F12 and crash #52782

New Issue

pistol ioan · 2017-09-16T17:54:25+02:00

pistol ioan commented

2017-09-16 17:54:25 +02:00

Windows 10 64, gtx 970

Blender 2.79

Just open the file and press f12 and crash

untitled3.blend

Windows 10 64, gtx 970 Blender 2.79 Just open the file and press f12 and crash [untitled3.blend](https://archive.blender.org/developer/F806017/untitled3.blend)

pistol ioan commented

2017-09-16 17:54:25 +02:00

Changed status to: 'Open'

pistol ioan commented

2017-09-16 17:54:25 +02:00

Added subscriber: @istoltoto

pistol ioan commented

2017-09-16 17:55:12 +02:00

in cycles using gpu render

Ray molenkamp commented

2017-09-16 18:19:29 +02:00

Added subscribers: @mont29, @LazyDodo

Bastien Montagne was assigned by Ray molenkamp

2017-09-16 18:19:29 +02:00

Ray molenkamp commented

2017-09-16 18:19:29 +02:00

It's a null ref in particles.c ( psys_thread_create_path ) @mont29 mind taking a peek here?

Bastien Montagne removed their assignment 2017-09-19 16:03:31 +02:00

Sergey Sharybin was assigned by Bastien Montagne

2017-09-19 16:03:31 +02:00

Bastien Montagne commented

2017-09-19 16:03:31 +02:00

Added subscriber: @Sergey

Bastien Montagne commented

2017-09-19 16:03:31 +02:00

Bah… yet another mismatch between rendering data and regular scene update data, here there are less child particles for rendering (5) than for 3DView (30), so when rendering data update process (re-)allocates memory for 5 children, but then regular scene update, which happens in parallel, tries to use that allocated cache for 30 children, it obviously breaks…

Simple way to "fix" this: never use less children for rendering than for 3DView!

@Sergey reassigning this to you, since I’d like to have your feelings on the issue before diving deeper into ugly particle code. :/

Bah… yet another mismatch between rendering data and regular scene update data, here there are less child particles for rendering (5) than for 3DView (30), so when rendering data update process (re-)allocates memory for 5 children, but then regular scene update, which happens in parallel, tries to use that allocated cache for 30 children, it obviously breaks… Simple way to "fix" this: never use less children for rendering than for 3DView! @Sergey reassigning this to you, since I’d like to have your feelings on the issue before diving deeper into ugly particle code. :/

Sergey Sharybin commented

2017-09-19 16:07:41 +02:00

There shouldn't be a parallel scene update at the same time if one renders with locked interface and/or fullscreen render. If that configuration still crashes, think it worth investigating why.

Bastien Montagne commented

2018-03-13 20:17:26 +01:00

@Sergey can confirm the crash still happens when trying to render with locked UI and full screen… Also happens with fixed single-thread rendering, and -t 1 option

@Sergey can confirm the crash still happens when trying to render with locked UI *and* full screen… Also happens with fixed single-thread rendering, and `-t 1` option

Sergey Sharybin removed their assignment 2019-01-23 16:32:03 +01:00

Sergey Sharybin commented

2019-01-23 16:32:03 +01:00

Not much i can add here. This isn't a regression, and the code is quite fragile and difficult to follow. Not something i want to be digging into.

Would rather facilitate the new node-based particles replacement instead! :)

Not much i can add here. This isn't a regression, and the code is quite fragile and difficult to follow. Not something i want to be digging into. Would rather facilitate the new node-based particles replacement instead! :)

Germano Cavalcante commented

2020-01-16 14:43:59 +01:00

Added subscriber: @mano-wii