Game: 6 seconds freeze entering castle town #2181

Open
opened 2026-11-22 01:56:53 +00:00 by conongnang · 4 comments

Longshot version

nightly-738-dd5ffe6a (source 8fd4dfe4)

Channel

Nightly

What kind of problem

Crash, freeze or black screen

What happened

Entering castle town freezes the game for about 10 s. This reproduced in both runs I logged. Presentation stops (0.2–0.3 fps over a 5–7 s window) and the game stops reading the controller.

Where in the game

Market_Child (child) Entering Castle Town

How often

Every time

Graphics card and driver version

AMD Radeon RX 6800, AMD proprietary driver 26.10.1 (AMD proprietary shader compiler)

CPU, RAM and operating system

13th Gen Intel(R) Core(TM) i7-13700 (24 threads), 32 GB, Windows 11 (build 26100)

Changes from the defaults

No response

longshot_log.txt

2 files below

Screenshots or video

No response

Before you send it

  • The problem is still there on the newest release or nightly.
  • I searched the open issues for it.
  • I'm not attaching game files, keys, the game's executable, its .bin caches or saves.
### Longshot version nightly-738-dd5ffe6a (source 8fd4dfe4) ### Channel Nightly ### What kind of problem Crash, freeze or black screen ### What happened Entering castle town freezes the game for about 10 s. This reproduced in both runs I logged. Presentation stops (0.2–0.3 fps over a 5–7 s window) and the game stops reading the controller. ### Where in the game Market_Child (child) Entering Castle Town ### How often Every time ### Graphics card and driver version AMD Radeon RX 6800, AMD proprietary driver 26.10.1 (AMD proprietary shader compiler) ### CPU, RAM and operating system 13th Gen Intel(R) Core(TM) i7-13700 (24 threads), 32 GB, Windows 11 (build 26100) ### Changes from the defaults _No response_ ### longshot_log.txt 2 files below ### Screenshots or video _No response_ ### Before you send it - [x] The problem is still there on the newest release or nightly. - [x] I searched the open issues for it. - [x] I'm not attaching game files, keys, the game's executable, its .bin caches or saves.
Member

Possibly related, for context:

  • PR #1827 fix(host): a CPU read or write of texels gives the thread's virtual core up while it waits for the GPU (no ~1 s freeze at area changes) — Merged fix addresses approximately one-second area-change freezes, but this reported longer, repeatable hitch may differ.
  • PR #1701 perf(vk): the texture pipelines made before the game, through the pipeline cache — Merged fix eliminates first-use texture pipeline compilation stalls; relevance to this repeatable transition freeze is uncertain.
  • issue #1831 Game: stutter at the Castle Town fountain on quick direction changes: Hyrule Field and Hyrule Castle are loaded ahead and dropped again at every turn (related: #996, #1639) — Related Castle Town performance stutter, but occurs with movement in the market rather than entering.
Possibly related, for context: - PR #1827 fix(host): a CPU read or write of texels gives the thread's virtual core up while it waits for the GPU (no ~1 s freeze at area changes) — Merged fix addresses approximately one-second area-change freezes, but this reported longer, repeatable hitch may differ. - PR #1701 perf(vk): the texture pipelines made before the game, through the pipeline cache — Merged fix eliminates first-use texture pipeline compilation stalls; relevance to this repeatable transition freeze is uncertain. - issue #1831 Game: stutter at the Castle Town fountain on quick direction changes: Hyrule Field and Hyrule Castle are loaded ahead and dropped again at every turn (related: #996, #1639) — Related Castle Town performance stutter, but occurs with movement in the market rather than entering.
Member

This is the simulated-turf stall that #2174 fixes, and the v3.1.0 release download does not have that fix.

Both of this issue's logs show it. On entering Market every simulated-turf list is released to 0 records. Then Market asks for a few, and they grow back from 0:

  • longshot_log.txt: blade list ... resized 40768 -> 0 records, pool ... resized 8608 -> 0 and 8128 -> 0 (lines 18103–18202), then a capture's wait for write-back 198357 timed out, 0.3 fps over the last 7.1 s, the game read no controller for 6.6 s, then resized 0 -> 32 / 0 -> 64.
  • longshot_log.1.txt: the same, at lines 10572–10818 (25536 -> 0 ... four write-back timeouts, 0.2 fps over the last 5.6 s, 0.3 fps over the last 7.1 s).

That is #2174's sequence: each list grows back from 0 with a new write-back that a capture then waits 250 ms for. #2174 keeps the lists at the game's own size (143360 / 138240 records), so they can no longer reach 0. It merged after nightly-738, which this report ran, and it is in nightly-745 (a2dc5385).

A second report shows it on NVIDIA with the v3.1.0 release. It was made with the Insert menu's bug report, and was sent to me rather than posted. Windows 11, RTX 5070 (NVIDIA 621.24), i5-13600KF, v3.1.0 (source 6f470170), in Market_Child:

  • simulated turf: blade list ... resized 6208 -> 0 records, 4864 -> 0, and the four pools to 0
  • then a capture's wait for write-back 121966 timed out (done 121960) and slow submit: 250.3 ms: dispatch 250.3, six times
  • nvn: GPU waits in 5 s: 32 sync waits (3498 ms), 20.7 fps over the last 8.5 s, then 0.5 fps over the last 6.0 s, the game read no controller for 6.0 s, then 37 and 56 fps, so about 25 s from the start before it ran near 60 fps again

The v3.1.0 release download was built from 9e10f6f7, one commit before #2174. The v3.1.0 tag points at a2dc5385. The release's own target_commitish is 9e10f6f7. Recomputing nx2/version.py's fingerprint over git archive <commit> host gpu nx2:

  • 9e10f6f7 gives 6f470170, the "source" that v3.1.0 prints
  • a2dc5385 gives 63363428
  • as a check, dd5ffe6a gives 8fd4dfe4, which matches nightly-738's line here

So everyone on the v3.1.0 release can get this stall. Its worst case in #2174's own testing was vk: FATAL: the GPU hung: no submission finished in 30 s, which players would see as a crash. Until the release is rebuilt, players can use nightly-745 or later, or put RX_VK_SIM_TURF_ADAPTIVE=0 in longshot_env.txt, which keeps the lists at stock as before #2093.

Found with AI assistance (Claude Code).

**This is the simulated-turf stall that #2174 fixes, and the v3.1.0 release download does not have that fix.** **Both of this issue's logs show it.** On entering Market every simulated-turf list is released to 0 records. Then Market asks for a few, and they grow back from 0: - `longshot_log.txt`: `blade list ... resized 40768 -> 0 records`, `pool ... resized 8608 -> 0` and `8128 -> 0` (lines 18103–18202), then `a capture's wait for write-back 198357 timed out`, `0.3 fps over the last 7.1 s`, `the game read no controller for 6.6 s`, then `resized 0 -> 32` / `0 -> 64`. - `longshot_log.1.txt`: the same, at lines 10572–10818 (`25536 -> 0` ... four write-back timeouts, `0.2 fps over the last 5.6 s`, `0.3 fps over the last 7.1 s`). That is #2174's sequence: each list grows back from 0 with a new write-back that a capture then waits 250 ms for. #2174 keeps the lists at the game's own size (143360 / 138240 records), so they can no longer reach 0. It merged after nightly-738, which this report ran, and it is in nightly-745 (`a2dc5385`). **A second report shows it on NVIDIA with the v3.1.0 release.** It was made with the Insert menu's bug report, and was sent to me rather than posted. Windows 11, RTX 5070 (NVIDIA 621.24), i5-13600KF, `v3.1.0 (source 6f470170)`, in Market_Child: - `simulated turf: blade list ... resized 6208 -> 0 records`, `4864 -> 0`, and the four pools to 0 - then `a capture's wait for write-back 121966 timed out (done 121960)` and `slow submit: 250.3 ms: dispatch 250.3`, six times - `nvn: GPU waits in 5 s: 32 sync waits (3498 ms)`, `20.7 fps over the last 8.5 s`, then `0.5 fps over the last 6.0 s`, `the game read no controller for 6.0 s`, then 37 and 56 fps, so about 25 s from the start before it ran near 60 fps again **The v3.1.0 release download was built from `9e10f6f7`, one commit before #2174. The `v3.1.0` tag points at `a2dc5385`.** The release's own `target_commitish` is `9e10f6f7`. Recomputing `nx2/version.py`'s fingerprint over `git archive <commit> host gpu nx2`: - `9e10f6f7` gives `6f470170`, the "source" that v3.1.0 prints - `a2dc5385` gives `63363428` - as a check, `dd5ffe6a` gives `8fd4dfe4`, which matches nightly-738's line here So everyone on the v3.1.0 release can get this stall. Its worst case in #2174's own testing was `vk: FATAL: the GPU hung: no submission finished in 30 s`, which players would see as a crash. Until the release is rebuilt, players can use nightly-745 or later, or put `RX_VK_SIM_TURF_ADAPTIVE=0` in `longshot_env.txt`, which keeps the lists at stock as before #2093. Found with AI assistance (Claude Code).
Member

From Discord:

Additional log from a player with Ryzen 9 9900X3D and RTX 5080: the log shows simulated-turf lists staying at stock sizes in Market and then adapting downward (e.g. 143360 -> 1920) after returning to ZoraFountain_Child; this differs from the zero-capacity/write-back timeout sequence diagnosed here. The log ends in a GPU hang, so please inspect the attached log for its exact trigger and version.

From [Discord](https://discord.com/channels/1569849200047427645/1573887236498391160/1573887236498391160): Additional log from a player with Ryzen 9 9900X3D and RTX 5080: the log shows simulated-turf lists staying at stock sizes in Market and then adapting downward (e.g. 143360 -> 1920) after returning to ZoraFountain_Child; this differs from the zero-capacity/write-back timeout sequence diagnosed here. The log ends in a GPU hang, so please inspect the attached log for its exact trigger and version.
Member

v3.1.1 is out with #2174: https://git.magicbean.dev/MagicBeanDev/Longshot/releases/tag/v3.1.1

It is built from a2dc5385, the commit v3.1.0's tag names, so it has the simulated-turf fix that v3.1.0's download lacked. The stable channel offers it now, and launchers on v3.1.0 get it as an update. On a nightly, nightly-745 or later already has it.

If the freeze on entering Market or going back into Hyrule Field still happens on v3.1.1, please post the new longshot_log.txt here.

For a maintainer with release rights: v3.1.1's page has no "What's new". Its notes cover the commits since v3.1.0's tag, and that tag is on the same commit. My account can push tags but not edit releases. This could go under its ## v3.1.1 heading:

v3.1.1 is v3.1.0 as it was meant to be. v3.1.0's download was built one commit too early (9e10f6f7), so it lacks two changes that are on its tag. Everything else in v3.1.0's notes is in this release too.

Fixes

  • vk: No more 5–25 s freezes on entering Market or going back into Hyrule Field, and no more GPU hangs there that closed the game: the simulated-turf lists keep the game's own size in areas without turf (#2174; reports in #2181)

Features

  • saves: castle-gate, child Link in Hyrule Field at Castle Town's gate, by day (#1722)

Players on v3.1.0 get v3.1.1 as an update from the launcher. Until they do, RX_VK_SIM_TURF_ADAPTIVE=0 in longshot_env.txt avoids the freeze on v3.1.0.

**v3.1.1 is out with #2174:** https://git.magicbean.dev/MagicBeanDev/Longshot/releases/tag/v3.1.1 It is built from `a2dc5385`, the commit v3.1.0's tag names, so it has the simulated-turf fix that v3.1.0's download lacked. The stable channel offers it now, and launchers on v3.1.0 get it as an update. On a nightly, nightly-745 or later already has it. If the freeze on entering Market or going back into Hyrule Field still happens on v3.1.1, please post the new `longshot_log.txt` here. **For a maintainer with release rights:** v3.1.1's page has no "What's new". Its notes cover the commits since v3.1.0's tag, and that tag is on the same commit. My account can push tags but not edit releases. This could go under its `## v3.1.1` heading: > **v3.1.1 is v3.1.0 as it was meant to be.** v3.1.0's download was built one commit too early (`9e10f6f7`), so it lacks two changes that are on its tag. Everything else in v3.1.0's notes is in this release too. > > ### Fixes > > - **vk:** No more 5–25 s freezes on entering Market or going back into Hyrule Field, and no more GPU hangs there that closed the game: the simulated-turf lists keep the game's own size in areas without turf (#2174; reports in #2181) > > ### Features > > - **saves:** castle-gate, child Link in Hyrule Field at Castle Town's gate, by day (#1722) > > Players on v3.1.0 get v3.1.1 as an update from the launcher. Until they do, `RX_VK_SIM_TURF_ADAPTIVE=0` in `longshot_env.txt` avoids the freeze on v3.1.0.
Sign in to join this conversation.
No milestone
No project
No assignees
3 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
MagicBeanDev/Longshot#2181
No description provided.