Summary
The CI / build workflow has been red on main since at least 2026-05-28 10:09Z (commit c871f7f). Three tests fail consistently:
1. E2E Node-CJS Codegen #4 — test_vscode_extension_adapter_override
Test in test/test_e2e.ml:3056. Asserts: when ~vscode_extension_adapter:"../local/adapter.cjs" is passed, the emitted CJS should not contain require("@hyperpolymath/affine-vscode").
The check currently fails because the embedded adapter source (packages/affine-vscode/mod.js, embedded at compile time per #380) contains literal require("@hyperpolymath/affine-vscode") strings in documentation comments / example usage that aren't actual require sites.
2. E2E Node-CJS Codegen #5 — test_vscode_extension_no_lc
Test in test/test_e2e.ml:3069. Same class of issue — asserts require("vscode-languageclient/node") is NOT present when no_lc:true, but the embedded adapter source has a delimiter comment mentioning that specifier.
Likely fix: either tighten the assertion to look for require(...) function-call form not the bare string (parse as JS rather than substring-match), or strip documentation comments from the embedded adapter source before substring assertions.
3. walker-phase2c-parity #2 — test_walker_finds_untyped_exception
Test in tools/res-to-affine/test/test_walker.ml:148. Walker reports lines 19 and 28 as having untyped-exception but misses line 22 (| Js.Exn.Error(_) => None in a catch-arm).
Root cause identified at tools/res-to-affine/walker.ml:434-440: the developer-comment-acknowledged fallback for catch-arm Js.Exn.* patterns only matches leaf nodes (node.children = []), but tree-sitter parses Js.Exn.Error(_) as a compound node with children, so the fallback never fires.
Likely fix: extend the fallback to recurse into compound nodes' leaves looking for Js.Exn-prefixed text, OR add a typed branch for tree-sitter's catch-arm constructor pattern node type.
Impact
Surfaced by
Suggested follow-up
Three small surgical fixes (one per test). Probably 1 PR per fix to keep history clean, or one bundled fix-ci PR.
Summary
The
CI / buildworkflow has been red on main since at least 2026-05-28 10:09Z (commitc871f7f). Three tests fail consistently:1.
E2E Node-CJS Codegen #4—test_vscode_extension_adapter_overrideTest in
test/test_e2e.ml:3056. Asserts: when~vscode_extension_adapter:"../local/adapter.cjs"is passed, the emitted CJS should not containrequire("@hyperpolymath/affine-vscode").The check currently fails because the embedded adapter source (
packages/affine-vscode/mod.js, embedded at compile time per #380) contains literalrequire("@hyperpolymath/affine-vscode")strings in documentation comments / example usage that aren't actual require sites.2.
E2E Node-CJS Codegen #5—test_vscode_extension_no_lcTest in
test/test_e2e.ml:3069. Same class of issue — assertsrequire("vscode-languageclient/node")is NOT present whenno_lc:true, but the embedded adapter source has a delimiter comment mentioning that specifier.Likely fix: either tighten the assertion to look for
require(...)function-call form not the bare string (parse as JS rather than substring-match), or strip documentation comments from the embedded adapter source before substring assertions.3.
walker-phase2c-parity #2—test_walker_finds_untyped_exceptionTest in
tools/res-to-affine/test/test_walker.ml:148. Walker reports lines 19 and 28 as having untyped-exception but misses line 22 (| Js.Exn.Error(_) => Nonein a catch-arm).Root cause identified at
tools/res-to-affine/walker.ml:434-440: the developer-comment-acknowledged fallback for catch-armJs.Exn.*patterns only matches leaf nodes (node.children = []), but tree-sitter parsesJs.Exn.Error(_)as a compound node with children, so the fallback never fires.Likely fix: extend the fallback to recurse into compound nodes' leaves looking for
Js.Exn-prefixed text, OR add a typed branch for tree-sitter's catch-arm constructor pattern node type.Impact
CIworkflow'sbuildjob fails)Surfaced by
Suggested follow-up
Three small surgical fixes (one per test). Probably 1 PR per fix to keep history clean, or one bundled fix-ci PR.