<html>
<head>
<base href="https://bugs.llvm.org/">
</head>
<body><table border="1" cellspacing="0" cellpadding="8">
<tr>
<th>Bug ID</th>
<td><a class="bz_bug_link
bz_status_NEW "
title="NEW - std::basic_string misbehavior when sizeof(CharT) > 8"
href="https://bugs.llvm.org/show_bug.cgi?id=51816">51816</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>std::basic_string misbehavior when sizeof(CharT) > 8
</td>
</tr>
<tr>
<th>Product</th>
<td>libc++
</td>
</tr>
<tr>
<th>Version</th>
<td>unspecified
</td>
</tr>
<tr>
<th>Hardware</th>
<td>PC
</td>
</tr>
<tr>
<th>OS</th>
<td>Windows NT
</td>
</tr>
<tr>
<th>Status</th>
<td>NEW
</td>
</tr>
<tr>
<th>Severity</th>
<td>enhancement
</td>
</tr>
<tr>
<th>Priority</th>
<td>P
</td>
</tr>
<tr>
<th>Component</th>
<td>All Bugs
</td>
</tr>
<tr>
<th>Assignee</th>
<td>unassignedclangbugs@nondot.org
</td>
</tr>
<tr>
<th>Reporter</th>
<td>georgthegreat@gmail.com
</td>
</tr>
<tr>
<th>CC</th>
<td>llvm-bugs@lists.llvm.org, mclow.lists@gmail.com
</td>
</tr></table>
<p>
<div>
<pre>Created <span class=""><a href="attachment.cgi?id=25247" name="attach_25247" title="Test using gtest">attachment 25247</a> <a href="attachment.cgi?id=25247&action=edit" title="Test using gtest">[details]</a></span>
Test using gtest
An attempt to compile std::basic_string with various weird character types
would lead to a multiple issues.
So far I have found the following:
1. stack allocation size would grow due to overallocation, i. e:
if constexpr (sizeof(Char) == 12) {
EXPECT_EQ(sizeof(s), 40u);
}
2. allocations would become power-of-two-exponential, i. e:
for (size_t i = 0; i < 47u; ++i) {
s.push_back({});
}
if constexpr (sizeof(Char) <= 4) {
// OK, do not request power-of-two blocks from allocator
EXPECT_EQ(
s.capacity(),
// capacity grows exponentially starting from 22 bytes
47u
);
} else if constexpr (sizeof(Char) == 8) {
// Bad
EXPECT_EQ(
s.capacity(),
// capacity grows exponentially as the power-of-two
63u
);
}
3. Memory leak is detected by ASan when both -fsized-deallocation is enabled
and stable ABI is used (i. e. _LIBCPP_ABI_VERSION == 2):
==1260518==ERROR: AddressSanitizer: new-delete-type-mismatch on 0x6080000004a0
in thread T0:
object passed to delete has wrong type:
size of the allocated type: 84 bytes;
size of the deallocated type: 72 bytes.
#0 0x4cdc02 in void std::__y1::__libcpp_operator_delete<void*, unsigned
long>(void*, unsigned long)
/home/thegeorg/arcadia/contrib/libs/cxxsupp/libcxx/include/new:245:3
#1 0x4cdc02 in void std::__y1::__do_deallocate_handle_size<>(void*,
unsigned long) include/new:271:10
#2 0x4cdc02 in std::__y1::__libcpp_deallocate(void*, unsigned long,
unsigned long) include/new:285:14
#3 0x4cdc02 in std::__y1::allocator<WeirdChar12>::deallocate(WeirdChar12*,
unsigned long) include/memory:850:13
#4 0x4cdc02 in std::__y1::allocator_traits<Allocator<WeirdChar12>
<span class="quote">>::deallocate(Allocator<WeirdChar12>&, WeirdChar12*, unsigned long)</span >
include/__memory/allocator_traits.h:484:14
This is caused by incorrect alignment in __recommend() and attempt to allocate
odd number of characters. String capacity() must be even as odd bit will be
overwritten by __long_mask.
Though C++ standard does not limit sizeof(CharT), I suggest
static_assert(sizeof(CharT) <= 4) to be added into libc++.
Though this could break existing codebases, I consider the break fast to be the
better option.
I do not think SSO is anyway helpful whenever sizeof(CharT) >= 8. Maybe we
should ensure that std::basic_strin is always long is such cases.</pre>
</div>
</p>
<hr>
<span>You are receiving this mail because:</span>
<ul>
<li>You are on the CC list for the bug.</li>
</ul>
</body>
</html>