Status: reported once, not yet reproduced, maintainer-filed to track it. A reproduction, or a trace captured while it is happening, is what this needs.
What happens
In the dual Mandelbrot/Julia view, the Julia panel sometimes stays at a reduced-resolution (blurry, upscaled) stage after motion, or lands on a flat-looking plane, while the Mandelbrot panel beside it is sharp. Reported at a moderate depth (Mandelbrot pane around 82×, Julia driven live by the cursor), with the Julia panel stuck upscaled rather than refining to full detail.
What is known
Two mechanisms are plausible and neither is confirmed:
- the per-panel frame-budget controller starving for the Julia view — its only lever is resolution, so a budget stuck low renders coarse indefinitely; or
- a display-side stall where the panel has rendered sharp but keeps presenting the stale coarse frame (a variant of a single-view problem fixed earlier, now possibly surfacing in one panel of the dual view).
A harness (--dualsettle) was built to chase it: it boots the dual view, drives the live Julia point along an arc through the real code path, then idles and measures sharpness. Across many trials, clean and under CPU load, every settle came back sharp within a second. The intermittent case has not been caught, so nothing has been changed — a fix without a reproduction would be a guess.
How you can help
If you see it, a short trace while it is on screen would likely identify which of the two mechanisms it is:
- Reproduce it (dual view, Julia driven by the cursor, after some motion).
- Relaunch with
FRACTADYNE_TRACE=gpu,tile and let it run about ten seconds while the panel is blurry.
- Attach the log, and if you can, the Performance panel readings (the resolution and budget lines name the mechanism).
Notes on your setup — depth, whether the Julia panel was being zoomed as well as re-centred, and whether a panel resize was involved — all help, since those are the profiles the harness has not yet driven together.
For maintainers
Tracked in TODO.md ("DUAL-VIEW JULIA SOMETIMES STAYS BLURRY…"); --dualsettle [DIR] is committed. Per the project's no-close-without-reproduction rule, this stays open until it reproduces.
Status: reported once, not yet reproduced, maintainer-filed to track it. A reproduction, or a trace captured while it is happening, is what this needs.
What happens
In the dual Mandelbrot/Julia view, the Julia panel sometimes stays at a reduced-resolution (blurry, upscaled) stage after motion, or lands on a flat-looking plane, while the Mandelbrot panel beside it is sharp. Reported at a moderate depth (Mandelbrot pane around 82×, Julia driven live by the cursor), with the Julia panel stuck upscaled rather than refining to full detail.
What is known
Two mechanisms are plausible and neither is confirmed:
A harness (
--dualsettle) was built to chase it: it boots the dual view, drives the live Julia point along an arc through the real code path, then idles and measures sharpness. Across many trials, clean and under CPU load, every settle came back sharp within a second. The intermittent case has not been caught, so nothing has been changed — a fix without a reproduction would be a guess.How you can help
If you see it, a short trace while it is on screen would likely identify which of the two mechanisms it is:
FRACTADYNE_TRACE=gpu,tileand let it run about ten seconds while the panel is blurry.Notes on your setup — depth, whether the Julia panel was being zoomed as well as re-centred, and whether a panel resize was involved — all help, since those are the profiles the harness has not yet driven together.
For maintainers
Tracked in
TODO.md("DUAL-VIEW JULIA SOMETIMES STAYS BLURRY…");--dualsettle [DIR]is committed. Per the project's no-close-without-reproduction rule, this stays open until it reproduces.