bug: circular deps error

[sroussey]

Issue description

ReferenceError: Cannot access 'isCI' before initialization.

Expected Behavior

node-llama-cpp’s dist/config.js has a circular dependency: config.js loads binariesGithubRelease.js, which imports back from config.js. While config.js is paused on await getBinariesGithubRelease(), another arm of the graph can load createModelDownloader.js, which imports isCI from config.js. In the published file, isCI was declared after that top-level await, so it was still in the temporal dead zone → ReferenceError: Cannot access 'isCI' before initialization.

Actual Behavior

ReferenceError: Cannot access 'isCI' before initialization.

Steps to reproduce

It is a timing error based on multiple vitest threads running tests in parallel

My Environment

Mac devcontainer, but also github actions

OS: Debian GNU/Linux 12 (bookworm) (arm64)
Node: 24.14.1 (arm64)

node-llama-cpp: 3.18.1
Prebuilt binaries: b8390

Vulkan: Vulkan is detected, but using it failed
To resolve errors related to Vulkan, see the Vulkan guide: https://node-llama-cpp.withcat.ai/guide/Vulkan

CPU model: unknown
Math cores: 12
Used RAM: 54.12% (12.71GB/23.49GB)
Free RAM: 45.87% (10.78GB/23.49GB)
Used swap: 9.02% (2.21GB/24.49GB)
Max swap size: 24.49GB
mmap: supported

Additional Context

https://github.com/workglow-dev/libs/actions/runs/25813888963/job/75837336404?pr=4941s
Run bun run test:bun:ai-provider-llamacpp
$ bun scripts/test.ts bun integration provider-llamacpp

Running integration tests in sections [provider-llamacpp] — 4 file(s)

bun test v1.3.14 (0d9b296a) 1x PARALLEL

packages/test/src/test/ai-provider/LlamaCpp_ChatWrapper.integration.test.ts:

packages/test/src/test/ai-provider/LlamaCpp_Generic.integration.test.ts:

packages/test/src/test/ai-provider/LlamaCpp_NativeToolCalling.integration.test.ts:

packages/test/src/test/ai-provider/LlamaCppProviderIntegration.integration.test.ts:
PermanentJobError: Provider LOCAL_LLAMACPP failed for ModelDownloadTask: Cannot access 'isCI' before initialization.

--- Error diagnostics ---
ReferenceError: Cannot access 'isCI' before initialization.
ReferenceError: Cannot access 'isCI' before initialization.
at _init (/home/runner/work/libs/libs/node_modules/.bun/node-llama-cpp@3.18.1+7524df1edfed9f02/node_modules/node-llama-cpp/dist/utils/createModelDownloader.js:284:27)
at async _create (/home/runner/work/libs/libs/node_modules/.bun/node-llama-cpp@3.18.1+7524df1edfed9f02/node_modules/node-llama-cpp/dist/utils/createModelDownloader.js:366:31)
at async (/home/runner/work/libs/libs/providers/node-llama-cpp/dist/ai-runtime.js:368:28)
at async execute (/home/runner/work/libs/libs/packages/ai/src/job/AiJob.ts:278:13)
at async execute (/home/runner/work/libs/libs/packages/ai/src/execution/QueuedExecutionStrategy.ts:59:17)
at async execute (/home/runner/work/libs/libs/packages/ai/src/task/base/AiTask.ts:165:20)
at async executeTask (/home/runner/work/libs/libs/packages/task-graph/src/task/TaskRunner.ts:480:36)
at async run (/home/runner/work/libs/libs/packages/task-graph/src/task/TaskRunner.ts:214:26)
at async runTask (/home/runner/work/libs/libs/packages/task-graph/src/task-graph/TaskGraphRunner.ts:514:39)
at async runLoop (/home/runner/work/libs/libs/packages/task-graph/src/task-graph/RunScheduler.ts:313:19)
at async runGraph (/home/runner/work/libs/libs/packages/task-graph/src/task-graph/TaskGraphRunner.ts:205:47)
at processTicksAndRejections (native:7:39)
PermanentJobError {
message: "Provider LOCAL_LLAMACPP failed for ModelDownloadTask: Cannot access 'isCI' before initialization.\n\n--- Error diagnostics ---\nReferenceError: Cannot access 'isCI' before initialization.\nReferenceError: Cannot access 'isCI' before initialization.\n at _init (/home/runner/work/libs/libs/node_modules/.bun/node-llama-cpp@3.18.1+7524df1edfed9f02/node_modules/node-llama-cpp/dist/utils/createModelDownloader.js:284:27)\n at async _create (/home/runner/work/libs/libs/node_modules/.bun/node-llama-cpp@3.18.1+7524df1edfed9f02/node_modules/node-llama-cpp/dist/utils/createModelDownloader.js:366:31)\n at async (/home/runner/work/libs/libs/providers/node-llama-cpp/dist/ai-runtime.js:368:28)\n at async execute (/home/runner/work/libs/libs/packages/ai/src/job/AiJob.ts:278:13)\n at async execute (/home/runner/work/libs/libs/packages/ai/src/execution/QueuedExecutionStrategy.ts:59:17)\n at async execute (/home/runner/work/libs/libs/packages/ai/src/task/base/AiTask.ts:165:20)\n at async executeTask (/home/runner/work/libs/libs/packages/task-graph/src/task/TaskRunner.ts:480:36)\n at async run (/home/runner/work/libs/libs/packages/task-graph/src/task/TaskRunner.ts:214:26)\n at async runTask (/home/runner/work/libs/libs/packages/task-graph/src/task-graph/TaskGraphRunner.ts:514:39)\n at async runLoop (/home/runner/work/libs/libs/packages/task-graph/src/task-graph/RunScheduler.ts:313:19)\n at async runGraph (/home/runner/work/libs/libs/packages/task-graph/src/task-graph/TaskGraphRunner.ts:205:47)\n at processTicksAndRejections (native:7:39)",
name: "PermanentJobError",
stack: "Error\n at classifyProviderError (/home/runner/work/libs/libs/packages/ai/src/job/AiJob.ts:210:14)\n at execute (/home/runner/work/libs/libs/packages/ai/src/job/AiJob.ts:287:13)\n at async execute (/home/runner/work/libs/libs/packages/ai/src/execution/QueuedExecutionStrategy.ts:59:17)\n at async execute (/home/runner/work/libs/libs/packages/ai/src/task/base/AiTask.ts:165:20)\n at async executeTask (/home/runner/work/libs/libs/packages/task-graph/src/task/TaskRunner.ts:480:36)\n at async run (/home/runner/work/libs/libs/packages/task-graph/src/task/TaskRunner.ts:214:26)\n at async runTask (/home/runner/work/libs/libs/packages/task-graph/src/task-graph/TaskGraphRunner.ts:514:39)\n at async runLoop (/home/runner/work/libs/libs/packages/task-graph/src/task-graph/RunScheduler.ts:313:19)\n at async runGraph (/home/runner/work/libs/libs/packages/task-graph/src/task-graph/TaskGraphRunner.ts:205:47)\n at async run (/home/runner/work/libs/libs/packages/task-graph/src/task-graph/Workflow.ts:268:39)",
retryable: false,
toString: [Function: toString],
}

Relevant Features Used

• Metal support
• CUDA support
• Vulkan support
• Grammar
• Function calling

Are you willing to resolve this issue by submitting a Pull Request?

Yes, I have the time, and I know how to start.

[sroussey]

applying a patchedDependencies

  "patchedDependencies": {
    "node-llama-cpp@3.18.1": "patches/node-llama-cpp@3.18.1.patch"
  },

patches/node-llama-cpp@3.18.1.patch:

diff --git a/dist/config.js b/dist/config.js
index 1ab00734010c423268a5cd01fcb1670071b20512..4c36fa189fb7091ab259a00f879e41e9987e075e 100644
--- a/dist/config.js
+++ b/dist/config.js
@@ -33,7 +33,8 @@ export const localXpacksCacheDirectory = path.join(xpackDirectory, "cache");
 export const buildMetadataFileName = "_nlcBuildMetadata.json";
 export const xpmVersion = "^0.16.3";
 export const builtinLlamaCppGitHubRepo = "ggml-org/llama.cpp";
-export const builtinLlamaCppRelease = await getBinariesGithubRelease();
+// Define before top-level await: createModelDownloader imports `isCI` while this module may still be
+// suspended on `getBinariesGithubRelease()` (config ↔ binariesGithubRelease circular dependency).
 export const isCI = env.get("CI")
     .default("false")
     .asBool();
@@ -41,6 +42,7 @@ export const isRunningInsideGoogleColab = env.get("COLAB_RELEASE_TAG")
     .default("")
     .asString() !== "";
 export const useCiLogs = isCI || isRunningInsideGoogleColab;
+export const builtinLlamaCppRelease = await getBinariesGithubRelease();
 export const defaultLlamaCppGitHubRepo = env.get("NODE_LLAMA_CPP_REPO")
     .default(builtinLlamaCppGitHubRepo)
     .asString();

[sroussey]

Basically, please move export const builtinLlamaCppRelease = await getBinariesGithubRelease(); to the end of the file.

[sroussey]

This seems to be an issue only in bun: oven-sh/bun#30651