Re: Upgrading Lua 5.4.2 to 5.4.3+ in a C++ project leads to PANIC

[Aleksandr Zenkov <webmaster@...>]

Ofcourse, how could I possibly forget about compiling my project in 
debug mode and getting a backtrace, getting old I guess. Thank you for 
the hint!

I've got a first hint from the backtrace. From here I need to explain 
how my application works..

As I said earlier, it is a TCP server which accepts clients that use a 
specific protocol to exchange data. This server has different plugins 
with registered callbacks, one such example callback is simply 
"OnUserLogin(name, addr)" which is called in all loaded plugins once a 
user logs in. Then there is a Lua plugin among the others, it loads Lua 
scripts that allow to receive callbacks inside the script, aswell as 
call registered functions that are located in separate table, one such 
example is "VH:GetUserIP(name)". This way the server can communicate 
with the Lua script and reverse.

Now to the actual issue that I just figured out..

The loaded Lua scripts are freely coded, some scripts use "VH:GetUserIP" 
helper, others don't. Some scripts actually use the callback 
"OnUserLogin" mentioned above,others don't. Previously in Lua 5.4.2 and 
earlier, a Lua script didn't have to actually have this function written 
in the code, in order to receive the callback, as I said, not all 
scripts require that specific callback, because they are written 
according the needs of the script.

My sample script that was crashing the server, didn't have the function 
"OnUserLogin", because it doesn't need it, it works with other type of 
callbacks instead. Once I added "OnUserLogin" function to my script, the 
server is no longer crashing.

So my next question is.. What type of security change was made in Lua 
5.4.3 to start panicking once a C++ code tried to call a piece of Lua 
code that does not exist? I mean, such a security checks never existed 
in Lua as far as I can remember. As of my server, it was always up to 
end user to create his own scripts as he wanted them to work. Ofcourse 
there are error captures on luaL_dofile and lua_pcall that are returned 
to the server to notify the user about possible issues. But this issues 
does crash whole server.

I hope I described everything as expected, and I'm really glad that we 
got one step closer to the solution. Also I see forward to some useful 
information that could get my issue resolved.

Regards.

On 2022-12-02 14:53, Roberto Ierusalimschy wrote:
> 1. Can you try bil til's suggestion? "set a breakpoint in the lua panic
> function and look at your call stack"
>
> 2. You could compile Lua with the option LUA_USE_APICHECK and see if
> semething pops up.
>
> -- Roberto