problem internal to GDB has been detected and debugging session dies in multi-thread environment

Question

My test code is like this:

#include <iostream>
#include <string>
#include <thread>
#include <unistd.h>

using namespace std;

class ns_string {
public:
    ns_string(string s) : _s(s) {}
    static ns_string make_ns_string(const string &s) {
        return ns_string(s);
    }
    string ns() const { return _s; }
private:
    string _s;
};


void f2(string s) {
    cout << "f2:" << s << endl;
}
void thread_func() {
    while (1) {
        cout << "------------------ another thread -> ";
        f2("thread_func");
        sleep(5);
    }
}

void f(int i, const ns_string& name) {
    cout << i << ", " << name.ns() << endl;
    f2(name.ns());
}

int main(int argc,char* argv[]) {
    thread t(thread_func);

    for (int i=0; i<10; i++) {
        f(i, ns_string::make_ns_string(std::to_string(i)));
    }

    t.join();
    return 0;
}

I want to set a breakpoint when f2 parameter s == "4".
I do that via b f2 if strcmp(s.c_str(), "4") == 0.

Here're three cases that I experiment.

GDB version is 11.1

case 1

DONOT start thread_func. Everything is ok. The program stops when s=="4"

(gdb) r
0, 0
f2:0
1, 1
f2:1
2, 2
f2:2
3, 3
f2:3
4, 4

Breakpoint 1, f2 (s="4") at gdb-crash.cc:21
21      cout << "f2:" << s << endl;

case 2

Start thread_func, but DONOT call f2. gdb outputs:

(gdb) r
[New Thread 0xffffbf17c1c0 (LWP 20299)]
------------------ another thread -> 0, 0
Error in testing breakpoint condition:
Unable to fetch general registers.: No such process.
An error occurred while in a function called from GDB.
Evaluation of the expression containing the function
(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >::c_str() const) will be abandoned.
When the function is done executing, GDB will silently stop.
Selected thread is running.
(gdb) c
Continuing.
f2:0
1, 1
f2:1
2, 2
f2:2
3, 3
f2:3
4, 4

Thread 1 "a.out" hit Breakpoint 1, f2 (s="4") at gdb-crash.cc:21
21      cout << "f2:" << s << endl;

However I try to continue, it still can stop at s=="4.

case 3

Start the thread_func, and call f2

Program die. gdb outputs:

Error in testing breakpoint condition:
The program stopped in another thread while making a function call from GDB.
Evaluation of the expression containing the function
(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >::c_str() const) will be abandoned.
When the function is done executing, GDB will silently stop.
[Switching to Thread 0x7ffff7a47740 (LWP 1832)]

Thread 1 "a.out" hit Breakpoint 1, f2 (s="0") at a.cc:30
30          cout << "f2:" << s << endl;
Missing separate debuginfos, use: dnf debuginfo-install libgcc-11.2.1-9.fc34.x86_64 libstdc++-11.2.1-9.fc34.x86_64
../../gdb/infcall.c:1284: internal-error: value* call_function_by_hand_dummy(value*, type*, gdb::array_view<value*>, void (*)(void*, int), void*): Assertion `call_thread->thread_fsm == sm' failed.
A problem internal to GDB has been detected,
further debugging may prove unreliable.

Obviously different threads access f2, which confuses gdb for some reason.

As a debugger, I don't care about which thread stops at that breakpoint. I only care about gdb makes it stop when my condition is met. How should I acheive that in multi-thread situation ?

Answer 1

This is a bug in GDB. Reproduced using current GDB trunk and GDB-10.0.

Assuming you are using g++ to build the code, here is a workaround:

(gdb) b f2 if s._M_dataplus._M_p[0] == '4' && s._M_dataplus._M_p[1] == '\0'

The most likely reason for the bug is that calling s.c_str() requires that the inferior (being debugged) program be resumed, and while it's running to evaluate s.c_str() another thread hits the breakpoint, causing GDB to become confused.

Similar (but not identical) bug.