Loss of idempotence of user type declarationd post v3.2.3

[ormshaw]

I am working on Fedora 37 using gcc 12.3.1 and lua 5.4.6. The following code works as expected with sol v 3.2.3.

#include "TestClass.h"

#define SOL_ALL_SAFETIES_ON 1
#include "sol/sol.hpp"

#include
#include

int main()
{
std::string script
{
R"(
print(testClass)
testClass:greet()
)"
};

try
{
    sol::state lua;
    lua.open_libraries();

    for(int i{0}; i<16; ++i)
    {
        sol::usertype<TestClass> luaUserType{lua.new_usertype<TestClass>("TestClass")};
        luaUserType.set("TestClass", sol::constructors<TestClass()>());
        luaUserType.set("greet", &TestClass::greet);

        TestClass testClass{};
        lua["testClass"] = &testClass;

        auto result{lua.safe_script(script)};
    }
}
catch( const sol::error& e )
{
    std::cout << "An unexpected error has occurred: " << e.what() << std::endl;
}

}

The output from this code is as follows:
sol.TestClass*: 0x2167dc8
Hello World!!!
userdata: 0x2169f98
Hello World!!!
userdata: 0x216b188
Hello World!!!
userdata: 0x216c378
Hello World!!!
userdata: 0x216d568
Hello World!!!
userdata: 0x21678d8
Hello World!!!
userdata: 0x2169a18
Hello World!!!
userdata: 0x216adf8
Hello World!!!
userdata: 0x216df18
Hello World!!!
userdata: 0x216eff8
Hello World!!!
userdata: 0x2170188
Hello World!!!
userdata: 0x2171378
Hello World!!!
userdata: 0x2172568
Hello World!!!
userdata: 0x216d088
Hello World!!!
userdata: 0x216c378
Hello World!!!
userdata: 0x2161cd8
Hello World!!!

If I change to using sol 3.3.x or 4.x the out put from the program is:

/home/ian/projects/lua/sol/build/source/idempotence> /home/ian/projects/lua/sol/build/source/idempotence/indempotence
sol.TestClass*: 0x7e2cd8
Hello World!!!
sol.TestClass*: 0x7e5718
Hello World!!!
sol.TestClass*: 0x7e7178
Hello World!!!
sol.TestClass*: 0x7e2998
An unexpected error has occurred: sol: runtime error: [string "..."]:3: attempt to index a sol.TestClass* value (global 'testClass')
stack traceback:
[string "..."]:3: in main chunk
[sol2] An error occurred and has been passed to an error handler: sol: runtime error: [string "..."]:3: attempt to index a sol.TestClass* value (global 'testClass')
stack traceback:
[string "..."]:3: in main chunk

I have seen similar code failing the first time the user type was re-declared. An other instance it was after 10 re-declarations.

My expectations is that the example code should work regardless of the number of re-declarations. I.e. it should be idempotent.

[ormshaw]

The same results are seen in Fedora 38 with gcc 13.2.1

[congard]

Creating multiple non-shared environments (i.e., without a parent) and then creating new usertypes within them results in quite similar behavior

[devgs]

Duplicate of #1492