[lejmr]

HOME : ${HOME}, with HOME set, was Unresolved. The spec says otherwise:

Here the ${foo} self-reference comes before foo has a value, so it is undefined, exactly as if the substitution referenced a path not found in the document. (examples-of-self-referential-substitutions.6)

A path not found in the document is looked up in the environment (substitution-fallback-to-environment-variables.1, substitutions.13). For ${?foo} the spec says the same thing: "as if it referred to a nonexistent value" (self-referential-substitutions.13). The blocking rule (substitution-fallback-to-environment-variables.4) is about a value the lookup finds, such as HOME : null. A self-reference with nothing below it has no such value. The spec's author said in lightbend/config#652 that it should work, and pyhocon has done it since chimpler/pyhocon#213. typesafe/config reports a cycle, which .7 rules out ("undefined" rather than "intractable cycle").

Behaviour

input, HOME set	before	after	typesafe/config	pyhocon
HOME : ${HOME}	Unresolved	"/home/…"	cycle error	"/home/…"
HOME : ${?HOME}	{}	"/home/…"	{}	"/home/…"
HOME : ${HOME}"/bin"	Unresolved	"/home/…/bin"	cycle error	"/home/…/bin"
HOME += x	[x]	UnsupportedConcatenation	[x]	"/home/… x"

+= is HOME = ${?HOME} [x], so the variable's text meets a list and the result is an error. A value set earlier, null included, still wins, and with --env=none nothing changes.

Code

• Found.gone is left for an optional that hit a cycle (env.an optional cycle never reads the variable is unchanged). A self-reference with nothing below is .absent, so it goes on to the variable. With --env=keep it waits for one.
• A kept self-reference from an include now always falls back to the path as written, so keep.inDocument goes. stringify's row 10 used to be a known loss and now round-trips.
• The tests that pinned the old rule in env.zig, keep.zig, Include.zig and stringify.zig are rewritten, with the reason in their comments.

Conformance

• substitution-fallback-to-environment/009 (${X}, java: unsupported) and 010 (${?X}, java: diverges).
• Measured again on the 478, after a restack onto fix(parse): a value that only starts like a number is text #30: zig-hocon (spec 478, java 419), typesafe/config 1.4.9 (spec 419, java 478), pyhocon (spec 337, java 308). hocon2, pulse-pyhocon, py.hocon and rs.hocon aren't installed here and stay marked as behind.
• FINDINGS.md, a CHANGELOG line, and "all 478" in the READMEs and MIGRATING.md, pyhocon 337.

Checks: zig build test passes, zig build test-python is at 100 %, zig fmt --check src is clean, and fill-expected.py --check reports 478 cases with 0 sidecars stale.

The case numbers may need to move when #28/#29/#30 land first. None of them touches this section.