This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[PATCH, C++, PR58282] Handle noexcept on transactions with -fno-exceptions
- From: Tom de Vries <Tom_deVries at mentor dot com>
- To: Jason Merrill <jason at redhat dot com>
- Cc: "gcc-patches at gcc dot gnu dot org" <gcc-patches at gcc dot gnu dot org>, Torvald Riegel <triegel at redhat dot com>
- Date: Tue, 3 Sep 2013 12:03:19 +0200
- Subject: [PATCH, C++, PR58282] Handle noexcept on transactions with -fno-exceptions
- Authentication-results: sourceware.org; auth=none
Jason,
Consider this testcase reduced from g++.dg/tm/noexcept-1.C:
...
struct TrueFalse
{
static constexpr bool v() { return true; }
};
int global;
template<typename T> int foo()
{
__transaction_atomic noexcept(T::v()) { global += 1; }
return __transaction_atomic noexcept(T::v()) (global + 2);
}
int f1()
{
return foo<TrueFalse>();
}
...
It fails with a SIGSEGV when run with -fno-exceptions:
...
$ g++ noexcept-1.C -fno-exceptions -fgnu-tm -std=c++0x -S
noexcept-1.C: In function ‘int foo() [with T = TrueFalse]’:
noexcept-1.C:10:43: internal compiler error: Segmentation fault
__transaction_atomic noexcept(T::v()) { global += 1; }
^
...
The noexcept(t::v()) operator evaluates to true, so the noexcept specifier on
the transaction is set to true. Then cp/semantics.c:build_transaction_expr
builds a MUST_NOT_THROW_EXPR to safe-guard the property at runtime:
...
if (noex)
{
expr = build_must_not_throw_expr (expr, noex);
SET_EXPR_LOCATION (expr, loc);
TREE_SIDE_EFFECTS (expr) = 1;
}
...
During gimplification of the MUST_NOT_THROW_EXPR in
cp/cp-gimplify.c:gimplify_must_not_throw_expr, we're trying to translate into a
try/catch:
...
gimplify_and_add (body, &try_);
mnt = gimple_build_eh_must_not_throw (terminate_node);
gimple_seq_add_stmt_without_update (&catch_, mnt);
mnt = gimple_build_try (try_, catch_, GIMPLE_TRY_CATCH);
...
However, terminate_node is NULL at that point because of -fno-exceptions, so in
gimple.c:gimple_build_eh_must_not_throw, this line causes a SIGSEGV because decl
== terminate_node == NULL:
...
gcc_assert (TREE_CODE (decl) == FUNCTION_DECL);
...
This patch fixes the SIGSEGV by not generating a MUST_NOT_THROW_EXPR with
-fno-exceptions.
Bootstrapped and reg-tested on x86_64.
OK for trunk?
Thanks,
- Tom
2013-09-02 Tom de Vries <tom@codesourcery.com>
PR c++/58282
* semantics.c (finish_transaction_stmt, build_transaction_expr): Handle
flag_exceptions.
diff --git a/gcc/cp/semantics.c b/gcc/cp/semantics.c
index ee3503c..c8b328c 100644
--- a/gcc/cp/semantics.c
+++ b/gcc/cp/semantics.c
@@ -5189,7 +5189,7 @@ finish_transaction_stmt (tree stmt, tree compound_stmt, int flags, tree noex)
/* noexcept specifications are not allowed for function transactions. */
gcc_assert (!(noex && compound_stmt));
- if (noex)
+ if (noex && flag_exceptions)
{
tree body = build_must_not_throw_expr (TRANSACTION_EXPR_BODY (stmt),
noex);
@@ -5210,7 +5210,7 @@ tree
build_transaction_expr (location_t loc, tree expr, int flags, tree noex)
{
tree ret;
- if (noex)
+ if (noex && flag_exceptions)
{
expr = build_must_not_throw_expr (expr, noex);
SET_EXPR_LOCATION (expr, loc);