[llvm-bugs] [Bug 36508] New: MSVC: noexcept aborts with longjmp

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=36508

            Bug ID: 36508
           Summary: MSVC: noexcept aborts with longjmp
           Product: clang
           Version: unspecified
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: LLVM Codegen
          Assignee: unassignedclangbugs at nondot.org
          Reporter: alex at crichton.co
                CC: llvm-bugs at lists.llvm.org, rnk at google.com

This is a sort of roundabout bug with respect to where it originated from, but
the gist of it is that this program:





#include <assert.h>
#include <setjmp.h>
#include <stdio.h>

static jmp_buf ENV;

static void test_end() {
  longjmp(ENV, 1);
  assert(0);
}

static void test_middle() noexcept {
  test_end();
}

static void test_start() {
  if (setjmp(ENV) != 0)
    return;
  test_middle();
  assert(0);
}

int main() {
  test_start();
  printf("good\n");
  return 0;
}




will abort when compiled with Clang and succeed when compiled with cl.exe. More
specifically when compiled with `cl.exe foo.cpp /EHsc` the program will succeed
and print "good", but when compiled with `clang-cl.exe foo.cpp /EHsc` it will
instead abort.

I believe the issue here is that the `test_middle` function, declared with
`noexcept`, will abort thinking that it threw an exception. It appears that on
MSVC the `longjmp` function is implemented with an exception-like mechanism
which I believe triggers the landing pad generated by `test_middle` to abort
when an exception is thrown.

This bug actually came from an upstream Rust bug [1] which has quite a long
thread! The gist of it is that in Rust 1.24.0 we effectively turned on
`noexcept` for a certain class of Rust functions by default. This change
regressed on MSVC for us in that it broke users who wanted to longjmp across
Rust frames back to C (for example with other languages like Lua and Ruby).

I did some investigation to see what Clang did (as right now Rust's
exception-related MSVC codegen attempts to closely match what Clang does) and I
figured y'all may be interested in this as well! This may or may not be a bug
in
Clang (I'm not sure), so feel free to close if this isn't relevant for clang.

[1]: https://github.com/rust-lang/rust/issues/48251

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20180225/667e445b/attachment.html>