[llvm-bugs] [Bug 51260] New: Incomplete implementation of the P1330R0 proposal / [expr.const]p5 and [class.union]p6 (N4892)
via llvm-bugs
llvm-bugs at lists.llvm.org
Wed Jul 28 20:46:08 PDT 2021
https://bugs.llvm.org/show_bug.cgi?id=51260
Bug ID: 51260
Summary: Incomplete implementation of the P1330R0 proposal /
[expr.const]p5 and [class.union]p6 (N4892)
Product: clang
Version: trunk
Hardware: PC
OS: Linux
Status: NEW
Severity: enhancement
Priority: P
Component: C++
Assignee: unassignedclangbugs at nondot.org
Reporter: matthewjbarichello at gmail.com
CC: blitzrakete at gmail.com, dgregor at apple.com,
erik.pilkington at gmail.com, llvm-bugs at lists.llvm.org,
richard-llvm at metafoo.co.uk
Created attachment 25087
--> https://bugs.llvm.org/attachment.cgi?id=25087&action=edit
Patch to fix the issue in 'releases/12.x'
All of the references and quotes in this report are concerned with the changes
in P1330R0 and the current working standard document, N4892:
- P1330R0: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2018/p1330r0.pdf
- N4892: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2021/n4892.pdf
Affected versions:
- trunk
- 12.0.1
- 12.0.0
- 11.0.1
- 11.0.0
- 10.0.1
- 10.0.0
- 9.0.1
- 9.0.0
Driver cmdline:
Note: For versions not supporting -std=c++20, -std=c++2a was used.
clang++ -std=c++20
Consider the following examples:
[Example A:
struct A {};
constexpr auto test() {
union U {
int i;
A a;
} u{.i = 0};
u.a = {};
return 0;
}
int main() {
constexpr auto r = test();
(void)r;
}
[Example B:
struct A {
int i;
constexpr A() noexcept : i(0) {}
constexpr A(int i) noexcept : i(i) {}
constexpr A& operator=(A const& other) noexcept {
i = other.i;
return *this;
}
};
constexpr auto test() {
union U {
int i;
A a;
} u{.i = 0};
u.a = {123};
return u.a.i;
}
int main() {
constexpr auto r = test();
(void)r;
}
According to [expr.constant]p5
An expression E is a core constant expression unless the evaluation of E,
following the rules of the abstract machine (6.9.1), would evaluate one of the
following:
... an invocation of an implicitly-defined copy/move constructor or
copy/move assignment operator for a union whose active member (if any) is
mutable, unless the lifetime of the union object began within the evaluation of
E ...
, and [class.union]p6
When the left operand of an assignment operator involves a member access
expression (7.6.1.5) that nominates a union member, it may begin the lifetime
of that union member ...
, the lifetime of `u.a` ("Example A":8) should start at the invocation of
`A`'s implicitly defined copy-assignment operator, however, under all listed
clang versions this is not the case as compilation yields the following error:
"Example A":13:17: error: constexpr variable 'r' must be initialized by a
constant expression
constexpr auto r = test();
^ ~~~~~~
"Example A":8:6: note: member call on member 'a' of union with active member
'i' is not allowed in a constant expression
u.a = {};
^
"Example A":13:21: note: in call to 'test()'
constexpr auto r = test();
^
1 error generated.
Likewise, the lifetime of `u.a` ("Example B":18) should begin at the
invocation of `A`'s user-defined constexpr copy-assignment operator, however
compilation yields the same error.
Addressing the issue:
I plan to formally submit a properly tested patch to trunk and backport it to
all affected major clang releases. Until then, I've attached a patch to address
this bug, based on the `release/12.x` tag:
commit fed41342a82f5a3a9201819a82bf7a48313e296b (grafted, HEAD ->
release/12.x, tag: llvmorg-12.0.1-rc4, tag: llvmorg-12.0.1,
origin/release/12.x)
Author: Tom Stellard <tstellar at redhat.com>
Date: Mon Jun 28 09:23:38 2021 -0700
Revert "Revert "[Coverage] Fix branch coverage merging in
FunctionCoverageSummary::get() for instantiation""
This reverts commit 33d312b2d731507327252fd597bac1b738870330.
The original patch was correct, so we need to restore it in the
release branch.
Notes:
- In testing both examples against other major compilers, such as MSVC and
GCC, it would seem that they compile and behave as expected.
--
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/20210729/0e2ffc57/attachment-0001.html>
More information about the llvm-bugs
mailing list