Game: On CachyOS, pipeline cache is rebuilt on every launch and build assets are not reused #1581

Open
opened 2026-11-18 13:02:26 +00:00 by BeanIssues · 3 comments
Collaborator

Reported on Discord

Longshot version

v3.0.0-rc.7 and the newest nightly (exact nightly build not provided).

Channel

Release candidate and nightly.

What happened

The reporter says there are two related cache-reuse problems on CachyOS:

  1. The builder rereads the ROM and performs a full build rather than reusing compiled assets. With six workers and 12 GB RAM, a full build takes 1.5–2 hours.
  2. After allowing shader preparation on the first launch, the game does not persist/reuse that state; it rebuilds on every launch, taking 45–60 minutes.

They also report multiple directories of files are saved locally but precompiled assets are still not reused. Expected: reuse valid compiled build assets and shader/pipeline caches on subsequent builds/launches.

Where in the game

Startup pipeline/shader preparation, before gameplay.

How often

Every build / every launch, according to the reporter.

Graphics card and driver version

The report says “3060x 306012gb”; exact GPU model and driver version are unknown.

CPU, RAM and operating system

CachyOS Linux; 12 GB RAM. CPU model is unknown.

Performance measurements

The reported durations are 1.5–2 hours for a full build and 45–60 minutes for pipeline preparation on each launch. No profiling measurements or logs were provided. For a performance profile, follow https://longshot.magicbean.dev/help/performance/ and include the resulting bug report folder and longshot_log.txt.

Still needed

  • Exact Longshot nightly build, launcher version, CPU model, GPU model and driver version.
  • Build log and longshot_log.txt from a repeated launch; reproduction steps and whether the same installed build is launched each time.
  • Confirmation of cache directory contents and whether any cleanup/free-space action or updates occurred between runs.
  • Longshot#1510 concerns shader-cache stamp invalidation that can force shader preparation again, but does not establish that this report has the same cause: #1510
  • Longshot#1551 reports pipeline caches not reused on Windows/Adreno, a different GPU/OS condition: #1551
  • Longshot#1418 fixed unnecessary pipeline rebuilds after a no-shader-change update and is included since nightly-476: #1418
Reported on [Discord](https://discord.com/channels/1569849200047427645/1572595870405500959/1572595870405500959) ### Longshot version v3.0.0-rc.7 and the newest nightly (exact nightly build not provided). ### Channel Release candidate and nightly. ### What happened The reporter says there are two related cache-reuse problems on CachyOS: 1. The builder rereads the ROM and performs a full build rather than reusing compiled assets. With six workers and 12 GB RAM, a full build takes 1.5–2 hours. 2. After allowing shader preparation on the first launch, the game does not persist/reuse that state; it rebuilds on every launch, taking 45–60 minutes. They also report multiple directories of files are saved locally but precompiled assets are still not reused. Expected: reuse valid compiled build assets and shader/pipeline caches on subsequent builds/launches. ### Where in the game Startup pipeline/shader preparation, before gameplay. ### How often Every build / every launch, according to the reporter. ### Graphics card and driver version The report says “3060x 306012gb”; exact GPU model and driver version are unknown. ### CPU, RAM and operating system CachyOS Linux; 12 GB RAM. CPU model is unknown. ### Performance measurements The reported durations are 1.5–2 hours for a full build and 45–60 minutes for pipeline preparation on each launch. No profiling measurements or logs were provided. For a performance profile, follow https://longshot.magicbean.dev/help/performance/ and include the resulting bug report folder and `longshot_log.txt`. ### Still needed - Exact Longshot nightly build, launcher version, CPU model, GPU model and driver version. - Build log and `longshot_log.txt` from a repeated launch; reproduction steps and whether the same installed build is launched each time. - Confirmation of cache directory contents and whether any cleanup/free-space action or updates occurred between runs. ### Related - Longshot#1510 concerns shader-cache stamp invalidation that can force shader preparation again, but does not establish that this report has the same cause: https://git.magicbean.dev/MagicBeanDev/Longshot/issues/1510 - Longshot#1551 reports pipeline caches not reused on Windows/Adreno, a different GPU/OS condition: https://git.magicbean.dev/MagicBeanDev/Longshot/issues/1551 - Longshot#1418 fixed unnecessary pipeline rebuilds after a no-shader-change update and is included since nightly-476: https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/1418
Author
Collaborator

From Discord:

Additional context from the reporter: they say Fritchii compiles shaders during gameplay and has no shader wait, and argue Longshot should not require all shader preparation before play. This is a comparison/request for a different shader compilation approach; they provided no reproduction details for Fritchii. Their existing report already covers the CachyOS build and pipeline cache reuse failures and the measured durations.

From [Discord](https://discord.com/channels/1569849200047427645/1572595870405500959/1572595870405500959): Additional context from the reporter: they say Fritchii compiles shaders during gameplay and has no shader wait, and argue Longshot should not require all shader preparation before play. This is a comparison/request for a different shader compilation approach; they provided no reproduction details for Fritchii. Their existing report already covers the CachyOS build and pipeline cache reuse failures and the measured durations.
Author
Collaborator

From Discord:

Reporter asks for a post acknowledging that, in their view, a build from roughly three builds ago was better. This is an opinion/communication request rather than new reproducible evidence or a fix; the tracked issue already includes their reported cache failures and timings.

From [Discord](https://discord.com/channels/1569849200047427645/1572595870405500959/1572595870405500959): Reporter asks for a post acknowledging that, in their view, a build from roughly three builds ago was better. This is an opinion/communication request rather than new reproducible evidence or a fix; the tracked issue already includes their reported cache failures and timings.
Author
Collaborator

From Discord:

The reporter clarifies the complaint as a straight regression: they say the builder repeatedly rereads the ROM and performs a full build, and shader preparation repeats on every launch. The issue already records these symptoms and reported durations, plus the Fritchii comparison and request for a public comparison of earlier builds; no exact additional hardware or logs were supplied.

From [Discord](https://discord.com/channels/1569849200047427645/1572595870405500959/1572595870405500959): The reporter clarifies the complaint as a straight regression: they say the builder repeatedly rereads the ROM and performs a full build, and shader preparation repeats on every launch. The issue already records these symptoms and reported durations, plus the Fritchii comparison and request for a public comparison of earlier builds; no exact additional hardware or logs were supplied.
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
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#1581
No description provided.