std::map is broken

[stevenskevin]

std::map is broken when compiling with ko-clang++. It works when using regular clang++.

Here's my test case ("test.cpp"). It sets up a std::map<std::string, std::string> and adds items to it one by one. At every step, it iterates over the whole map to print its contents, and also searches for a specific item with .find().

#include <cstdio>
#include <map>
#include <string>

typedef std::map<std::string, std::string> string_map_t;

void search_for_target(string_map_t &string_map) {
    printf("Current entries:");
    for (auto it = string_map.begin(); it != string_map.end(); ++it) {
        printf("  %s", it->first.c_str());
    }

    printf("\nSearching for \"key_target\": ");
    auto found = string_map.find("key_target");
    if (found != string_map.end()) {
        printf("Found! \"%s\"\n", found->second.c_str());
    } else {
        printf("Not found...\n");
    }
}

int main(int argc, const char *argv[]) {
    string_map_t string_map = {};
    search_for_target(string_map);

    string_map["key_A"] = "value_A";
    search_for_target(string_map);

    string_map["key_B"] = "value_B";
    search_for_target(string_map);

    string_map["key_target"] = "value_target";
    search_for_target(string_map);

    string_map["key_C"] = "value_C";
    search_for_target(string_map);

    string_map["key_D"] = "value_D";
    search_for_target(string_map);

    return 0;
}

Compile with standard clang++ and ko-clang++ commands (you can also optionally add -stdlib=libc++ -lc++ -lc++abi to the clang++ call to make it more like the ko-clang++ one, but it doesn't affect the overall outcome):

$ clang++ -o test test.cpp
$ ko-clang++ -o test_ko test.cpp

Output when compiled with regular clang++ (correct): The map grows as expected, and once "key_target" is added, .find() starts finding it:

$ ./test
Current entries:
Searching for "key_target": Not found...
Current entries:  key_A
Searching for "key_target": Not found...
Current entries:  key_A  key_B
Searching for "key_target": Not found...
Current entries:  key_A  key_B  key_target
Searching for "key_target": Found! "value_target"
Current entries:  key_A  key_B  key_C  key_target
Searching for "key_target": Found! "value_target"
Current entries:  key_A  key_B  key_C  key_D  key_target
Searching for "key_target": Found! "value_target"

Output when compiled with ko-clang++ (incorrect): Once an item is added to the map, iteration and searching both break. Iterators only ever yield the first item, and .find() always returns the most recently added item.

$ ./test_ko
==27699==WARNING: DataFlowSanitizer: call to uninstrumented function pthread_cond_broadcast
==27699==WARNING: DataFlowSanitizer: call to uninstrumented function pthread_cond_broadcast
==27699==WARNING: DataFlowSanitizer: call to uninstrumented function pthread_cond_broadcast
==27699==WARNING: DataFlowSanitizer: call to uninstrumented function pthread_cond_broadcast
==27699==WARNING: DataFlowSanitizer: call to uninstrumented function mbtowc
==27699==WARNING: DataFlowSanitizer: call to uninstrumented function __ctype_get_mb_cur_max
==27699==WARNING: DataFlowSanitizer: call to uninstrumented function newlocale
==27699==WARNING: DataFlowSanitizer: call to uninstrumented function pthread_cond_broadcast
==27699==WARNING: DataFlowSanitizer: call to uninstrumented function pthread_cond_broadcast
==27699==WARNING: DataFlowSanitizer: call to uninstrumented function pthread_cond_broadcast
==27699==WARNING: DataFlowSanitizer: call to uninstrumented function mbtowc
==27699==WARNING: DataFlowSanitizer: call to uninstrumented function __ctype_get_mb_cur_max
Current entries:
Searching for "key_target": Not found...
Current entries:  key_A
Searching for "key_target": Found! "value_A"
Current entries:  key_A
Searching for "key_target": Found! "value_B"
Current entries:  key_A
Searching for "key_target": Found! "value_target"
Current entries:  key_A
Searching for "key_target": Found! "value_C"
Current entries:  key_A
Searching for "key_target": Found! "value_D"
==27699==WARNING: DataFlowSanitizer: call to uninstrumented function wcrtomb
==27699==WARNING: DataFlowSanitizer: call to uninstrumented function wcrtomb
==27699==WARNING: DataFlowSanitizer: call to uninstrumented function wcrtomb
==27699==WARNING: DataFlowSanitizer: call to uninstrumented function wcrtomb

My environment, if it helps: LLVM 6.0.1 in an Ubuntu 18.04 Docker container.