fix(translate): reject short ok arrays and negative numeric error codes

[devrishik]

Follow-up on merged #23 (boss review findings 3+4). Findings 1/2/5 deferred as taste.

Finding #3

parse_translate_response accepted a per-string ok array shorter than sent. An all-true short array (e.g. ok:[true] for three texts) passed and the missing flags silently covered unsent fields.

Fix: require flags.len() == sent.len(); mismatch is a malformed envelope (bail!). Same-length all-true still passes.

Finding #4

Numeric negative error codes ("code": -1) were skipped: as_u64() is None on signed JSON numbers and as_str() is None on numbers.

Fix: parse as i64 (as_i64 first, then as_u64→i64, then string). Negative codes are error envelopes. 0 and 200 stay success. Do not fold i64→u64 (try_from would drop negatives again).

Deferred (taste)

• feat(translate): zola translate subcommand + self-hosted release CI (ADR-008 Z-2) #1 Default impl for OpenRouterClient
• feat(graph): zola graph migrate (once) + refresh (local) CLI #2 fields param on parse_translate_response
• docs: translate restored after deploy #5 URL scheme check on TRANSLATE_URL

Tests

• ok_array_shorter_than_sent_is_malformed
• numeric_negative_error_code_is_a_hard_failure

cargo test --bin zola -- ok_array_shorter_than_sent numeric_negative_error_code ok_false ok_array_with_a_false mirrored_error_code_1003 ok_all_true → 7 passed.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Team

Run ID: 93fe2334-4290-44b7-a865-608826f80d25

---

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.}

[devrishik]

Automated GLM review — human review still required.

Scope: src/cmd/translate.rs changes only (loops/ docs excluded). Both fixes verified correct; tests re-run locally (cargo test --bin zola -- … → 7 passed, 0 failed). No approval, no merge, no code changes from this review.

1. should-fix — src/cmd/translate.rs:265 — u64-range positive error codes are now silently dropped
v.as_u64().and_then(|n| i64::try_from(n).ok()) is dead for its stated purpose: any u64 that fits i64 was already consumed by as_i64() at :264, so the only values reaching this arm are > i64::MAX, where try_from fails → None → code silently treated as success. Net regression vs old code: {"code": 18446744073709551615} used to bail! via as_u64; now it passes. Same gap via the string arm at :266: "code": "18446744073709551615" parses as neither i64 (overflow) nor the old u64 inference → dropped. The comment ("u64 → i64 is only for values that did not fit as_i64") claims the opposite of what executes. Practically absurd codes, but this PR's whole point is "error codes must not be dropped." Suggestion: drop the middle arm; instead, when v.is_number() (or v.as_str() numeric) but all parse attempts yield None, bail!("translate endpoint: unparsable code in {key}"). That closes both gaps and deletes the dead arm.

2. nit — src/cmd/translate.rs:238 — longer ok arrays now also rejected; endpoint contract unverified
The length guard rejects flags.len() > sent.len() too. Symmetric with the pre-existing translations length check, so consistent — but body-only chunk requests send sent.len()==1, and if the nllb endpoint ever echoes a fixed 3-flag ok array regardless of texts count, every long-page body chunk hard-fails. Test at :912 only covers the shorter case. Suggestion: add ok_array_longer_than_sent_is_malformed, and confirm the endpoint mirrors texts length in ok (one curl against the live endpoint settles it).

3. nit — src/cmd/translate.rs:954 — string negative code path untested
:266 now also makes "code": "-1" a hard failure (string parse target changed u64→i64). That's new behavior with no test. Suggestion: mirror the numeric test with {"code":"-1","translations":["T"]}.

4. nit — src/cmd/translate.rs:247 — unwrap_or("?") fallback now unreachable
Harmless belt-and-suspenders after the :238 length guard; could drop .copied().unwrap_or("?") for a direct index. Leave if you prefer the guard-independent form.

Checked and clean: as_bool().unwrap() at the position closure is short-circuit-safe (unwrap only reached on Value::Boolean); per-chunk envelope validation stays consistent with 6000-char body chunking (each chunk gets its own sent/ok/translations triple); empty fields never sent; parse_translate_response signature untouched (deferred findings 1/2/5 respected); error messages carry both the code value and the offending key.

Findings: 0 blockers, 1 should-fix, 3 nits.