Re: [PATCH] eh: Use tree_could_throw_1/lhs_could_trap_p for lhs in non-call exception case [PR127183]

[Richard Biener]

On Wed, Sep 2, 2026 at 8:56 PM Andrea Pinski
<andrew.pin...@oss.qualcomm.com> wrote:
>
> Since `this` argument was added as a non-trapping, we incorrectly had thought
> it would not cause an non-call exception when it was on the LHS.  With the 
> fix for
> PR127133, we can change the places which were checking for trapping on the lhs
> to use lhs_could_trap_p/tree_could_throw_1 instead of using 
> tree_could_throw_p.
> That fixes the non-call exceptions on the lhs of assignments with respect to
> this argument.
>
> Bootstrapped and tested on x86_64-linux-gnu.

OK.

Richard.

>         PR tree-optimization/127183
>
> gcc/ChangeLog:
>
>         * tree-eh.cc (lower_eh_constructs_2): Use lhs_could_trap_p for lhs.
>         (stmt_could_throw_1_p): Use tree_could_throw_1 for lhs.
>         (tree_could_throw_p): Likewise.
>
> gcc/testsuite/ChangeLog:
>
>         * g++.dg/eh/noncall-this-method-1.C: New test.
>
> Signed-off-by: Andrea Pinski <andrew.pin...@oss.qualcomm.com>
> ---
>  .../g++.dg/eh/noncall-this-method-1.C         | 25 +++++++++++++++++++
>  gcc/tree-eh.cc                                |  6 ++---
>  2 files changed, 28 insertions(+), 3 deletions(-)
>  create mode 100644 gcc/testsuite/g++.dg/eh/noncall-this-method-1.C
>
> diff --git a/gcc/testsuite/g++.dg/eh/noncall-this-method-1.C 
> b/gcc/testsuite/g++.dg/eh/noncall-this-method-1.C
> new file mode 100644
> index 00000000000..544bf0396b8
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/eh/noncall-this-method-1.C
> @@ -0,0 +1,25 @@
> +// PR tree-optimization/127183
> +// { dg-do compile }
> +// { dg-options "-O2 -fnon-call-exceptions -fdump-tree-optimized" }
> +
> +struct s1
> +{
> +    int f();
> +    int a;
> +};
> +
> +int s1::f()
> +{
> +    try {
> +      this->a = 1;
> +    }catch(...)
> +    {
> +        __builtin_trap ();
> +    }
> +    return 0;
> +}
> +
> +// An write access to this should be still considered as trapping
> +// and an throwable for non-call exceptions.
> +
> +// { dg-final { scan-tree-dump "__builtin_trap " "optimized" } } */
> diff --git a/gcc/tree-eh.cc b/gcc/tree-eh.cc
> index 2d2d391f0a9..d2c2f3b3157 100644
> --- a/gcc/tree-eh.cc
> +++ b/gcc/tree-eh.cc
> @@ -2118,7 +2118,7 @@ lower_eh_constructs_2 (struct leh_state *state, 
> gimple_stmt_iterator *gsi)
>        if (stmt_could_throw_p (cfun, stmt)
>           && gimple_has_lhs (stmt)
>           && gimple_stmt_may_fallthru (stmt)
> -         && !tree_could_throw_p (gimple_get_lhs (stmt))
> +         && !lhs_could_trap_p (gimple_get_lhs (stmt))
>           && is_gimple_reg_type (TREE_TYPE (gimple_get_lhs (stmt))))
>         {
>           tree lhs = gimple_get_lhs (stmt);
> @@ -3057,7 +3057,7 @@ stmt_could_throw_1_p (gassign *stmt)
>      }
>
>    /* First check the LHS.  */
> -  if (tree_could_trap_p (gimple_assign_lhs (stmt)))
> +  if (tree_could_trap_1 (gimple_assign_lhs (stmt), true))
>      return true;
>
>    /* Check if the main expression may trap.  */
> @@ -3144,7 +3144,7 @@ tree_could_throw_p (tree t)
>    if (TREE_CODE (t) == MODIFY_EXPR)
>      {
>        if (cfun->can_throw_non_call_exceptions
> -          && tree_could_trap_p (TREE_OPERAND (t, 0)))
> +         && tree_could_trap_1 (TREE_OPERAND (t, 0), true))
>          return true;
>        t = TREE_OPERAND (t, 1);
>      }
> --
> 2.43.0
>