[kovarex] [1.1.6] MP BP-library related crash. The stacktrace. Each time.

[KeepResearchinSpoons]

TL;DR:

Not a mission critical level. I am NOT locked out of the game per se.

So. I have a book. The book is in my lib. I join a MP game.
I get the book to my slot 3 or slot 2 or whatever. I take the book into the cursor.
I try to scroll it in a cursor with a wheel and it crashes.


I can reproduce it each time if you need more stacktraces.
First crash happened in a scenario-map with no mods. The later crashes just need a MP server.
I CANT reproduce this in the single player mode.
I CANT reproduce this in the host-savefile mode.

Fun fact is that If i "scroll" with a keyboard arrows alias it works fine. No crash, flips as desired. Though 1 at a time. AND it even fixes the wheel scrolling. Up untill I re-join the game, aaaand here we go the crashing steps yet again if necessary.

So, the warriors:

• for the defending team:
  • 3 awesome servers.
  • Factorio 1.1.6 linux build 57355.
• for the offending team:
  • 1 dark book of a wall sorcery.
• The goal:
  • to crash or not to crash.

The details

The blueprints lib "wins". Each time. With crashing colors! As fast as 2 minutes in since game launch, if necessary.

Well, the book has... another book inside, which has some fine prints with "global grid" relation set... And whatever the server i join i can get a crash with this particular book if I want to?
Seems I got myself the cursed book.

the valuable part of the log with a stack trace

The crash usually starts with smth like

Code: Select all

 111.873 Error CrashHandler.cpp:621: Received SIGSEGV
Factorio crashed. Generating symbolized stacktrace, please wait ...

or

Code: Select all

66.645 Error CrashHandler.cpp:621: Received SIGSEGV
Factorio crashed. Generating symbolized stacktrace, please wait ...

bruh, the forum can't keep up with this full-sized text quite fast enough... so here's the attachment file.
The file contains the *valuable* (imo) part with a symbolic stack trace only.

factorio_redacted.log

(4.84 KiB) Downloaded 130 times

For now I have one log of the first game crash and 3 repros on 3 servers. All -v mode.

As I see some

Code: Select all

0x0000000001073500 in ?? at ??:0

("??" crapstuff) in the midst of these...
I _suppose_ i have some headers/debug symbols missing? Or am I? Is this just enough yet?

the rest

I have all the logs and the bp storage backed up.
I am at linux64 distro right now and can install dbg packets/ change pathes/ update tools if necessary.
I am not into sharing the full log and/or bp storage here if possible. Not because of security reasons/issues but rather because of a common privacy idea.
I am willing to provide files necessary and assistance required assuming the related transfered data is by design a subject to non-public disclosure-terms such as PM here or e.g. any discord non-public options available.


As the "send crashes" is turned on and the logs pretty much say it got sent I assume the full logs are already delivered as well.
So as not to strain "the logstash" at the receiving end I won't try new crashes for testing purposes and shall leave the things as they are right now.

A little post- question

Regarding security related issues in general:
not really asking for the security.txt file as we have... support@factorio.com at home.
But imagine, say some easy in-game `/mute` bypass gets discovered and posted righty onto the active-topics forums, lol.
Ofc it is a harmless cherry picked one. And by a common sense you want this stuff to be disclosed either simply "never in public" or after the most clients get the related patches. Not a week before the patch is about to (maybe) be released at all. So the Forums are off the board imo. Alas FAQ bears no other way either. At least the support mail desk is mentioned as per endpoint. But well, who's gonna write a mail about the bug inside a mp inside a game... Probably even less then those who get to the forums at the first place.
So, do we report these impact-confusing issues to mods (hoping they would send the word flying the right direction)? Do we simply wrap-and-mail it? Or do we post it on forums as per usual bugs?

[kovarex]

Hello, I would need a way to reproduce it. Ideally a blueprint-storage.dat file with steps how to reproduce it.
I can't do much until then.

[KeepResearchinSpoons]

Sent a PM with bpst.dat files. Also the steps are included. Bp storage it too big for forums so zipped it down.

[kovarex]

Hm, I wasn't able to reproduce any of the crashes even with the extra data you gave me in the PM, but since I really trusted that you are not making this up, I checked the relevant code surroundings little bit more carefuly, and I found some potential problem that would actually only cause problems randomly and typically in start of the game, not later on.

Technical:
The problem was related to invalidation of std::vector content by inserting to it after some (lower recursion levels) already got reference to its value to be used as index for iterating in the blueprintbook (record).
This makes it even more probable to be the problem when I see the random address in the stacktrace related to usage of the integer used as index being reference from the std::vector.
This problem was repeated on 3 places where the game needs to keep custom external book active indexes for each player, so when one player scrolls a book, other isn't affected.

I fixed these 3 places and there is quite a big chance that this crash could go away, but since I wasn't able to reproduce it, I can't guarantee, so if you can, let me know if this gets fixed for you once we release 1.1.9.
But my guts tells me, that there is 95% chance that the fix is related to crash.

[KeepResearchinSpoons]

We shall see... But well it appears as if it is one of those latency squad things. The issue is still as stable as ever. But well something different might have *just* been improved so that it won't break in long-games with mods anymore when you try to get a bunch of books off your sleeve without actually making a verbatim copy for them.

The new PM (just yet another message lol) has the server side logs that might hint the latency as per reason. And I really hope it is not of the nature of the fast vs reliable consistency problem things this time around uuh. Cuz ofc I like it fast. But ofc I like it reliable just as much Good luck finding the nook and cranny for the solution to cut the marvel!

edit: FIXED! Now only the PM related issue is still alive. Case closed.