Repository navigation
CMake: declare BUILD_TESTING before any subproject; llama.cpp b11457 → b11462 - #480
Merged
Merged
Conversation
The CUDA build of the Publish run on main (37550451678) failed linking jllama_test: gtest discovery runs the binary at build time and it needs libcuda.so.1, which the GPU-less runner does not have. The job does not ask for tests at all. Since the llama.cpp bump our CUDA builds pass -DGGML_CUDA_CCCL_VERSION=v3.4.3, so ggml fetches CCCL, and CCCL's CMakeLists calls include(CTest) unconditionally -- which creates BUILD_TESTING as a cache variable defaulting to ON before our option(BUILD_TESTING ... OFF) runs, turning it into a no-op. The last green CUDA build (36935020061) never built jllama_test. The option is now declared before the first FetchContent_MakeAvailable(). Reproduced without CUDA by injecting include(CTest) into ggml's project() call (CMAKE_PROJECT_ggml_INCLUDE): before the change BUILD_TESTING ends up ON and jllama_test is generated, after it stays OFF; -DBUILD_TESTING=ON still enables the tests. The Windows CUDA job passes the same pin. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AytmJF9faEiQEVt6eetQS2
One upstream commit, #30067: the Hexagon backend's copy/concat overhaul. No natives jar builds Hexagon, so nothing in it reaches a shipped library; 112 KiB, but a single commit cannot be split. All ten patches apply unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AytmJF9faEiQEVt6eetQS2
Four commits, backends only: SYCL drops its separate flash-attention KV buffers (#27689; its CMakeLists globs the sources) and gains an IQ3_S multi-column MMVQ (#29500), Vulkan reads AMD UMA memory through the device copy path instead of the slow write-combined mapping (#30049), WebGPU (#27069, not built). All ten patches apply unchanged; b11462 is the newest release. CLAUDE.md now records why BUILD_TESTING must be declared before the subprojects (CCCL's include(CTest)), next to the CCCL pin. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AytmJF9faEiQEVt6eetQS2
bernardladenthin
had a problem deploying
to
startgate
October 7, 2026 10:21 — with 
GitHub Actions
Error
GitHub Actions
Error
bernardladenthin
had a problem deploying
to
maven-central
October 7, 2026 10:21 — with
GitHub Actions
Failure
GitHub Actions
Failure
bernardladenthin
had a problem deploying
to
maven-central
October 7, 2026 10:21 — with
GitHub Actions
Failure
GitHub Actions
Failure
|
Review Summary This PR fixes a CMake BUILD_TESTING cache-variable shadowing issue and bumps llama.cpp to b11462. CMake Fix Status: Correct. The option(BUILD_TESTING OFF) is now declared before any FetchContent_MakeAvailable(), preventing CCCLs unconditional include(CTest) from creating a conflicting cache variable. The explanation in both CMakeLists.txt and CLAUDE.md is clear and references the exact CUDA job failure. Users can still enable tests with -DBUILD_TESTING=ON. Version Bump Status: Methodical and correct. All 10 patches apply unchanged. Backend changes only upstream (Hexagon, SYCL, Vulkan, WebGPU). All version references updated: CMakeLists.txt, README.md, CLAUDE.md, LlamaCppVersion.java, CHANGELOG.md, docs/history/llama-cpp-breaking-changes.md. Test Coverage
Verdict: No issues found. Ready to merge. |
|
7 of 8 tasks
This branch had an error being deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
jllama_test's gtest discovery could not loadlibcuda.so.1(Linux) ornvcuda.dll(Windows,0xc0000135).-DGGML_CUDA_CCCL_VERSION=v3.4.3) makes ggml fetch CCCL, and CCCL callsinclude(CTest)unconditionally. That createsBUILD_TESTINGas a cache variable defaulting to ON. Ouroption(BUILD_TESTING ... OFF)was declared after the firstFetchContent_MakeAvailable(), so it was a no-op and both CUDA jobs built the tests.llama/CMakeLists.txt, before any subproject.-DBUILD_TESTING=ONstill enables the tests.CLAUDE.mdnotes why the option has to stay there..githubchanges upstream.0016stays.docs/history/llama-cpp-breaking-changes.md; the pin is updated in CMake, README, CLAUDE.md andLlamaCppVersion.Test plan
BUILD_TESTING=ONandjllama_testgenerated;-DBUILD_TESTING=ONstill builds the tests.ctest598/598 passed, including the Kolibri-1 tests.mvn clean testwithNativeLibraryLoadSmokeTestandLlamaLoggerTest: 10 tests, 0 skipped. This includes the check that the linked build info matchesLlamaCppVersion(b11462).Related issues / PRs
Follows #479. Fixes the CUDA failures of run 37550451678.
🤖 Generated with Claude Code
https://claude.ai/code/session_01AytmJF9faEiQEVt6eetQS2
Generated by Claude Code