[转]New Features of C++0x

What's in C++0x

You can get a very good feel for the contents of C++0x by looking at the first partial draft mentioned above. That draft contains the bulk of the extensions the committee pulled in from the first C++ Standard Library Extensions Technical Report (TR1). It does not, however, contain text for all major features yet, and has placeholders for the major features that I'll summarize in this section.

Barring a disaster, here are the following additional major features not yet in the above partial draft that will be in C++0x. Every bullet heading links to a good "read this first" paper about the feature, and for some I've included additional links.

Concepts [N2081]

From that paper:

"Concepts introduce a type system for templates that makes templates easier to use and easier to write. By checking the requirements that templates place on their parameters prior to template instantiation, concepts can eliminate the spectacularly poor error messages produced by today’s template libraries, making them easier to use. ... At the core of concepts is the idea of separate type checking for templates. Template declarations are augmented with a set of constraints (requirements): if the definition of the template typechecks against these requirements, and a use of a template meets these requirements, then the template should not fail to instantiate."

Imagine, if you will, an error message like 'sorry, YourType isn't a BidirectionalIterator' instead of three pages of arcane messages about what templates couldn't be instantiated. Among other things, concepts will let us write the standard's own container and iterator requirements in code, rather than as English "cookbooks" of programming conventions.

The concepts proposal is coauthored by Bjarne Stroustrup and Doug Gregor. You can already start trying them out: There is a ConceptGCC proof-of-concept implementation that covers most of the proposal. For more details about how concepts can be applied to the C++ standard library, see also papers N2082, N2083, N2084, and N2085.

Garbage collection [N2129]

A C++0x compiler must provide garbage collection, which the programmer can use on request; "don't GC this region of code" is still the default. This "opt-in" model leaves existing programs' semantics unchanged, while providing real GC support for programs that want it. It is also intended to enable garbage collection for objects allocated and manipulated by most legacy libraries, which both makes it much easier to convert existing code to a garbage-collected environment and helps "repair" legacy code with deficient memory management. Of course, any C++98/03/0x compiler is already free to garbage-collect the whole C++ heap, and a number of current implementations do that; but they aren't required to do so, and this change would require GC to be available in a way that programmers can use and rely on. This work has been primarily driven by Hans Boehm and Mike Spertus. You can find a detailed paper here: N2128.

Memory model for concurrency [N2138]

As I wrote in "The Free Lunch Is Over", chip designers and compiler writers "are under so much pressure to deliver ever-faster CPUs that they’ll risk changing the meaning of your program, and possibly break it, in order to make it run faster." This only gets worse in the presence of multiple cores and processors.

A memory model is probably of the lowest-level treaty between programmers and would-be optimizers, and fundamental for any higher-level concurrency work. Quoting from my memory model paper: "A memory model describes (a) how memory reads and writes may be executed by a processor relative to their program order, and (b) how writes by one processor may become visible to other processors. Both aspects affect the valid optimizations that can be performed by compilers, physical processors, and caches, and therefore a key role of the memory model is to define the tradeoff between programmability (stronger guarantees for programmers) and performance (greater flexibility for reordering program memory operations)." This work is being driven primary by Hans Boehm, Clark Nelson, and myself.

Concurrency libraries [N2094]

The committee will likely release the concurrency libraries in two parts, both of which are likely to be released at about the same time.

In C++0x itself: Support at the level of basic threads and locks, as well as atomic operations for lock-free coding, and related features like thread local storage.

In the second C++ Standard Library Extensions Technical Report (aka Library TR2): More advanced extensions including futures and upgradable locks, so that we can gain experience with them before considering them for the standard itself post-C++0x. This follows the same path the committee took with the first set of library extensions, including shard_ptr and the hash-based containers: Those were first put into a separate TR (TR1), completed in 2004, and then based on experience most TR1 features were voted into the standard itself in April 2006 so that they are now part of the current partial draft of C++0x. The TR2 features are planned to be on a similar path, but for now are being released as a separate specification and not part of C++0x.

For more information and snapshots of current draft proposals still under discussion and development, see: N1907, N1966, N2047, and N2090.

What's not in C++0x

The following major features were cut from C++0x because they require more work and would delay the next standard. Once C++0x is done, the committee may pick them up again.

Modules [N2073]

It's definitely desirable to have a module system for C++ to package libraries and their implementations, and to provide an alternative to C++'s current build model based on textual source file inclusion. Reliance the current source file inclusion model poses technical difficulties for large C++ programs, notably: (a) long build times for large projects; and (b) difficulty of managing dependencies between interfaces and implementations. The current work has been spearheaded primarily by Daveed Vandevoorde (of EDG fame). The committee still intends to do work on this, but it's now planned as a separate Technical Report on Modules after C++0x is complete rather than as part of the C++ standard.

Dynamic libraries [N1496]

Although C++ programmers routinely use shared libraries in C++ today, the C++ standard doesn't specify their semantics in a portable way. Indeed, there are some fundamental semantic differences between some popular approaches, such as between Windows static and dynamic libraries and Linux shared libraries. Note that C++'s broad cross-platform support, and the existence of many existing approaches on those different platforms, makes it more difficult to specify portable library semantics than in languages like Java that started fresh with their own (then-new) run-time environment. Several committee members, notably Pete Becker, have done work to try to reconcile the existing approaches in a standard way that implementers could broadly support, but this effort will not be ready in time for C++0x.

 

Originally posted in Herb Sutters blog:http://herbsutter.spaces.live.com/blog/cns!2D4327CC297151BB!159.entry