fix(ip): make an empty ip -4/-6 fail like Go instead of exiting 0

[GeiserX]

tnet ip -6 on a node whose tailnet holds no IPv6 printed
(no matching tailnet address) on stdout and exited 0. A script that
tests the exit status — if tnet ip -6 "$host"; then ... — read that as
"the host has an IPv6 address". It does not.

Upstream is cmd/tailscale/cli/ip.go (runIP) @
53a0d659afa51835dd7a9283873cca44261454f8. Its match loop sets match
only when it printed an address, and the function ends:

if !match {
    if ipArgs.want4 { return errors.New("no Tailscale IPv4 address") }
    if ipArgs.want6 { return errors.New("no Tailscale IPv6 address") }
}
return nil

Those are errors.New returns, not outln calls. That is the whole
reason upstream wrote them as errors: the CLI turns a returned error into
a stderr line and a non-zero exit, so the command is usable as a
predicate. This port had re-derived the branch as a printed placeholder.

WHAT CHANGED

format_ip_filtered and format_service_ips now return
Result<String, &'static str>. The empty-selection branch consults the
new ip_no_match_error, which is Go's !match tail verbatim: -4 gives
no Tailscale IPv4 address, -6 gives no Tailscale IPv6 address, and
with neither flag it gives None. run_ip prints an Ok and
anyhow::bail!s an Err, which reaches the shell as stderr plus exit 1.

Three shape decisions worth naming:

• The branch is on the RAW -4/-6 flags, not on a derived "both
  families wanted" pair. Go branches on ipArgs.want4/want6, not on
  the local v4 that defaults to true, so an unflagged empty answer
  correctly stays a non-error and keeps the placeholder.
• format_service_ips moves too. Both formatters feed the same let out
  expression in run_ip, so they share a return type; and Go really does
  run one match loop over one ips slice however the target resolved, so
  the Service arm owes the same error. Not scope creep — forced.
• The unflagged empty answer keeps the placeholder rather than borrowing
  Go's other message. Go covers that state EARLIER, with
  no current Tailscale IPs; state: %v, built from BackendState — see
  QUESTIONS.

HOW IT WAS VERIFIED

New tests/tnet_ip_family_filter_exit_status.rs runs the built tnet
against a stub daemon on a Unix socket, the way the lock/cert/kubeconfig
tests do, and pins the surface a script actually sees: exit status,
stdout, stderr. Four cases — -6 against an IPv4-only reply, -4
against an IPv6-only reply, a positive control (-4 against a node that
has one, so exit 0 with the address on stdout), and the unflagged empty
answer (still exit 0). The control is there so a broken socket path
cannot make the two negative cases pass vacuously.

Checked against the pre-fix binary: the two negative cases fail with
stdout: (no matching tailnet address) and a zero exit, while the
control and the unflagged case pass either way. The unit tests beside
format_ip_filtered pin which state is an error; this file pins the
wiring, so a future edit that turned the bail! back into an
eprintln! + Ok(()), or that printed the message on stdout, fails
here.

Gate, all four green:
cargo fmt --all --check; cargo clippy --all-targets -- -D warnings;
cargo test --all-targets (437 + 302 unit tests plus every integration
suite, 0 failed); cargo build --release --bins.

WHAT TO READ FIRST

ip_no_match_error in src/bin/tnet.rs — it is the ported tail, and its
else { None } arm is the part an over-eager fix gets wrong. Then the
match out at the end of run_ip, then the new test file.

STALE ELSEWHERE

docs/PARITY_GAP_ANALYSIS.md is not touched here, but its tailscale ip
row no longer holds: the -4/-6 family filter now matches upstream's
error return and exit status, so only the address-less unflagged case
(below) still diverges.

DECISIONS

• The error text is a &'static str carried through the formatters,
  rather than an error type. anyhow::bail! on a &'static str is
  already the idiom in this same function (ip_usage_refusal does it),
  and Go's strings are fixed, so nothing needs formatting.
• The -1 evaluation-order comments in both formatters are unchanged.
  Upstream's -1 reads as a break inside the match branch rather than
  a slice truncation ahead of it; the two are indistinguishable in
  behaviour here and ip_usage_refusal makes the difference unreachable,
  so rewording them belongs in its own change.
• ip_refusal_covers_the_service_arm_that_would_answer_emptily is
  renamed to ..._that_would_select_nothing, and its rationale rewritten.
  The old name and the sentence under it said the command "would answer
  that empty set" — after this change it would not; it would report the
  family error. The refusal is still worth testing, only its
  justification had to move.

QUESTIONS

Q: Should no current Tailscale IPs; state: %v be ported too, by putting
the backend state on the ip response? It is the last piece of runIP's
tail that still exits 0 where Go exits 1.

A: Yes, but as its own change — it needs a new field on Response::Ip in
src/localapi.rs plus the daemon side, which is a wire-DTO change, and
this one is a CLI-side fix. Two notes for whoever takes it. First, the
remaining gap is narrower than it sounds: with -4 or -6 set, a node
holding no address at all now exits non-zero here, with the family text
rather than the state text. What still exits 0 is only the unflagged (or
-1-only) call against an address-less node. Second, Go checks
len(ips) == 0 BEFORE the match loop, so no current Tailscale IPs must
win over no Tailscale IPv6 address when both would apply.

---

_{bead tsd-aud319 · session tsd-aud319-c581-r2-s1n0 · base refs/heads/main}

[coderabbitai]

Warning

Review limit reached

Next included review available in 29 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available.

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Team

Run ID: 5f65cf05-65c9-457f-be2b-feba42575a9b

📥 Commits

Reviewing files that changed from the base of the PR and between ec4a7cb and 06befda.

📒 Files selected for processing (2)

• src/bin/tnet.rs
• tests/tnet_ip_family_filter_exit_status.rs

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[GeiserX]

Superseded by a sibling pull request from the same task, which has already merged — this one would land the same change twice, so it can never be merged. Closing it to free the factory's open-pull-request budget. The branch is kept: reopen this if you want this variant instead.