https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115285
Bug ID: 115285
Summary: std::unordered_set can have duplicate values
Product: gcc
Version: 15.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: libstdc++
Assignee: unassigned at gcc dot gnu.org
Reporter: bugs at phlexo dot com
Target Milestone: ---
When a derived class, which uses its base class' hasher and performs a
conversion during the base class constructor call, is used as the value_type of
a std::unordered_set object, the std::unordered_set will contain duplicates if
it's constructed from iterators of a container holding base class objects that
result in duplicate values after the conversion.
Example:
#include <string>
#include <vector>
#include <unordered_set>
#include <cassert>
class TrimmedStr : public std::string {
static std::string trim_str(std::string const &str) {
auto start = str.find_first_not_of(" \r\n\t");
return start == std::string::npos
? str
: str.substr(start, str.find_last_not_of(" \r\n\t") -
start + 1);
}
public:
TrimmedStr(std::string const &arg) : std::string{trim_str(arg)} {}
TrimmedStr(char const *arg) : TrimmedStr{std::string{arg}} {}
};
namespace std {
template<> struct hash<TrimmedStr> : public std::hash<std::string> {};
}
int main() {
assert(TrimmedStr{"foo "} == std::string{"foo"});
assert(TrimmedStr{" foo"} == std::string{"foo"});
assert(TrimmedStr{" foo "} == std::string{"foo"});
std::unordered_set<TrimmedStr> set_from_initializer_list{"foo", "bar",
" foo ", " bar "};
assert(set_from_initializer_list.size() == 2); //Passes.
std::vector<std::string> args{"foo", "bar", " foo ", " bar "};
std::unordered_set<TrimmedStr> set_from_iterators{args.begin(),
args.end()};
assert(set_from_iterators.size() == 2); //Fails.
}
The final assertion in this code fails with libstdc++ v13 and above, but works
as expected with libstdc++, and libstdc++ < v13.
Bisecting suggests this was introduced in commit
dc9b92facf87a6f2d8b0e5d5fc404f30c3b15a74.
