impose some reasonable maximum size for records and arrays (e.g. 2**32 bits), which is an upper bound (“capacity” limit) on the size, whether chosen by default or by being specified by the user. The largest size supported for records need not be the same as the largest size supported for arrays. 

Implementation Advice: The recommended level of support for the Size attribute should be followed.

Ramification: {AI95-00291-02} There is no recommendation to support any nonconfirming Size clauses for types not mentioned above. Remember that 13.1 requires support for confirming Size clauses for all types. 

{AI95-00114-01} We specify the meaning of Size in much more detail than Ada 83. This is not technically an inconsistency, but it is in practice, as most Ada 83 compilers use a different definition for Size than is required here. This should have been documented more explicitly during the Ada 9X process.

The requirement for a nonnegative value in a Size clause was not in RM83, but it's hard to see how it would make sense. For uniformity, we forbid negative sizes, rather than letting implementations define their meaning. 

Ramification: The value of this attribute is never negative, since it is impossible to “reserve” a negative number of storage elements.

If the implementation chooses to allocate an initial amount of storage, and then increase this as needed, the Storage_Size cannot include the additional amounts (assuming the allocation of the additional amounts can raise Storage_Error); this is inherent in the meaning of “reserved.”

The implementation is allowed to allocate different amounts of storage for different tasks of the same subtype.

Storage_Size is also defined for access subtypes — see 13.11. 

Ramification: The implementation is allowed to round up a specified Storage_Size amount. For example, if the implementation always allocates in chunks of 4096 bytes, the number 200 might be rounded up to 4096. Also, if the user specifies a negative number, the implementation has to normalize this to 0, or perhaps to a positive number. 

Implementation Note: The intent is that the value of X'Component_Size is always nonnegative. If the array is stored “backwards” in memory (which might be caused by an implementation-defined pragma), X'Component_Size is still positive. 

Ramification: For an array object A, A'Component_Size = A(I)'Size for any index I. 

Ramification: For example, if Storage_Unit = 8, and Word_Size = 32, then the user is allowed to specify a Component_Size of 1, 2, 4, 8, 16, and 32, with no gaps. In addition, n*32 is allowed for positive integers n, again with no gaps. If the implementation accepts Component_Size = 3, then it might allocate 10 components per word, with a 2-bit gap at the end of each word (unless packing is also specified), or it might not have any internal gaps at all. (There can be gaps at either end of the array.) 

Implementation Advice: The recommended level of support for the Component_Size attribute should be followed.

Implementation defined: The default external representation for a type tag.

Implementation defined: What determines whether a compilation unit is the same in two different partitions.

Reason: These requirements are important because external tags are used for input/output of class-wide types. These requirements ensure that what is written by one program can be read back by some other program so long as they share the same declaration for the type (and everything it depends on).

The user may specify the external tag if (s)he wishes its value to be stable even across changes to the compilation unit in which the type is declared (or changes in some unit on which it depends).

{AI95-00114-01} We use a String rather than a Stream_Element_Array to represent an external tag for portability. 

Ramification: Note that the characters of an external tag need not all be graphic characters. In other words, the external tag can be a sequence of arbitrary 8-bit bytes. 

Ramification: An implementation may specify that an implementation-defined attribute is specifiable for certain entities. This follows from the fact that the semantics of implementation-defined attributes is implementation defined. An implementation is not allowed to make a language-defined attribute specifiable if it isn't. 

{extensions to Ada 83} The syntax rule for length_clause is replaced with the new syntax rule for attribute_definition_clause, and it is modified to allow a name (as well as an expression). 

The syntax rule for attribute_definition_clause now requires that the prefix of the attribute be a local_name; in Ada 83 this rule was stated in the text.

{AI95-00114-01} In Ada 83, the relationship between a aspect_clause specifying a certain aspect and an attribute that queried that aspect was unclear. In Ada 95, they are the same, except for certain explicit exceptions. 

{8652/0009} {AI95-00137-01} Corrigendum: Added wording to specify for each attribute whether it is an operational or representation attribute.

{8652/0040} {AI95-00108-01} Corrigendum: Added wording to specify that External_Tag is never inherited.

{AI95-00051-01} {AI95-00291-01} Adjusted the Recommended Level of Support for Alignment to eliminate nonsense requirements and to ensure that useful capabilities are required.

{AI95-00051-01} {AI95-00291-01} Adjusted the Recommended Level of Support for Size to eliminate nonsense requirements and to ensure that useful capabilities are required. Also eliminated any dependence on whether an aspect was specified (a confirming representation item should not affect the semantics).

{AI95-00133-01} Added the definition of machine scalar.

{AI95-00247-01} Removed the requirement that specified alignments for a composite type cannot override those for their components, because it was never intended to apply to components whose location was specified with a representation item. Moreover, it causes a difference in legality when a confirming alignment is specified for one of the composite types.

{AI95-00291-02} Removed recommended level of support rules about types with by-reference and aliased parts, because there are now blanket rules covering all recommended level of support rules.

{AI95-00291-02} Split the definition of Alignment for subtypes and for objects. This simplified the wording and eliminated confusion about which rules applied to objects, which applied to subtypes, and which applied to both.