[danusha2345]

Test-only, no changelog entry.

Problem

CODEGRAPH_KERNEL=0 is the switch that sends every language through wasm, and running the suite with it is how the wasm path gets exercised on a machine that has the kernel built. Two suites fail in that run instead of skipping:

$ CODEGRAPH_KERNEL=0 npx vitest run __tests__/kernel-*.test.ts
 FAIL  __tests__/kernel-deep-nesting.test.ts > kernel deep-nesting guard (#1581)
 FAIL  __tests__/kernel-retry-materialize.test.ts > kernel buffer-transport storage (#1541) > storeExtractionResult persists the decoded nodes of a raw kernel result
 Test Files  2 failed | 16 passed (18)

kernel-deep-nesting asserts in beforeAll that at least one language is routed to the kernel, and none is. kernel-retry-materialize asks for a raw kernel result and gets null. Both skip only when the .node file is missing. The other kernel-* suites call the kernel directly or clear the variable themselves and pass either way.

Fix

The two describe.skipIf(!kernelBuilt) blocks also skip when CODEGRAPH_KERNEL === '0' — the same outcome as a checkout without a kernel. The kernel-independent zero-node row self-heal block in kernel-retry-materialize and the CODEGRAPH_KERNEL_EXPECT=1 presence checks are untouched.

Tests

All 18 kernel-* files plus c-fnptr-kernel-sweep:

• with the kernel (CODEGRAPH_KERNEL_EXPECT=1): 18 files, 258 tests pass, nothing newly skipped;
• with CODEGRAPH_KERNEL=0: 17 files pass, kernel-deep-nesting is skipped, and one test of kernel-retry-materialize is skipped while its self-heal test runs. On main the same command ends with the two failures above.

Update (rebased onto ed199e6). #2426 added a third kernel-only block, "C++ brace scopes: native kernel parity (error-extract hatch)" in __tests__/cpp-brace-scopes.test.ts, which fails the same way under CODEGRAPH_KERNEL=0 (4 tests). It gets the same one-line condition here. The three files now give 26/26 with the kernel and 13 passed / 13 skipped with CODEGRAPH_KERNEL=0.

🤖 Generated with Claude Code

[colbymchenry]

Thanks @danusha2345! A kernel-off run that fails instead of skipping hides real regressions on the wasm path, so this is a good fix, and your condition matches the loader's own kill switch exactly.

We landed your commit unchanged in #2848, with you as the author. We also ran every other test file that references the kernel with CODEGRAPH_KERNEL=0, and these three were the only ones that failed that way. Closing this in favor of #2848, which is now merged.