fix(toml): emit TOML literals for Infinity/NaN/Date in arrays

[spokodev]

Primitive and mixed array stringification went through code paths that produced invalid TOML for non-finite numbers and Date values. The issue author flagged six distinct cases in #7162; this PR addresses the four with a deterministic spec-aligned fix and leaves the two null-related ones for maintainer guidance, since they are a design call.

What changed

#arrayDeclaration (primitive arrays) previously routed values through JSON.stringify, which has no representation for Infinity, -Infinity, NaN, or TOML datetime literals. #printAsInlineValue (mixed arrays) returned numbers verbatim (so Infinity printed as the JS identifier) and wrapped Dates in string quotes.

A new private helper #printPrimitive produces TOML-compliant literals for Date, string, RegExp, number (including inf/-inf/nan) and boolean. Both array paths now route primitives through it.

Behaviour for null/undefined inside arrays is unchanged - the code still throws Should never reach. The issue author flagged this as a design call (TOML has no null), so it stays as-is until maintainers decide between a clearer error message, silent skip, or another shape.

Before/after

// primitive array
stringify({x: [Infinity, -Infinity, NaN]})
// before: x = [null,null,null]
// after:  x = [inf,-inf,nan]

stringify({x: [new Date(0)]})
// before: x = [\"1970-01-01T00:00:00.000Z\"]
// after:  x = [1970-01-01T00:00:00.000]

// mixed array
stringify({x: [Infinity, -Infinity, NaN, {}]})
// before: x = [Infinity,-Infinity,NaN,{}]   (invalid TOML)
// after:  x = [inf,-inf,nan,{}]

stringify({x: [new Date(0), {}]})
// before: x = [\"1970-01-01T00:00:00.000\",{}]
// after:  x = [1970-01-01T00:00:00.000,{}]

Tests

Six new cases in toml/stringify_test.ts:

• positive: Infinity/-Infinity/NaN in primitive arrays render as inf/-inf/nan
• positive: Date in primitive arrays renders as a bare TOML datetime literal
• positive: Infinity/-Infinity/NaN in mixed arrays
• positive: Date in mixed arrays
• regression-check: finite numbers in primitive arrays keep their JSON-style representation (covers the new #printPrimitive path that replaced JSON.stringify)
• regression-check: scalar Infinity/NaN/Date declarations (not array elements) are unaffected

One existing test (stringify() handles mixed array) had a date inside a mixed inline-table that was previously emitted with quotes. The expected output is updated to the spec-aligned bare datetime literal.

`deno test --allow-all toml/` -> 689 passed, 0 failed, 111 ignored.
`deno fmt` and `deno lint` clean on the changed files.

Fixes #7162 (partial - null handling left for maintainer guidance).

[CLAassistant]

All committers have signed the CLA.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94.84%. Comparing base (cdf74a8) to head (3c68f02).
⚠️ Report is 18 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #7163      +/-   ##
==========================================
+ Coverage   94.57%   94.84%   +0.26%     
==========================================
  Files         636      617      -19     
  Lines       52142    51682     -460     
  Branches     9401     9358      -43     
==========================================
- Hits        49315    49016     -299     
+ Misses       2249     2121     -128     
+ Partials      578      545      -33     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[spokodev]

recheck

[bartlomieju]

Thanks for this — really thorough PR, and the write-up is excellent (clear before/after, honest about the deferred null design call, and well-scoped). I verified the positive fixes locally and they all work: [Infinity,-Infinity,NaN] -> [inf,-inf,nan], dates emit unquoted and reparse as Date, finite numbers are preserved, and deno test -A toml/ is 689/0 with fmt+lint clean.

There's one regression introduced by the refactor that needs a fix before this can land — RegExp values inside mixed arrays get dropped. Details inline. Once that's handled (plus a small regression test) this is in good shape.

Heads-up: #7171 is an open PR fixing the same issue (#7162) with the same inf/-inf/nan + unquoted-Date approach — worth a look so the two efforts don't collide.

[bartlomieju]

This branch catches RegExp as well as plain objects, which silently drops RegExp values in mixed arrays. Verified against main:

stringify({ x: [/foo/, {}] })
// main:     x = ["/foo/",{}]   ✅
// this PR:  x = [{},{}]        ❌  RegExp lost

Reason: typeof /foo/ === "object" and a RegExp isn't a Date, so a RegExp enters this inline-table branch, Object.keys(/foo/) is [], and it serializes as {}. The old code handled RegExp before the object branch (via the string || RegExp arm), so it never reached here.

Note the asymmetry that hides it: primitive arrays go through #arrayDeclaration -> #printPrimitive, which handles RegExp fine (["/a/","/b/"]). Only the mixed path is broken, and there's no test for RegExp-in-mixed-array, so the suite stays green.

Suggested fix — exclude RegExp the same way Date is:

} else if (
  value && typeof value === "object" &&
  !(value instanceof Date) && !(value instanceof RegExp)
) {

and a regression test, e.g. assertEquals(stringify({ x: [/foo/, {}] }), 'x = ["/foo/",{}]\n').

[spokodev]

Thanks for the careful review and the precise repro. Fixed: excluded RegExp from the inline-table branch the same way Date is, so a RegExp in a mixed array now serializes through the primitive path ({ x: [/foo/, {}] } becomes x = ["/foo/",{}]). Added a regression test. deno test -A toml/ is 690/0, fmt and lint clean.

On #7171: thanks for the heads-up. This PR is broader (it also covers -Infinity, NaN, and the deferred null design), so I will leave the call to you on which to land; I can rebase around the other if you prefer it.