[Bug c/86765] New: invalid -Wmisleading-indentation for double macro expansion with pragmas

[ndesaulniers at google dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86765

            Bug ID: 86765
           Summary: invalid -Wmisleading-indentation for double macro
                    expansion with pragmas
           Product: gcc
           Version: 8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ndesaulniers at google dot com
  Target Milestone: ---

Developing a patch for the Linux kernel, I hit a -Wmisleading-indentation
warning for a particular configuration in gcc 6, 7, and 8.

This is the paired down test case: https://godbolt.org/g/b5dsqH

Looking at the output, the warning is technically correct (puts() was called
unconditionally), but I don't think the compiled code is.  Compared to clang,
clang compiles the conditional check (as expected), but gcc does not.

The kernel patch is trying to add compiler specific pragma's to ignore
clang-7's -Wreturn-stack-address.  We could be better about not emitted
_Pragma's for push/pop if we're not clang, but this still seems like a
miscompilation from gcc to me.

Is there a special rule about GNU statement expressions [0] that I'm missing?

[0] https://gcc.gnu.org/onlinedocs/gcc/Statement-Exprs.html