Hacker News Reader

Comment by Animus7 | Hacker News Reader

Awesome! I've been looking forward to the day I'm no longer relying on "experimental" C++0x support! Here's a smaller tl;dr for those who want it:

-foreach loop

-first-class rvalue ("temporary") types

-lambda functions + closures

-implicit typing (auto keyword)

-decltype(), getting "declared type" of any expression

-variadic ... templates

-expanded STL -- incl. threading and RNG's

-construction from C-style initializer list

-Unicode literals

-enum class that doesn't auto-decay to int, enums with configurable base type

-explicit strong nullptr constant; no more NULL macro nonsense

The rest of the stuff is (in my opinion) less general/noteworthy.

View on HN

RNG's... Did they include Mersenne Twister?

http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/emt.html

EDIT: Apologies... the only reason I asked was to contribute to the discussion, but now I realize it probably looked like I was being lazy.

Yes - there are several options, which MT is one of: http://en.wikipedia.org/wiki/C_0x#Extensible_random_number_f...

Mersenne Twister is hardly perfect. It's slow, causes a lot of cache misses, and IIRC the output is purely XOR of previous outputs. Do people just bring it up because they like the name?

Deserved or not, Mersenne Twister is somewhat famous among laymen like myself for being very fast relative to it's reasonably high quality. If you can point me at an RNG that is both faster and higher quality, I will be genuinely in your debt.

My layman's understanding is that George Marsaglia's xorshift generator[1] is considerably faster than the Mersenne Twister while still providing high quality randomness.

I've also seen WELL512[2] mentioned in a few places, but I don't know much more about it than the name.

[1]: First introduced here: http://groups.google.com/group/comp.lang.c/msg/e3c4ea1169e46...

[2]: http://www.iro.umontreal.ca/~panneton/WELLRNG.html

Let me know if you are ever in San Francisco and I'll open the tab at one of our many fine beer-snob bars (or a wine-snob bar, if you prefer).

This thread & invite is still open to anyone who can provide further suggestions!

> explicit strong nullptr constant; no more NULL macro nonsense

What is the point of this? C++ defines the null pointer to be always 0. So I never needed the NULL macro in C++ anyway, as I'm allowed to simply type 0 instead.

In how far is that new nullptr constant preferable to writing simply 0?

It's all about compile time warnings and static type information. The resulting code should be the same whether you're using 0, NULL or nullptr.

not e.g. with 64-bits, va args and literal 0 (cf FrankBooth example)

You may know that a null pointer is always 0, but you don't know that 0 is always a null pointer -- it may be the result of subtracting an integer from itself. That's the difference.

Consider:

  int execl(const char *path, const char *arg, ...);

called like so:

  execl("foo", "bar", 0);

Particularly when sizeof(int) != sizeof(void *).

In overloaded or argument-deduced contexts, then "0" is preferentially an int, but nullptr is never an int.

The standard example is:

    void f(int x);
    void f(char *x);
    ...
    f(0);  // calls void f(int)

That may seem contrived, but you may not know of the f(int) overload, especially in combination with templates.

NULL, if #define'd as (void *)0, prevents that error.

NULL is actually #defined to 0 in C++ as void* is not implicitly convertible to any pointer type as it is in C. So unlike in C there has been no safety in using NULL. Until now.

Don't forget the auto keyword.

The question is, how long before I can actually use this stuff?

auto falls under "implicit typing". Clarified this.

And we've been able to use a lot of C++0x for a while now:

http://gcc.gnu.org/projects/cxx0x.html

Visual Studio 2010 has about half the C++11 features (obviously it came out before the full standard was finished). Next release I guess will add more.

Or maybe the next service pack. Thinking of VS2008 feature pack/SP1 which added TR1.

And status for clang: http://clang.llvm.org/cxx_status.html

With the C++98 standard it took almost a decade to get decent support in mainstream compilers.

This time it's different. Most of C++0x is already implemented in GCC and MSVC. However the implemented subsets are a bit different, but most of it is already there. The last time I checked, MSVC had lots of missing stuff in the standard library. In particular, threads and clocks were missing.

I've been writing C++0x with GCC for two years now.

No love for the low level concurrency support?

Yeah, threads, clocks and atomics are missing from this list.

Yeah, I'm building my subset of C++B right now. Much of the new stuff is either too complex or you will use it too rarely to make use without looking it up.

Reply on news.ycombinator.com