# Bug Report: Shared Object Cache Fails to Reuse Compiled Objects Across Nightly Updates on LinuxInstall: #1906

Open
opened 2026-11-20 20:07:51 +00:00 by bigwiggler · 3 comments

What you used

Linux AppImage

Version

148

Channel

Nightly

Where it stopped

Installing or updating the launcher or the builder

The error

# Bug Report: Shared Object Cache Fails to Reuse Compiled Objects Across Nightly Updates on Linux

**Platform:** CachyOS Linux  
**Architecture:** x86_64  
**Launcher:** Longshot Launcher nightly-148-985cc614  
**Affected builds:** nightly-618-2210931c and nightly-625-a104182b

## Summary

The Longshot shared object cache does not appear to reuse previously compiled objects when switching between nightly builds.

I performed a full build using nightly 618 specifically to establish a populated cache and test whether object reuse would work. I then updated to nightly 625 to verify the fix.

Nightly 625 proceeded to compile again instead of reusing the objects generated by nightly 618. The same issue had already occurred when testing nightly 618.

This defeats the purpose of the shared object store and causes unnecessary, lengthy recompilation on nightly updates.

## Build log evidence

Nightly 625 reported:

```text
INF build: the game is built for this PC's CPU (znver1_znver1_9a7fa803d6)
INF [Read the game files] Reading your game files
INF the game was read from this .nup before: reading it again is skipped
INF [Translate] Translating the game to PC code
INF the game was translated with these tools before: translating it again is skipped
INF [Compile] Compiling
INF build: shared A64 runtime helpers at O2 (RX_A64_RT_O2=0: cold-code level)
INF build: 58 lifted chunks without -O1's module passes (LONGSHOT_LIGHT_PASSES, functions under 1024 KB)
INF object keys: game-code headers 16f3cadc1c, flags 775a1bd4ce, hot flags 2d5dfdb1c0, lifted code 7d2aed8281 (81 files)
INF object store /home/[USER]/.cache/longshot/launcher/work/obj-store/3fcd1e7be0def429: 0 objects from it
INF 561 files, 6 jobs, -DLONGSHOT_NATIVE_CPU=znver1_znver1_9a7fa803d6 -O2
```

## Cache inspection

The shared cache is present and populated.

**Object store directory:**
```text
~/.cache/longshot/launcher/work/obj-store/3fcd1e7be0def429/
```

**Cached compiled objects:** 1,890 `.o` files.

Example cached object filenames:

```text
001117fd1a554da05cc137560e8ad36cda22b234.o
00217621b50fecf99485e16a8e6f7d9103c4e7b9.o
0115e5e75fce8d024361291f3a9da714f6f4e47f.o
01475e4c41e47ba00be15f74c83d160f8da373c4.o
0182b9b7d98ed4a4ee291c2bf9c79a1307abde5f.o
```

There is only one compiler/system-header partition directory, rather than separate directories for each nightly.

The object cache and lift store are also populated.

## Expected behavior

After completing a build on nightly 618, switching to nightly 625 should reuse compatible previously compiled objects from the shared object store.

Only objects affected by actual source, compiler, or build-flag changes should need recompilation.

## Actual behavior

The nightly 625 build reports:

```text
0 objects from it
```

It then compiles 561 files despite the existing populated object store.

The same general behavior occurred during the nightly 618 test.

## Request

Please investigate why the shared object store is finding zero reusable objects across nightly builds.

In particular, please verify:

1. Whether the compiler/system-header partition key is stable across nightly updates.
2. Whether object keys are changing unexpectedly between builds.
3. Whether build flags or profile paths are preventing otherwise compatible objects from matching.
4. Whether the launcher is correctly passing `LONGSHOT_OBJ_STORE` and `LONGSHOT_LIFT_STORE` to every nightly build.
5. Whether object-store lookup and hard-link reuse are functioning correctly on Linux.

This appears to be a regression or configuration issue in the shared build cache, not an empty or missing cache.

**The existing cache has been preserved; its contents have not been deleted.**

Operating system and free space on the output drive

Cachy OS

Visual Studio or Build Tools version, and Windows SDK version

No response

Log

There is no log. It builds fine it just ingnores all the old work and leaves it all over your machine. Its a huge waste of time.

ON older ryzen cpus it takes 2hrs to build from scratch. You keep fixing it and breaking it again! its ridiculous. Just leave it alone when it works ffs

Before you send it

  • It still fails on the newest release.
  • I searched the open issues for this error.
  • I'm not attaching game files or anything built from them.
### What you used Linux AppImage ### Version 148 ### Channel Nightly ### Where it stopped Installing or updating the launcher or the builder ### The error ````text # Bug Report: Shared Object Cache Fails to Reuse Compiled Objects Across Nightly Updates on Linux **Platform:** CachyOS Linux **Architecture:** x86_64 **Launcher:** Longshot Launcher nightly-148-985cc614 **Affected builds:** nightly-618-2210931c and nightly-625-a104182b ## Summary The Longshot shared object cache does not appear to reuse previously compiled objects when switching between nightly builds. I performed a full build using nightly 618 specifically to establish a populated cache and test whether object reuse would work. I then updated to nightly 625 to verify the fix. Nightly 625 proceeded to compile again instead of reusing the objects generated by nightly 618. The same issue had already occurred when testing nightly 618. This defeats the purpose of the shared object store and causes unnecessary, lengthy recompilation on nightly updates. ## Build log evidence Nightly 625 reported: ```text INF build: the game is built for this PC's CPU (znver1_znver1_9a7fa803d6) INF [Read the game files] Reading your game files INF the game was read from this .nup before: reading it again is skipped INF [Translate] Translating the game to PC code INF the game was translated with these tools before: translating it again is skipped INF [Compile] Compiling INF build: shared A64 runtime helpers at O2 (RX_A64_RT_O2=0: cold-code level) INF build: 58 lifted chunks without -O1's module passes (LONGSHOT_LIGHT_PASSES, functions under 1024 KB) INF object keys: game-code headers 16f3cadc1c, flags 775a1bd4ce, hot flags 2d5dfdb1c0, lifted code 7d2aed8281 (81 files) INF object store /home/[USER]/.cache/longshot/launcher/work/obj-store/3fcd1e7be0def429: 0 objects from it INF 561 files, 6 jobs, -DLONGSHOT_NATIVE_CPU=znver1_znver1_9a7fa803d6 -O2 ``` ## Cache inspection The shared cache is present and populated. **Object store directory:** ```text ~/.cache/longshot/launcher/work/obj-store/3fcd1e7be0def429/ ``` **Cached compiled objects:** 1,890 `.o` files. Example cached object filenames: ```text 001117fd1a554da05cc137560e8ad36cda22b234.o 00217621b50fecf99485e16a8e6f7d9103c4e7b9.o 0115e5e75fce8d024361291f3a9da714f6f4e47f.o 01475e4c41e47ba00be15f74c83d160f8da373c4.o 0182b9b7d98ed4a4ee291c2bf9c79a1307abde5f.o ``` There is only one compiler/system-header partition directory, rather than separate directories for each nightly. The object cache and lift store are also populated. ## Expected behavior After completing a build on nightly 618, switching to nightly 625 should reuse compatible previously compiled objects from the shared object store. Only objects affected by actual source, compiler, or build-flag changes should need recompilation. ## Actual behavior The nightly 625 build reports: ```text 0 objects from it ``` It then compiles 561 files despite the existing populated object store. The same general behavior occurred during the nightly 618 test. ## Request Please investigate why the shared object store is finding zero reusable objects across nightly builds. In particular, please verify: 1. Whether the compiler/system-header partition key is stable across nightly updates. 2. Whether object keys are changing unexpectedly between builds. 3. Whether build flags or profile paths are preventing otherwise compatible objects from matching. 4. Whether the launcher is correctly passing `LONGSHOT_OBJ_STORE` and `LONGSHOT_LIFT_STORE` to every nightly build. 5. Whether object-store lookup and hard-link reuse are functioning correctly on Linux. This appears to be a regression or configuration issue in the shared build cache, not an empty or missing cache. **The existing cache has been preserved; its contents have not been deleted.** ```` ### Operating system and free space on the output drive Cachy OS ### Visual Studio or Build Tools version, and Windows SDK version _No response_ ### Log There is no log. It builds fine it just ingnores all the old work and leaves it all over your machine. Its a huge waste of time. ON older ryzen cpus it takes 2hrs to build from scratch. You keep fixing it and breaking it again! its ridiculous. Just leave it alone when it works ffs ### Before you send it - [x] It still fails on the newest release. - [x] I searched the open issues for this error. - [x] I'm not attaching game files or anything built from them.
Member

Possibly related, for context:

  • issue #1581 Game: On CachyOS, pipeline cache is rebuilt on every launch and build assets are not reused — CachyOS reports compiled build assets not being reused; current issue isolates object-store misses across nightlies.
  • issue #1076 Idea: preserve reusable build outputs after unrelated post-link failures — Prior report/fix context for removing compiled objects from shared store.
  • PR #371 perf(build): one object store for every work folder, so a new one compiles only what no other built — Merged implementation of shared object-store lookup and reuse.
  • PR MagicBeanDev/LongshotLauncher#72 feat(build): Longshot's object and lift stores in the build files folder, so switching versions reuses earlier builds' work — Merged launcher integration passing object/lift store paths to builds.
Possibly related, for context: - issue #1581 Game: On CachyOS, pipeline cache is rebuilt on every launch and build assets are not reused — CachyOS reports compiled build assets not being reused; current issue isolates object-store misses across nightlies. - issue #1076 Idea: preserve reusable build outputs after unrelated post-link failures — Prior report/fix context for removing compiled objects from shared store. - PR #371 perf(build): one object store for every work folder, so a new one compiles only what no other built — Merged implementation of shared object-store lookup and reuse. - PR MagicBeanDev/LongshotLauncher#72 feat(build): Longshot's object and lift stores in the build files folder, so switching versions reuses earlier builds' work — Merged launcher integration passing object/lift store paths to builds.
bigwiggler changed title from Install: to # Bug Report: Shared Object Cache Fails to Reuse Compiled Objects Across Nightly Updates on LinuxInstall: 2026-11-20 20:08:07 +00:00
Author

You are turning a functional recomp into cycles of regression. PLS do not break features in your new updates.

Longshot should have stayed part of hookshot ffs. You guys are just hording ideas when 1 master project would have been the best result.

Now we have 5+ half ass recomps instead of one amazing one.

You are turning a functional recomp into cycles of regression. PLS do not break features in your new updates. Longshot should have stayed part of hookshot ffs. You guys are just hording ideas when 1 master project would have been the best result. Now we have 5+ half ass recomps instead of one amazing one.
Contributor

Same cause as #2108, corrected with the code and the 618->625 history here: #2108 (comment) — the chunk key already hashes the generated C, your update reused the translation, and the chunks recompiled because #1727 changed the splitter flag (and most chunks' split text). A content-keyed fix for the split stage helps future splitter edits that leave a chunk's pieces unchanged; it does not undo that update. The host/renderer half is #2199.

Same cause as #2108, corrected with the code and the 618->625 history here: https://git.magicbean.dev/MagicBeanDev/Longshot/issues/2108#issuecomment-72234 — the chunk key already hashes the generated C, your update reused the translation, and the chunks recompiled because #1727 changed the splitter flag (and most chunks' split text). A content-keyed fix for the split stage helps future splitter edits that leave a chunk's pieces unchanged; it does not undo that update. The host/renderer half is #2199.
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#1906
No description provided.