[yuefdev]

Running add on []any{map[string]any(nil), map[string]any{"x": 1}} panics: cloning the first map leaves a nil destination for the next maps.Copy.

Keep maps.Clone for the first map and initialise an empty map only when the clone is nil. Clone the first slice with append([]any{}, x...) so a typed nil slice produces an empty array rather than null with json.Marshal.

Regressions cover both nil maps and nil slices through the public query API, using add and add(.[]). The new slice cases fail before the follow-up and pass afterwards.

Validation: go test ./..., go vet ./..., changed-file gofmt and git diff --check pass. The race suite could not run for this follow-up because cgo is disabled in the current local environment.

[itchyny]

This is slower than maps.Clone because this rehashes and inserts. I would suggest checking for nil and assign map[string]any{} after maps.Clone.

[yuefdev]

I've restored maps.Clone(x) and only allocate map[string]any{} when the clone is nil (31af8e3). That keeps the fast cloning path for populated maps while retaining the nil-map fix. The existing regression passes for both add and add(.[]).

[itchyny]

We'll also need to avoid nil slice because I want to guarantee that the emit does not emit nil slice (which may cause confusing result with json.Marshal). Replace slices.Clone(x) with append([]any{}, x...).

[yuefdev]

I've also switched the slice case to append([]any{}, x...) in 31af8e3, so a typed nil slice produces an empty, non-nil array. Added a public-API regression for both add and add(.[]); both cases fail before the change and pass afterwards.

go test ./..., go vet ./... and the changed-file formatting checks pass. I couldn't rerun the race suite in this environment because cgo is disabled. I've updated the PR description to reflect that.

[itchyny]

Thank you.