constexpr

Welcome to the home of the ISO C++0x constexpr feature as implemented in GCC. Say yes to magic! For more details on this feature, also called generalized constant expressions, see N2235 for background and language changes and N2976 for library changes .

Git it

Development is hosted in git. Try it, you'll like it.

Starting from a git working tree as described in GitMirror you can set up to use the constexpr repository as follows:

git checkout -b constexpr origin/constexpr

To stay up to date, use:

git pull

Status

2010-11-02

This branch has been merged to trunk: all future work on constexpr will be part of the usual gcc efforts.

The branch merging, in patches:

C++0x constexpr PATCH #1: improve setting of CLASSTYPE_LITERAL_P

C++0x constexpr PATCH #2: improve type checking of constexpr declarations

C++0x constexpr patch #3: implicit constexpr

C++0x constexpr PATCH #4: is_literal_type

C++0x constexpr PATCH #5: two more small tweaks

C++0x constexpr patch #6: fix for constexpr setting on implicit default constructor with --disable-checking

C++0x constexpr PATCHes #7-10: the rest of the compiler support

C++0x constexpr PATCHes #11-13: library usage

Since the merge, many bugs have been fixed. Please use gcc trunk, not the constexpr branch, if you care about constexpr and wish to play with it.

G++:

See bugzilla query for current status http://gcc.gnu.org/bugzilla/buglist.cgi?quicksearch=constexpr

LIBSTDC++:

Notes:

There is an attempt to communicate the extensions in the GNU implementation to the ISO C++ LWG. That paper will be in the post-Batavia meeting mailing, and the draft can be previewed now. Constexpr Library Additions

1) <chrono>

Remove the following:

Add the following:

2) <bitmask>, Done, except for bool operator[], may not be able to be constexpr. NEEDS SPECIFICATION

3) <ios> ios_base:: bitmask types new operators are not constexpr, don't know if this is intended. This requires a signature change from ref to value.

4) <atomic> atomic_integral::is_atomic could be constexpr, except this changes some design decisions about compile-time vs. run-time. Current design is for run-time resolution. Note, some unresolved store/load const issues. Meta picture of how const variables interact needs clarification, tangential. Needs updating to Adjusting C++ Atomics for C Compatibility

5) <initializer_list> initializer_list members are all constexpr in implementation, GNU extensions

7) <regex>. As done as regex. std::regex_constants needs some normalization WRT bitmask types (same as ios_base). Testsuite for random seems to be numbered and have other oddities.

8) <array>, seems like most of this could be constexpr

9) <string> basic_string's npos, size/length/max_size could be constexpr

10) <complex>

Add the following:

11) <utility>

Add the following:

12) <future>

13) <tuple> make more ctors constexpr. Do it do it do it.

Add the following:

META 1: Use of constexpr and some form of exception specifications, either throw specifications or noexcept. It turns out this is not redundant: constexpr functions can throw.

Jason says: A constant expression is noexcept even if it involves a call to a constexpr function that isn't declared noexcept.

Here's an example of a valid constexpr that throws:

constexpr int f (int i) { return i > 0 ? i : throw 42; }

META-2: Use of constexpr specifier with default constructors vs. defaulted constructors.

META-3: Opportunistic use of constexpr on member functions of class templates that are literal types, and boolean comparison operators of literal types.

META-4: Use of constexpr and const in member functions. Some places have both: this is redundant. Pick one, assume the one picked will be constexpr.

META-5: Opportunistic use of constexpr on constructors of class templates of non-literal type to make static initialization explicitly constant (and not dynamic.)

META-6: Static const vs. Static constexpr. Pick static constexpr.

None: Constexpr (last edited 2013-08-23 15:10:25 by FrankChEigler)