Bug #1954
open
Added by John Robertson 8 months ago.
Updated 13 days ago.
Description
Latest dev [Qt]
Race condition `real_menus_update()` is sometimes called before `real_menus_init()`.
For the Qt client this results in accessing an uninitialized `multiplayer_menu` variable.
Files
This could also be cause from a delay in the loading of the ruleset (e.g. `!game.client.ruleset_ready`).
If `real_menu_init() returns without initializing, because the ruleset took too long to load, is it ever called again?
I tend to startup the server and client at the same time. I think `real_menus_init()` returns because the ruleset has not yet been loaded, causing the `multiplayer_menu` variable to never become initialized.
Reduced, but not perfect, risk.
Perfectly remove the SIGSEGV, at the sacrifice of the menus. The menus are constantly rebuilt/updated, so eventually, the `real_menus_init()` should succeed.
Before fix
`multiplayer_menu` is NULL in `real_menus_update()` if `real_menus_init()` exits because the server ruleset is not ready.
John Robertson wrote in #note-2:
I tend to startup the server and client at the same time. I think `real_menus_init()` returns because the ruleset has not yet been loaded, causing the `multiplayer_menu` variable to never become initialized.
That makes no sense. Server loads rulesets before accepting any client connections, so server start time shouldn't have any role in this.
However, it's possible that client tries to use menus before it has received rulesets from the server. We've had bugs of that type in the past. There's mystery, though, in why you are seeing this and others (at least myself) are not. One would assume it to be deterministic.
Also available in: Atom
PDF