Compiler error: constexpr with bitfields.

Created attachment 44538
preprocessed file of a small test case for the constexpr crash

Hello,

I've created a small test case that produces a compiler crash. You can find attached the preprocessed file (main.ii) made by running the following command, with `G++ 7.3.1 20180303 (Red Hat 7.3.1-5)` on a CentOS 7 (with the devtoolset-7 installed), but it seems to exist on the last version too (see https://wandbox.org/permlink/zGZWdHIrRNJKfh5Z):

g++ -O3 -Wall -Wextra -o main -save-temps -v main.cpp

I give at the end of the mail the output of the command.

You can find a run on the last G++ version (9) here: https://wandbox.org/permlink/zGZWdHIrRNJKfh5Z

-----------------------
Here is the code:

#include <iostream>
struct Bugged {
    double x; bool isfreex; bool isfreey;
    constexpr bool operator==(const Bugged& other) const noexcept {
        return (x == other.x) && (isfreex == other.isfreex) && (isfreey == other.isfreey);
    }
};
int main () {
    std::cout << std::boolalpha << (Bugged{} == Bugged{}) << std::endl;
    return 0;
}
-----------------------

The bug only appears if I enable optimization (-02 or -O3), I guess constexpr optimization is not done otherwise.

The problem occurs on the operator==(). The compiler crashes because it can't compute the result of the constexpr. Note that if I change the struct so it doesn't hold booleans, but double instead, or if I remove any of the parentheses term in the operator== function, it doesn't crash.

Note that this code compiles on MSVC 2017 and on clang 6 (though it doesn't mean anything).

I search for a similar bug report but didn't find anything, sorry if this is a duplicate.

-----------------------
Below the output of the command:

Using built-in specs.
COLLECT_GCC=g++
COLLECT_LTO_WRAPPER=/opt/rh/devtoolset-7/root/usr/libexec/gcc/x86_64-redhat-linux/7/lto-wrapper
Target: x86_64-redhat-linux
Configured with: ../configure --enable-bootstrap --enable-languages=c,c++,fortran,lto --prefix=/opt/rh/devtoolset-7/root/usr --mandir=/opt/rh/devtoolset-7/root/usr/share/man --infodir=/opt/rh/devtoolset-7/root/usr/share/info --with-bugurl=http://bugzilla.redhat.com/bugzilla --enable-shared --enable-threads=posix --enable-checking=release --enable-multilib --with-system-zlib --enable-__cxa_atexit --disable-libunwind-exceptions --enable-gnu-unique-object --enable-linker-build-id --with-gcc-major-version-only --enable-plugin --with-linker-hash-style=gnu --enable-initfini-array --with-default-libstdcxx-abi=gcc4-compatible --with-isl=/builddir/build/BUILD/gcc-7.3.1-20180303/obj-x86_64-redhat-linux/isl-install --enable-libmpx --enable-gnu-indirect-function --with-tune=generic --with-arch_32=i686 --build=x86_64-redhat-linux
Thread model: posix
gcc version 7.3.1 20180303 (Red Hat 7.3.1-5) (GCC)
COLLECT_GCC_OPTIONS='-O3' '-Wall' '-Wextra' '-o' 'main' '-save-temps' '-v' '-shared-libgcc' '-mtune=generic' '-march=x86-64'
 /opt/rh/devtoolset-7/root/usr/libexec/gcc/x86_64-redhat-linux/7/cc1plus -E -quiet -v -D_GNU_SOURCE main.cpp -mtune=generic -march=x86-64 -Wall -Wextra -O3 -fpch-preprocess -o main.ii
ignoring nonexistent directory "/opt/rh/devtoolset-7/root/usr/lib/gcc/x86_64-redhat-linux/7/include-fixed"
i...

Created attachment 44539
Small test case

Confirmed. G++ 5 compiles it fine.

struct B {
  double x;
  bool isfreex;
  bool isfreey;

  constexpr bool operator==(const B& other) const noexcept
  {
    return (x == other.x) && (isfreex == other.isfreex) && (isfreey == other.isfreey);
  }

  constexpr bool operator!=(const B& other) const noexcept { return !(*this == other); }
};

int
main ()
{
  bool b = B{} == B{};
}

Started with r230365 -- Merge C++ delayed folding branch..

Another reason why we badly need to finish PR81601.
The constexpr evaluator really can't cope well with these BIT_FIELD_REFs, they read multiple fields at once, or could also read some uninitialized bits in between and mask them off, etc.
So at minimum we'd need to disable these optimizations while folding constexpr declared functions and if the fields don't have INTEGRAL_TYPE or the bit field reference covers more than one field.

GCC 6 branch is being closed

The patch for 89285 seems to have fixed this on trunk.

It seems to be fixed on the last version available on wandbox.org (gcc HEAD 9.0.1 201904): https://wandbox.org/permlink/Tu4T8jEXDDtDw0OS

Though it doesn't work on any other versions (8.3.0...).
Any chance to see the fix in the 8 branch?

Anyway, thanks a lot :)

Author: jakub
Date: Tue Apr 16 19:06:41 2019
New Revision: 270396

URL: https://gcc.gnu.org/viewcvs?rev=270396&root=gcc&view=rev
Log:
 PR c++/86953
 * g++.dg/cpp0x/constexpr-86953.C: New test.

Added:
    trunk/gcc/testsuite/g++.dg/cpp0x/constexpr-86953.C
Modified:
    trunk/gcc/testsuite/ChangeLog

The GCC 7 branch is being closed, re-targeting to GCC 8.4.

GCC 8.4.0 has been released, adjusting target milestone.

The GCC 8 branch is being closed, fixed in GCC 9.1.