[Bug c++/68074] New: [C++14] Complex ConstExpr Evaluation

[Bug c++/68074] New: [C++14] Complex ConstExpr Evaluation

[arbmind at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68074

            Bug ID: 68074
           Summary: [C++14] Complex ConstExpr Evaluation
           Product: gcc
           Version: 5.2.0
               URL: http://goo.gl/T7h3Qg
            Status: UNCONFIRMED
          Keywords: rejects-valid, wrong-code
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: arbmind at gmail dot com
  Target Milestone: ---

Please take a look at my constexpr experiment.
http://goo.gl/T7h3Qg

GCC 5.2.0 and 5.1.0 has multiple issues with this code.
It compiles well in all variants on Clang 3.7 and 3.8 snapshots.

The goal of the experiment is to implement a multiplication table at compile
time. Peter Sommerlad showed it on CppCon 2015. Video
https://www.youtube.com/watch?v=DM-RXeiSCmc

This implementation makes much more use of constexpr features of C++14.

The issues are marked with comment in the code.

1. Currently the code compiles, but prints the strings backwards. Generating
the strings at runtime, does not schod this behavior. Clang gets this right as
well.

2. The auto generated copy & move operators of carray do behave different. In
compile time execution only the first element is handled. Using the class at
runtime works well. Clang does the same during compile time.

3. Adding an exception to the string generation triggers a strange compiler
error. "sorry, unimplemented: unexpected AST of kind loop_expr" Clang works
well.

4. GCC does not allow me to use range based for loops for carray. Clang has no
problem with that.

Thanks for all the good work. May my experiment help to improve g++.

[Bug c++/68074] [C++14] Complex ConstExpr Evaluation

[arbmind at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68074

--- Comment #1 from Andreas Reischuck <arbmind at gmail dot com> ---
Fixed some issues in my example.
http://goo.gl/Rxq7TP

The ranged for loops now compile on clang. The problem is not the loop, but the
address calculation.