Add submission package for microsoft/TypeScript #55091

contrib/typescript-55091/PR_BODY.md

@@ -0,0 +1,74 @@
+Fixes #55091
+
+### Problem
+
+Enum members whose initializers evaluate to `Infinity`, `-Infinity`, or `NaN` are
+emitted as the global identifiers of those names. Identifiers resolve in the scope of
+the *emitted* code rather than at the enum declaration, so a local binding that shadows
+`Infinity` or `NaN` silently changes the member's value at runtime:
+
+```ts
+{
+  let Infinity = 3;
+  enum A {
+    X = 1 / 0
+  }
+  console.log(A.X); // 3, expected Infinity
+}
+```
+
+emits
+
+```js
+A[A["X"] = Infinity] = "X";
+```
+
+This compiles with no diagnostic. `const enum` members are protected here by TS2477 and
+TS2478, but plain enums are not, so the value is wrong with nothing to indicate it.
+
+### Change
+
+`constantExpression` now emits `1 / 0`, `-1 / 0`, and `0 / 0` for these three values.
+They evaluate to exactly the same numbers and cannot be shadowed, which is the same
+reasoning behind emitting `void 0` rather than `undefined`.
+
+The `-Infinity` case prints as `-1 / 0` rather than `-(1 / 0)`; unary minus binds tighter
+than division, so no parentheses are needed.
+
+### Testing
+
+I added `tests/cases/conformance/enums/enumComputedNonFiniteShadowed.ts`, covering all
+three values both under a shadowing binding and without one.
+
+By hand, I built the compiler from this branch and ran the case above through it. Before
+the change the emitted program printed `Pos: 3 Neg: -3 Nan: 7`; after it, it printed
+`Pos: Infinity Neg: -Infinity Nan: NaN`.
+
+Five existing emit baselines change, in each case only `Infinity`/`NaN` becoming the
+division form: `enumConstantMembers`, `enumNaNValues`, `enumAutoIncrementValue`,
+`fakeInfinity2`, and `fakeInfinity3`. No `.types`, `.symbols`, or `.d.ts` baseline
+changes, since checking is untouched. `enumShadowedInfinityNaN` is unaffected, as that
+test references the identifiers directly rather than computing a constant.
+
+I ran `go test ./...` under `tsc/`: 62 packages pass. `internal/astnav` fails in my
+environment because it loads `node_modules/typescript/lib/typescript.js`, which I had not
+installed; I confirmed it fails identically with this change stashed. I did not run the
+`npx hereby` wrappers for that reason, but `gofmt -l` and `go vet ./internal/transformers/...`
+are both clean.
+
+### Key points
+
+The tradeoff is emit readability: `E[E["a"] = 1 / 0] = "a"` reads less clearly than
+`= Infinity` for the common, non-shadowed case. I chose the unconditional form because it
+is correct without needing to resolve `Infinity`/`NaN` at the emit site. If you would
+rather keep the identifier when it is demonstrably not shadowed, I am happy to rework it
+that way.
+
+This does not address declaration emit, which has the same underlying problem: the
+committed `fakeInfinity3.d.ts` baseline still carries TS1066 and TS2749 errors. That
+looked like a separate change, so I left it out rather than widen this one.
+
+### AI assistance
+
+This patch was written with AI assistance (Claude Code). I have read and understand the
+change and will be responding to review feedback myself.

contrib/typescript-55091/README.md

@@ -0,0 +1,82 @@
+# microsoft/TypeScript #55091 - submission package
+
+Status: **prepared, not submitted.** Submission is a manual step for the repository
+owner, for the reasons in "Why you submit this" below.
+
+## Record
+
+| Field | Value |
+|---|---|
+| Organization | Microsoft |
+| Repository | microsoft/TypeScript |
+| Issue | [#55091](https://github.com/microsoft/TypeScript/issues/55091) - `Help Wanted`, unassigned, milestone `Dormant` |
+| Prior art | [#55107](https://github.com/microsoft/TypeScript/pull/55107) by Andarist, closed 2026-03-24 in the post-6.0 mass closure, not rejected on merit |
+| Base commit | `9b323de` era clone of `main` (Go compiler; verify against latest before submitting) |
+| Files changed | 1 source, 1 new test, 8 baselines |
+| Diff | +43 / -24 in source; baselines mechanical |
+| PR body | `PR_BODY.md` |
+| Patch | `fix.patch` |
+
+## The bug
+
+`tsc/internal/transformers/tstransforms/utilities.go`, `constantExpression()` emitted
+bare `Infinity` / `-Infinity` / `NaN` identifiers for enum members whose initializers
+fold to those values. Identifiers resolve in the emitted code's scope, so shadowing
+changes the value silently. Reproduced on published TypeScript 7.0.2:
+
+```
+$ cat repro.ts
+{
+  let Infinity = 3;
+  enum A { X = 1 / 0 }
+  console.log(A.X);
+}
+$ tsc --target es2020 repro.ts && node repro.js
+3          # expected: Infinity
+```
+
+No diagnostic is issued. `const enum` is guarded by TS2477/TS2478; plain `enum` is not.
+
+## Verification performed
+
+- `go build ./...` under `tsc/` - clean
+- `go test ./internal/testrunner/ -run TestLocal` - passes with updated baselines
+- `go test ./...` - 62 packages pass; `internal/astnav` fails on a missing
+  `node_modules/typescript`, confirmed identical with the fix stashed
+- `gofmt -l` clean, `go vet ./internal/transformers/...` clean
+- Built the patched compiler and diffed real emit/runtime before and after
+
+Not run: `npx hereby test` / `lint` / `check:format`, because npm dependencies were not
+installed in the build environment. Run these before submitting.
+
+## Why you submit this
+
+microsoft/TypeScript `CONTRIBUTING.md` states that a pull request is acceptable only if
+"a specific human operator has chosen this specific issue, intends to shepherd the change
+through review themselves, and will be the one responding to feedback in their own
+personal workflow." It also requires that AI assistance be disclosed in the PR
+description, and closes undisclosed AI-authored PRs without review. `PR_BODY.md` contains
+that disclosure - keep it.
+
+So: read the diff until you can defend it, then submit it yourself and answer review
+yourself. That is the path the project actually accepts.
+
+## Steps
+
+1. Read `fix.patch` and make sure you can explain why `1 / 0` is correct and why the
+   identifier form was not.
+2. Fork `microsoft/TypeScript`, clone your fork, branch.
+3. `git am < fix.patch` (or `git apply`), rebase onto current `main`, re-run the tests
+   above plus the `hereby` wrappers.
+4. Push and open the PR using `PR_BODY.md`.
+5. Note the PR template requires an associated issue in the `Backlog` milestone; #55091
+   is currently `Dormant`. Expect maintainers to re-milestone it or to say the change is
+   not wanted right now - that is a real possibility worth accepting up front.
+
+## Honest assessment
+
+Merge probability is moderate, not high. In favour: the bug is real, silent, reproducible
+on the shipping compiler, the fix is small, and the approach has precedent. Against: a
+maintainer called this scenario unlikely on the earlier PR, the issue sits in `Dormant`,
+and the emit-readability tradeoff may draw a request to make it conditional. It is a
+legitimate contribution either way.