Skip to content

GPU tests that create an app with gpu=true take over 120 s in renderer_init and look like hangs #90

Open

Description

@5hif7y

Related to #10

What happens

Host and editor tests that call host.app_create({gpu=true,...}) take more than 120 s in renderer_init here (RX 580 (no ray queries, so the software tracer), amdvlk 31.0.21924.61, Windows 10, i5-7400, 16 GB).

  • mirai test kills a test after 120 s by default, so they look like hangs.
  • They pass with --timeout 900.
  • Reproduced on a clean main 8c5a2a6 with a_lost_gpu_leaves_the_frames_running_without_drawing.

Cause, from the run: with app.root="" there is no pipeline store, so every run compiles the renderer's pipelines from nothing, with validation on.

Expected

A GPU test of the host does not compile the whole renderer cold every run. Either it gets a store (a shared one for tests), or it asks only for the pipelines it uses, so the default timeout fits.

🤖 Generated with Claude Code

Activity

  1. self-assigned this
    on Nov 15, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions