[Bug c++/100409] C++ FE elides pure throwing call

[rguenth at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100409

--- Comment #10 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #7)
> Jason, any idea?

diff --git a/gcc/cp/except.c b/gcc/cp/except.c
index a8cea53cf91..cbc94dff790 100644
--- a/gcc/cp/except.c
+++ b/gcc/cp/except.c
@@ -1050,7 +1050,7 @@ check_noexcept_r (tree *tp, int * /*walk_subtrees*/, void
* /*data*/)
       if (concept_check_p (fn))
        return NULL_TREE;
       tree type = TREE_TYPE (fn);
-      gcc_assert (INDIRECT_TYPE_P (type));
+      gcc_assert (INDIRECT_TYPE_P (type) || type == unknown_type_node);
       type = TREE_TYPE (type);

       STRIP_NOPS (fn);
@@ -1073,7 +1073,7 @@ check_noexcept_r (tree *tp, int * /*walk_subtrees*/, void
* /*data*/)
             to evaluate the noexcept-specifier's operand here, but that could
             cause instantiations that would fail.  */
        }
-      if (!TYPE_NOTHROW_P (type))
+      if (type == unknown_type_node || !TYPE_NOTHROW_P (type))
        return fn;
     }


fixes all but the ICE for g++.dg/warn/noeffect4.C where we end up with
an invalid CALL_EXPR:

        arg:0 <sizeof_expr 0x7ffff66ce940 type <integer_type 0x7ffff65617e0
long unsigned int>
            readonly
            arg:0 <call_expr 0x7ffff6550738 type <integer_type 0x7ffff65615e8
int>
                side-effects tree_6
                fn <component_ref 0x7ffff66cf300 type <method_type
0x7ffff66a5f18>


also notice how sizeof() doesn't have side-effects but the call has.  I
suppose that's somehow an "unevaluated" sizeof() expression we don't
expect to hit expr_noexcept_p?  OTOH noexcept(sizeof(foo())) might be
"valid"?

Changing the assert to if (INDIRECT_TYPE_P (type)) conditionalizing the
use of 'type' beyond of course works.  Aka

 diff --git a/gcc/cp/except.c b/gcc/cp/except.c
index a8cea53cf91..623b83b8dc2 100644
--- a/gcc/cp/except.c
+++ b/gcc/cp/except.c
@@ -1050,8 +1050,6 @@ check_noexcept_r (tree *tp, int * /*walk_subtrees*/, void
* /*data*/)
       if (concept_check_p (fn))
        return NULL_TREE;
       tree type = TREE_TYPE (fn);
-      gcc_assert (INDIRECT_TYPE_P (type));
-      type = TREE_TYPE (type);

       STRIP_NOPS (fn);
       if (TREE_CODE (fn) == ADDR_EXPR)
@@ -1073,7 +1071,8 @@ check_noexcept_r (tree *tp, int * /*walk_subtrees*/, void
* /*data*/)
             to evaluate the noexcept-specifier's operand here, but that could
             cause instantiations that would fail.  */
        }
-      if (!TYPE_NOTHROW_P (type))
+      if (!INDIRECT_TYPE_P (type)
+         || !TYPE_NOTHROW_P (TREE_TYPE (type)))
        return fn;
     }

which leaves us with

FAIL: g++.dg/warn/noeffect4.C  -std=gnu++98  (test for warnings, line 84)
FAIL: g++.dg/warn/noeffect4.C  -std=gnu++14  (test for warnings, line 84)
FAIL: g++.dg/warn/noeffect4.C  -std=gnu++17  (test for warnings, line 84)
FAIL: g++.dg/warn/noeffect4.C  -std=gnu++2a  (test for warnings, line 84)

not diagnosing

  sizeof (x++);                // { dg-warning "no effect" }

not sure what should happen to the x++ side-effect (throwing).