From: Jakub Jelinek <jakub@redhat.com>
To: Jason Merrill <jason@redhat.com>
Cc: gcc-patches@gcc.gnu.org
Subject: [PATCH] c++: Fix ICE with __builtin_bit_cast [PR98469]
Date: Wed, 30 Dec 2020 10:13:43 +0100 [thread overview]
Message-ID: <20201230091343.GU3788@tucnak> (raw)
Hi!
On the following testcase we ICE during constexpr evaluation (for warnings),
because the IL has ADDR_EXPR of BIT_CAST_EXPR and ADDR_EXPR case asserts
the result is not a CONSTRUCTOR.
I've tried to force a temporary for those in call.c next to:
if (convs->need_temporary_p
|| TREE_CODE (expr) == CONSTRUCTOR
|| TREE_CODE (expr) == VA_ARG_EXPR)
but that resulted in a lot of ICEs, so this patch just punts on lval
evaluation of BIT_CAST_EXPR instead, normally __builtin_bit_cast is called
from std::bit_cast which is constexpr and therefore the BIT_CAST_EXPR
isn't evaluated there during parsing or tsubst and when evaluating the call
to std::bit_cast the NRV optimized return is assigned to some temporary or
variable and so BIT_CAST_EXPR is not evaluated as lval.
Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
2020-12-30 Jakub Jelinek <jakub@redhat.com>
PR c++/98469
* constexpr.c (cxx_eval_constant_expression) <case BIT_CAST_EXPR>:
Punt if lval is true.
* g++.dg/cpp2a/bit-cast8.C: New test.
* g++.dg/cpp2a/bit-cast9.C: New test.
--- gcc/cp/constexpr.c.jj 2020-12-23 22:44:05.398093175 +0100
+++ gcc/cp/constexpr.c 2020-12-29 10:32:44.865030881 +0100
@@ -6900,6 +6900,15 @@ cxx_eval_constant_expression (const cons
return t;
case BIT_CAST_EXPR:
+ if (lval)
+ {
+ if (!ctx->quiet)
+ error_at (EXPR_LOCATION (t),
+ "address of a call to %qs is not a constant expression",
+ "__builtin_bit_cast");
+ *non_constant_p = true;
+ return t;
+ }
r = cxx_eval_bit_cast (ctx, t, non_constant_p, overflow_p);
break;
--- gcc/testsuite/g++.dg/cpp2a/bit-cast8.C.jj 2020-12-29 10:35:31.547140723 +0100
+++ gcc/testsuite/g++.dg/cpp2a/bit-cast8.C 2020-12-29 10:34:26.431879120 +0100
@@ -0,0 +1,11 @@
+// PR c++/98469
+// { dg-do compile { target c++20 } }
+// { dg-options "-Wall" }
+
+struct S { int s; };
+
+S
+foo ()
+{
+ return __builtin_bit_cast (S, 0);
+}
--- gcc/testsuite/g++.dg/cpp2a/bit-cast9.C.jj 2020-12-29 10:35:35.018101365 +0100
+++ gcc/testsuite/g++.dg/cpp2a/bit-cast9.C 2020-12-29 10:35:05.905431494 +0100
@@ -0,0 +1,15 @@
+// PR c++/98469
+// { dg-do compile { target c++20 } }
+// { dg-options "-Wall" }
+
+template<typename T, typename F>
+constexpr T
+bit_cast (const F &f) noexcept
+{
+ return __builtin_bit_cast (T, f);
+}
+struct S { int s; };
+constexpr int foo (const S &x) { return x.s; }
+constexpr int bar () { return foo (bit_cast<S> (0)); }
+constexpr int x = bar ();
+static_assert (!x);
Jakub
next reply other threads:[~2020-12-30 9:13 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-12-30 9:13 Jakub Jelinek [this message]
2021-01-04 20:44 ` Jason Merrill
2021-01-04 20:48 ` Jakub Jelinek
2021-01-04 21:01 ` Jason Merrill
2021-01-05 15:26 ` [PATCH] c++, v2: " Jakub Jelinek
2021-01-05 16:00 ` Jason Merrill
2021-01-07 21:11 ` Jason Merrill
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20201230091343.GU3788@tucnak \
--to=jakub@redhat.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=jason@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).