ICE when attempting to construct a struct using placement new

[Clint Bland]

- release version
  arm-none-eabi-gcc (GNU Tools for ARM Embedded Processors) 6.2.1 20161205 (release) [ARM/embedded-6-branch revision 243739]

- whether the toolchain was rebuilt or you are using or binary package
  Using the binary package off launchpad ( https://launchpad.net/gcc-arm-embedded )

- host machine
  Linux leaf 4.8.13-1-ARCH #1 SMP PREEMPT Fri Dec 9 07:24:34 CET 2016 x86_64 GNU/Linux

- preprocessed testcase for us to reproduce
  see attachment: `tu_preprocessed.cc`

- symptom(s)
  internal compiler error when attempting to compile the file

- error/warning message encountered
[green@leaf bug]$ arm-none-eabi-g++ -Os -std=c++14 -c tu_preprocessed.cc
tu.cc: In function 'void fn()':
tu.cc:17:6: internal compiler error: in assign_temp, at function.c:961
 void fn (void)
      ^~
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

===================================================================================

Commentary:

* The issue seems to require specifically using placement new:
    replacing the line using placement new with one simply using new will not trigger the ICE.

* The issue requires both a data member AND a destructor to be present. The presence or absence of a constructor does not change whether the problem presents UNLESS the `create` method is available for inlining.

* The issue does not seem to present when using: g++ (GCC) 6.2.1 20160830

Let me know if you need anything else!

[Clint Bland (bland-cr)]

Oh, I forgot one additional thing:

This won't occur with -O0, -O1, -O2, -O3, -O4, -Og but it does occur with -Os.

[Thomas Preud'homme (thomas-preudhomme)]

The bug is due to s_buf not having the same alignment as struct MM. s_buf is 1-byte aligned (because the array element are char) but struct MM is 4-byte aligned because of the int attribute. I agree this should give an error rather than crash. That makes the fix lower priority though.

Best regards.

[Thomas Preud'homme (thomas-preudhomme)]

FYI, this has been raised against latest GCC version:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79085

Best regards.

[Thomas Preud'homme (thomas-preudhomme)]

A patch has been posted that solved the ICE I get when compiling Clint's testcase. Once it gets committed and backported to stable branches of GCC we'll include it in our toolchain.

Best regards.