[boskid][0.18.15] Crash on save: "incomingBlock->getOutboundSignals().count(const_cast<RailSignalBase*>(this))" failed

[Mat46]

Hello

Il have a intermittent problem in the game on manual save or autosave.

In the log, i see 2 errors :
- the first with visual studio 2017
- the second with RailSignalBase

Is this a problem with a mod or something else ?

Thank you for your answer.
Mat

Factorio Log :

factorio-current.log

(75.79 KiB) Downloaded 136 times

Mod List :

mod-list.json

(1.75 KiB) Downloaded 138 times

My save :

Solo.zip

(5.91 MiB) Downloaded 136 times

[Loewchen]

If you have repeated unrelated crashes you can not reliably reproduce then the issue is your hardware/OS/driver and not the game.

[Mat46]

Ok, i check this.

Thank you
Mat

[Loewchen]

I'd start with: https://www.memtest.org/ let it run for several complete loops or until an error.

[Mat46]

Hello

I run Memtest 3 times in the row (for the moment 0 errors). I will continue the test in the next week.

For info, my pc have 5 month. I changed everything except the graphics card and 2 hard drives. (PC Assembled).

I checked all the update for windows, drivers, .. and all is ok.
I uninstalled all versions of visual basic and installed only visual basic 2019 -> Same Problem.
I uninstalled factorio and reinstalled -> Same Problem.

I didn't have the problem in previous versions.

I noticed that I make a manual backup right after loading my save, I don't have the problem.
If I start playing for 5 minutes and save, I have the bug.

I'm going to do other tests by deactivating mods

Thank You
Mat

[Mat46]

Hello

I think i found the problem.

The problem comes from the "Train Construction Site" mod. If I put a rail signal next to the block (see the photo attached), I have the problem.
I will contact the modder to tell him the problem.

Thank you
Mat

Factorio.png (3.02 MiB) Viewed 5501 times

[Loewchen]

Mods generally should not be able to cause crashes so this should still be fixed game side. Is the save you provided the save you can reproduce this with?

[Mat46]

Hello

Yes, this is the save with which we can reproduce the problem.
I tested it several times with an automatic and manual backup.

Thank You
Mat

[dpacbach]

"incomingBlock->getOutboundSignals().count(const_cast<RailSignalBase*>(this))" failed

note to devs... const_cast is a code smell!

[Rseding91]

"incomingBlock->getOutboundSignals().count(const_cast<RailSignalBase*>(this))" failed

note to devs... const_cast is a code smell!

And yet, it doesn't matter.

[boskid]

Ok, after wasting too much time trying to find a nice solution i went berserk and fixed it the hard way:
- as stated in 86385, every rail signals will have to have collision masks that collide with each other to prevent multiple rail signals from being placed on the same tile.
- when trying to build a rail signal not on a tile center, the rail signal build position will be changed to be on a tile center (aka floor(x)+0.5)
- flag "placeable-off-grid" on any rail signal will be ignored.
- rail signal collision box is hardcoded to {{-0.2, -0.2}, {0.2, 0.2}} to prevent defining collision box that would not contain a {0,0} point inside preventing multiple signals at the same position from colliding with each other. This is also to make sure signal will be properly found when looking for a too-close signal on adjacent tile.
- if by any chance (or using editor's clone or scripted clone without snap-to-grid) a rail signal would land not on a tile center, it will not be considered valid so it will not bind to the rails
- all the saves from before 1.1.0 will have rail segments recomputed so if there are any signals not on a tile center, they will detach from rails immediately.

Unfortunately there are some assumptions that are important for rail signals logic to work so i will not remove any of those restrictions.