|Project:||Programming Language C++|
|Reply to:||Alisdair Meredith <email@example.com>|
The register keyword was deprecated in the 2011 C++ standard, as its effect was already implicit in the language. It remains reserved for future use by the standard, and is time to remove its vestigial specification.
One theme of C++17 appears to be cleaning up older features, such as trigraphs (N4086), operator++(bool) (Core issue 1653), auto_ptr and the deprecated adaptable library function API (N4190). The register keyword has no normative function, yet occupies a place in the grammar that must be specified, so seems a good candidate for similar cleanup.
The register keyword was deprecated in the 2011 C++ standard in order to preserve our ability to repurpose a useful English keyword in future proposals, much as auto was repurposed in C++11. Its usage was deprecated, rather than removed, at the time as there was no conflicting immediate need for a proposed re-use at the time, unlike with auto. Similarly, the meaning of the export keyword was simply removed in C++11 due to implementation difficulty and a general lack of compilers implementing the feature. It was deemed appropriate to be more conservative with the widely implemented register keyword. However, users will have seen two standards published with the keyword deprecated, and with the general theme of clean-up occurring in C++17, should expect this advertised (potential) removal to finally occur.
One reason to consider retaining the keyword is compatibility with the C language. In particular, the register keyword can appear in a function parameter list. However, this usage has no meaning in C++, which may be problematic if calling into C code that ascribes a specific meaning to this hint.
However, C has further restrictions on register that do not translate to C++, that would impact only inline function definitions in shared headers. Code is ill-formed in C if it tries to take the address of a variable with the register storage-class-specifier. If an array is declared with the register storage-class-specifier in C, then it is undefined behavior to trigger array-to-pointer decay on that array.
Another comparison with C is the use of C99 keyword restrict in specifying interfaces. This seems much more widespread than the use of register, and has not proved to be a serious compatibility problem in the following 15 years.
Rather than try to resolve the differences with C, this paper prefers to remove the otherwise redundant feature from C++17.
The impact on the standard is minimal. The keyword is specified to be valid in a couple of places in the grammar, but explicitly has no meaning when used. It can safely be removed without impacting the integrity of the standard.
Looking beyond the standard: most of the early feedback to this paper has been that there appears, anecdotally, to be minimal use of the deprecated register keyword but considerably more use of the gcc extension based on this keyword. This suggests that future applications of the keyword may face practical constraints. Similarly, we should think very carefully before unreserving the keyword, even if we do not make immediate use of it again in the standard.
The purpose in removing the usage of register is to preserve the ability to use this valuable keyword for future proposals, such as modules. It is therefore retained as a reserved keyword, with a note on intent, similar to the handling of export in C++11. All other references to the register keyword are removed.
If interface compatibility with C is important enough to retain that usage, then we should undeprecate the use of register when declaration function parameters, potentially restricted to parameters of functions declared with extern "C" linkage.
We might also consider support for restrict as a contextual keyword for functions declared with extern "C" linkage as part of the same design concern.
An updated paper along these lines would continue to advocate the removal of register as a storage-class-specifier for local variables though.
The other suggestion received while authoring this paper is that it does not go far enough, and the register keyword should be released as an identifier to the user community.
register is certainly a flexible word offering significant use as both a verb (popular choice for function names, such as registering with a factory facility) and as a noun (for variables and class types).
If no significant proposal comes forward making use of register in the near future, it would be simple to amend 2.11 to no longer reserve the keyword. This paper recommends the more conservative approach for now though, of continuing to reserve a valuable term for standard use, as it would be difficult to reclaim once released.
Make the following edits (relative to
removals marked like so:
1 The identifiers shown in Table 3 are reserved for use as keywords (that is, they are unconditionally
treated as keywords in phase 7) except in an attribute-token (7.6.1)
[ Note: The export keyword
unused but is reserved for future use. — end note ]:
1 Block-scope variables
explicitly declared register or not explicitly declared
static or extern have automatic storage duration. The storage for these
entities lasts until the block in which they are created exits.
1 The storage class specifiers are
registerstatic thread_local extern mutable
At most one storage-class-specifier shall appear in a given decl-specifier-seq, except that thread_local may appear with static or extern. If thread_local appears in any declaration of a variable it shall be present in all declarations of that entity. If a storage-class-specifier appears in a decl-specifier-seq, there can be no typedef specifier in the same decl-specifier-seq and the init-declarator-list of the declaration shall not be empty (except for an anonymous union declared in a named namespace or in the global namespace, which shall be declared static (9.5)). The storage-class-specifier applies to the name declared by each init-declarator in the list and not to any names declared by other specifiers. A storage-class-specifier other than thread_local shall not be specified in an explicit specialization (14.7.3) or an explicit instantiation (14.7.2) directive.
The register specifier shall be applied only to names of variables declared in a block (6.3)
or to function parameters (8.4). It specifies that the named variable has automatic storage duration (3.7.3).
A variable declared without a storage-class-specifier at block scope or
declared as a function parameter has automatic storage duration by default.
3 A register specifier is a hint to the implementation that the variable so declared will be heavily used.
[ Note: The hint can be ignored and in most implementations it will be ignored if the address of the
variable is taken. This use is deprecated (see D.2). — end note ]
1 An alignment-specifier may be applied to a variable or to a class data member,
but it shall not be applied to a bit-field, a function parameter, an
, or a variable declared with the register
storage class specifier. An alignment-specifier may also be applied to the
declaration or definition of a class (in an elaborated-type-specifier (220.127.116.11)
or class-head (Clause 9), respectively) and to the declaration or definition of an
enumeration (in an opaque-enum-declaration or enum-head, respectively (7.2)).
An alignment-specifier with an ellipsis is a pack expansion (14.5.3).
2 A static, thread_local, extern,
mutable, friend, inline, virtual, or typedef
specifier applies directly to each declarator-id in an init-declarator-list;
the type specified for each declarator-id depends on both the decl-specifier-seq
and its declarator.
5 A member shall not be declared with the extern
storage-class-specifier. Within a class definition, a member shall not be declared with
the thread_local storage-class-specifier unless also declared static.
registerstatic thread_local extern mutable
1 The use of the register keyword as a storage-class-specifier (7.1.1) is deprecated.
Thanks for early reviews of this paper, while acknowledging that does not imply endorsing its contents, by Alan Bellingham, Bret Brown, John Fletcher, Al Grant, Kevlin Henney, Mike Miller, Nathan Myers, Roger Orr, Richard Smith, Ville Voutilainen, Jonathan Wakely, and Derek Woolverton.