So the x86 assertion failure I fixed yesterday where I didn't have a testcase? Martin L has just tripped over it and filed a bug for it overnight.
I'm adding his testcase to the C++ regression testsuite. I also retro-actively added the PR marker to the x86 commit which fixes this problem.
Jeff
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog index 4cc2cedc0dc..0a08fe2ed5c 100644 --- a/gcc/testsuite/ChangeLog +++ b/gcc/testsuite/ChangeLog @@ -1,5 +1,8 @@ 2017-11-03 Jeff Law <l...@redhat.com> + PR target/82823 + * g++.dg/torture/pr82823.C: New test. + * gcc.target/i386/stack-check-12.c: New test. 2017-11-03 Jakub Jelinek <ja...@redhat.com> diff --git a/gcc/testsuite/g++.dg/torture/pr82823.C b/gcc/testsuite/g++.dg/torture/pr82823.C new file mode 100644 index 00000000000..dab369e7ad3 --- /dev/null +++ b/gcc/testsuite/g++.dg/torture/pr82823.C @@ -0,0 +1,26 @@ +// { dg-do compile } +// { dg-additional-options "-fstack-clash-protection" } +// { dg-require-effective-target supports_stack_clash_protection } + + +class a +{ +public: + ~a (); + int b; +}; +class c +{ +public: + a m_fn1 (); +}; +class d +{ + int e (); + c f; +}; +int +d::e () +{ + return f.m_fn1 ().b; +}