You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
render_far_from_the_centre_looks_as_at_it_and_a_jump_keeps_the_history fails on main with the software tracer #89
On main8c5a2a6 (RX 580 (no ray queries, so the software tracer), amdvlk 31.0.21924.61, Windows 10, i5-7400, 16 GB), this test fails alone and in the full suite:
a view 15 km from the origin is the one at it: mean 0.00011748416, most 0.13720703
The mean difference is tiny, but one pixel differs by 0.137.
A view far from the origin looks as one at it, within the test's tolerance, with the software tracer too. Or the tolerance accounts for what is expected to differ.
Not investigated: whether it happens only with the software tracer.
Still fails on main4aae7bb4, on the same machine as above: RX 580 (no ray queries, so the software tracer), amdvlk, Windows 10. The numbers are the same:
a view 15 km from the origin is the one at it: mean 0.00011748416, most 0.13720703
It ran twice on a clean origin/main checkout, while verifying #110. That branch does not touch this path.
Activity
5hif7y commentedon Nov 16, 2026
Still fails on
main4aae7bb4, on the same machine as above: RX 580 (no ray queries, so the software tracer), amdvlk, Windows 10. The numbers are the same:It ran twice on a clean
origin/maincheckout, while verifying #110. That branch does not touch this path.🤖 Generated with Claude Code