Importing SVG's in 4.3.dev with commit 73589f6 is crashing

[RevoluPowered]

Tested versions

• Reverting ThorVG: update from v0.12.1 to v0.12.3 #87612 resolves the crashing
• Crash is not always present on windows but sometimes is
• Linux could repro the crash each time booting the engine provided you nuke the .godot/ folder

System information

Linux - Fedora 38 - As up to date as possible

Issue description

When importing SVG files in multiple threads the engine crashes with a realloc() crash, I did try to workaround this but had the same crash

Steps to reproduce

• Import a pool of SVG's (doesn't seem to matter the type)

Minimal reproduction project (MRP)

TESTPROJECTREPRO.zip

Images taken from another source, our project files are completely unrelated and they also crash.

[akien-mga]

CC @capnm

[capnm]

Unfortunately I can't reproduce this myself, I even tested the PR importing a large amount of SVGs (debian 12, Vulkan API 1.3.230 - Forward+):
svg.zip

I have now run it again on the current tip 9adb7c7 and didn't see any issue
(with or without

[rendering]
driver/threads/thread_model=2

)

If anyone is able to reproduce this, perhaps could try to bisect the mentioned full ThorVG changelog.

The most likely would be thorvg/thorvg@cfb6a7e

[hermet]

@RevoluPowered Hello, I'm wondering if you can use the thread sanitizer to catch any clues?

[Rubonnek]

If the project is configured to use single threads (i.e. modify project.godot to include):

[editor]

import/use_multiple_threads=false

The SVGs import fine.

I'm able to reproduce the crash but I don't think it's related to ThorVG -- the crash happens at different places and due to double frees or some memory corruption. The root cause is most likely a race condition.

I stumbled upon a similar issue in #84364 with the same workaround, but it's far easier to reproduce the crash by importing these SVGs.

[akien-mga]

@RevoluPowered @Rubonnek I couldn't reproduce this bug on Linux (Mageia 9), but I suspect it's related to #87962 (which I can reproduce) and should be fixed by #88053. Could you test to confirm?

[Rubonnek]

@akien-mga can confirm -- I can't reproduce the crash described in this issue anymore easily with a build from 36e943b.

Edit: there's still some strange things going on under the hood due to the data race in EditorFileSystem::_reimport_thread but that problem merits its own issue.