Projects take considerably longer to start if another instance exists (editor or another game) (possibly shader caching bug)

[nathanfranke]

Godot version

3f39625
Can't to reproduce in 3.x (tested 3.4.4.stable and 3.5.rc7)

System information

Arch on 5.18.12-arch1-1

Issue description

See #43351 (comment)

$ time ./bin/godot.linuxbsd.opt.tools.64.llvm --path ~/workspace/empty --quit

...

real    0m1.804s
user    0m2.805s
sys     0m0.386s

Again, in another terminal right after:

$ time ./bin/godot.linuxbsd.opt.tools.64.llvm --path ~/workspace/empty --quit

...

real    0m7.352s
user    0m6.795s
sys     0m0.394s

Steps to reproduce

1. Run a godot project with --path
2. Run a godot project with --path
   the second instance will take considerably longer to start

Minimal reproduction project

empty.zip

[Calinou]

Can you reproduce this in 3.x to confirm whether this is due to 4.0's shader cache?

[Zylann]

I experience really slow startups as well. When starting a game from the editor, it shows a white window for about 5 seconds before it starts showing the splash. It's nowhere near startup times in Godot 3.

[nathanfranke]

Can you reproduce this in 3.x to confirm whether this is due to 4.0's shader cache?

Cannot reproduce at all in 3.4.4.stable or 3.5.rc7. However I did discover a similar bug #63460

I experience really slow startups as well. When starting a game from the editor, it shows a white window for about 5 seconds before it starts showing the splash. It's nowhere near startup times in Godot 3.

Issue #43351 may be more related to your issue, this issue is only about starting multiple instances simultaneously.

[KoBeWi]

Can't reproduce, both instances take the same time to launch.
@nathanfranke Can you confirm that this is fixed?

[KoBeWi]

Closing. You can deconfirm it.

[Zireael07]

Is "deconfirm" a typo for "reconfirm"?

[nathanfranke]

I will check it soon

[KoBeWi]

Is "deconfirm" a typo for "reconfirm"?

No, it refers to

Can you confirm that this is fixed?

I just closed the issue assuming it's fixed. So if it's not, it would "deconfirm" it.

[nathanfranke]

I don't see this issue on 4.0.stable, probably fixed.