[LLVMbugs] [Bug 22353] New: __wrap_iter's private constructor runs into a bug with gcc
bugzilla-daemon at llvm.org
bugzilla-daemon at llvm.org
Tue Jan 27 09:32:58 PST 2015
http://llvm.org/bugs/show_bug.cgi?id=22353
Bug ID: 22353
Summary: __wrap_iter's private constructor runs into a bug with
gcc
Product: libc++
Version: unspecified
Hardware: PC
OS: All
Status: NEW
Severity: normal
Priority: P
Component: All Bugs
Assignee: unassignedclangbugs at nondot.org
Reporter: nicolasweber at gmx.de
CC: llvmbugs at cs.uiuc.edu, mclow.lists at gmail.com
Classification: Unclassified
When using libc++ with gcc, this program doesn't compile:
#include <string>
void f(const std::string& s) {
s.begin();
}
#include <vector>
void AppendTo(const std::vector<char>& v) {
v.begin();
}
[reply] [-] Description Nico Weber 2015-01-24 23:40:08 CST
clang accepts the following code, which gcc, cl, icc all reject. icc says
"warning #525: type "A<void>::const_pointer" (declared at line 3) is an
inaccessible type (allowed for cfront compatibility)", so this is probably a
clang bug (?)
$ cat repro.ii
template <class T1> class A;
template <> class A<void> {
typedef void *const_pointer;
};
template <class T1> class A {
void m_fn1(A<void>::const_pointer);
};
class F {
A<int> const_pointer;
};
$ bin/clang -c repro.ii
$ gcc-4.8.1 -c repro.ii
gcc-4.8.1: warning: couldn’t understand kern.osversion ‘14.0.0
repro.ii: In instantiation of ‘class A<int>’:
repro.ii:9:10: required from here
repro.ii:3:17: error: ‘typedef void* A<void>::const_pointer’ is private
typedef void *const_pointer;
^
repro.ii:6:23: error: within this context
void m_fn1(A<void>::const_pointer);
^
[reply] [-] Comment 1 Nico Weber 2015-01-24 23:40:32 CST
Oh, and libc++ seems to depend on this bug :-/
[reply] [-] Comment 2 Nico Weber 2015-01-25 01:09:30 CST
Here's a snippet that doesn't build with gcc and libc++ but does build with
clang and libc++:
#include <string>
void f(const std::string& s) {
s.begin();
}
#include <vector>
void AppendTo(const std::vector<char>& v) {
v.begin();
}
$ bin/clang -c repro.cc -I ~/src/libcxx/include/ -isysroot $(xcrun
-show-sdk-path)
$ gcc-4.8.1 -c repro.cc -I ~/src/libcxx/include/ -isysroot $(xcrun
-show-sdk-path)
gcc-4.8.1: warning: couldn’t understand kern.osversion ‘14.0.0
In file included from /Users/thakis/src/libcxx/include/memory:604:0,
from /Users/thakis/src/libcxx/include/algorithm:628,
from /Users/thakis/src/libcxx/include/string:439,
from repro.cc:1:
/Users/thakis/src/libcxx/include/vector: In instantiation of
‘std::__1::vector<_Tp, _Allocator>::const_iterator std::__1::vector<_Tp,
_Allocator>::__make_iter(std::__1::vector<_Tp, _Allocator>::const_pointer)
const [with _Tp = char; _Allocator = std::__1::allocator<char>;
std::__1::vector<_Tp, _Allocator>::const_iterator = std::__1::__wrap_iter<const
char*>; typename std::__1::__vector_base<_Tp, _Allocator>::const_pointer =
const char*; std::__1::vector<_Tp, _Allocator>::const_pointer = const char*]’:
/Users/thakis/src/libcxx/include/vector:1480:38: required from
‘std::__1::vector<_Tp, _Allocator>::const_iterator std::__1::vector<_Tp,
_Allocator>::begin() const [with _Tp = char; _Allocator =
std::__1::allocator<char>; std::__1::vector<_Tp, _Allocator>::const_iterator =
std::__1::__wrap_iter<const char*>; typename std::__1::__vector_base<_Tp,
_Allocator>::const_pointer = const char*]’
repro.cc:42:11: required from here
/Users/thakis/src/libcxx/include/iterator:1228:31: error:
‘std::__1::__wrap_iter<_Iter>::__wrap_iter(std::__1::__wrap_iter<_Iter>::iterator_type)
[with _Iter = const char*; std::__1::__wrap_iter<_Iter>::iterator_type = const
char*]’ is private
_LIBCPP_INLINE_VISIBILITY __wrap_iter(iterator_type __x) _NOEXCEPT :
__i(__x) {}
^
In file included from repro.cc:7:0:
/Users/thakis/src/libcxx/include/vector:1463:30: error: within this context
return const_iterator(__p);
Depending on how you reduce, this turns up different issues. My best reduction
so far shows a bug in gcc: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64816
(but note point (1.) at the bottom, this kind of suggests that I over-reduced –
might still be the same bug though, so best to see if a fix for that bug helps
here too). But since vector is a friend of __wrap_iter, this is clearly some
gcc bug.
I found a clang bug while reducing this too (issue 22350), but on second look
it appears unrelated.
I have a libc++-side workaround for this that makes everything work with
current gccs. With that workaround, I'm able to build chromium/android with
libc++ and gcc. WIthout the workaround, 2 (out of 20000) translation units fail
to build, on innocuous code.
I'll send out the workaround to the list.
--
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/20150127/86ee447b/attachment.html>
More information about the llvm-bugs
mailing list