some new intel on the lobby bug

[rootbeer23]

i have run FA.exe in a debugger and got some results:

from the segfault report that is printed when fa crashes:
0x0048169b - this is the stack frame in which the bug happens
0x0047fbe0 - this is the function that calls the above function in its body
0x0047fe54 - this is the location in 0x0047fbe0 where 0x0048169b was called

0047fbe0 is only executed when someone joins the game, so it was triggered maybe a dozen times during the test.
(possibly also called when i kicked some1, need to confirm that)
0048169b was called 2 times during my test run, both when the game was already almost full, so it was a while
before it was triggered the first time.
the second call to that function triggered the segfault.
all calls to 0047fbe0 were in the "Main Thread"

so now it looks that the bug is triggered in a piece of code that is run when someone joins the game if a certain condition is met.
so the good news is that the code section containing the bug is not generic, like for example a part of the lua interpreter but only called in certain conditions, i.e. someone joining (possibly also leaving).
it would be nice to be able to do a lua backtrace when one of the breakpoints is hit. any idea how to do that?

[Ze_PilOt]

Well, the harder part will be to replicate it for sure to know the exact conditions.

There are some few things to take in consideration :
- If the player did some alt-tabbing (FA does NOT like that)
- If the player joining made lobby "hanging" (The lobby "freeze" when negotiating the connection, can take some seconds due to some firewall rejecting the first packet or something I guess).

Supreme commander had a debug mode, but not FA sadly.

[Adraius]

- If the player did some alt-tabbing (FA does NOT like that)

I didn't know that. Should I avoid alt-tabbing in the future, or is it not that big of a concern?

[rootbeer23]

0x0048169b - this is the stack frame in which the bug happens
0x0047fbe0 - this is the function that calls the above function in its body
0x0047fe54 - this is the location in 0x0047fbe0 where 0x0048169b was called

ok, furthermore:
0x0047fbe0 is called when someone joins or leaves but not when kicked. tested twice.

[Ze_PilOt]

- If the player did some alt-tabbing (FA does NOT like that)

I didn't know that. Should I avoid alt-tabbing in the future, or is it not that big of a concern?

You should avoid it.

One of my crazy theory is that in GPGNET, the bug was there too, but people were alt tabbing way less because the chat was not available after you host game.

[Softly]

alt-tabbing?

Can we test this because this is a regular for me and I'm not sure that it contributes to crashes

[Ze_PilOt]

It sure doesn't help at all.

While a connection is in progress, you can see that your lobby does a mini-freeze (more obvious when you connect : The cancel button stop making sounds).

If you alt-tab during these, there is a fair chance that you can cause a crash/grey screen/whatever.

[rootbeer23]

is there a lua function that aborts FA.exe with a printout of the lua backtrace?
If there is, i can catch the abort with a breakpoint and from the stacktrace i can find out which function to call
to get a lua backtrace.
also if there is a way to generate any kind of backtrace with a single lua function, i can break at ntdll:FileWrite and
find out which address gets called during a lua stackdump.
if i can insert a call to generate a backtrace when that specific pointer is zero, that would give a lua backtrace at the moment the segfault would occur.

[Adraius]

I ask because I (used to) alt-tab a LOT. I never noticed any problems I could link back to it; occasionally I have nil to somebody, but that could be anything. The only thing I did concretely notice is if I alt-tab at the right moment as a replay is loading, when I return to the 'loading' replay, the screen will go black and only show my mouse, and I can't alt-tab back out. Not sure if that's new info for you.

[Ze_PilOt]

It won't produce nil or anything.

But it can cause your client not sending a command or failing to answer to one, causing the whole system to crash.
Probably randomly.

Well, test to host a game with players that wont alt-tab EVER (put a password to filter who is coming), and see if it crashes ?

For your other issues, I'm not kidding, FA doesn't like alt-tab. So, don't