fix: decode 8-bit string fields as latin-1 instead of ascii

[johansolve]

The problem

Node's 'ascii' encoding masks off bit 7 rather than validating its input, so any byte at 0x80 or above silently decodes to a different character.

A B&G Vulcan 7 sends PGN 129285 route names as STRING_LAU with control byte 1 and latin-1 characters. The Swedish route name Hättan-Askim arrives on the bus as:

48 e4 74 74 61 6e 2d 41 73 6b 69 6d 00
H  ä  t  t  a  n  -  A  s  k  i  m

and canboatjs decodes it as Hdttan-Askim — 0xe4 becomes 0x64. The plotter itself displays the name correctly, so the corruption happens on our side. Downstream this reaches Signal K as navigation.courseRhumbline.activeRoute.name, which then no longer matches the same route stored in the resources API, where the name is intact because it took an HTTP path instead of an N2K one.

Buffer.from([0x48, 0xe4, 0x74, 0x74, 0x61, 0x6e]).toString('ascii')   // 'Hdttan'
Buffer.from([0x48, 0xe4, 0x74, 0x74, 0x61, 0x6e]).toString('latin1')  // 'Hättan'

The change

'ascii' → 'latin1' in the three readers that decode 8-bit string fields: STRING_LAU, STRING_FIX, and String with start/stop byte.

This is backwards compatible where it counts. For genuine 7-bit ASCII the two encodings produce byte-identical output, so no currently correct decode can break. Above 0x7f, latin-1 at least preserves the byte value — a consumer that knows better can reinterpret with Buffer.from(s, 'latin1') — whereas 'ascii' destroys it irrecoverably.

It follows the same reasoning as 82d12b2, which moved STRING_LZ off 'ascii' for this class of problem.

toPgn already writes these fields with charCodeAt, i.e. latin-1, so this removes a round-trip asymmetry for high bytes rather than introducing one.

Visible behaviour change: padding or junk bytes at 0x80 and above in STRING_FIX now render as U+0080–U+00FF instead of their bit-7-stripped equivalents. Both are garbage; the output differs.

I checked test/pgns/ for existing expectations containing non-ASCII characters before and after — there are none besides the one added here.

Test case

A real 129285 captured off the boat, including the full fast-packet reassembly, so the regression is pinned to a packet a shipping plotter actually sends.

It carries skipEncoderTest, and the reason is worth stating because it is not what it first looks like: the Vulcan NUL-terminates its STRING_LAU fields and counts the terminator in the length byte, while the encoder omits it. Re-encoding therefore yields three bytes fewer (0x0f vs 0x0e for the route name, and the same for each waypoint name). The reader handles both forms; changing the encoder would move the bytes under every other STRING_LAU encode test, so that is deliberately left alone. The 0x7fffffff lat/lon in the first waypoint, which is what I first assumed was the cause, round-trips exactly.

Known follow-up, deliberately not in this PR

STRING_LZ is now the one remaining inconsistency: it is read as 'utf-8' but written with charCodeAt, so 'ä' encodes to 0xe4 and reads back as U+FFFD. That affects 21 fields across the PGN definitions. Since the UTF-8 read was a deliberate choice in #296, it seemed wrong to reverse it as a side effect of this fix — but the read and write sides do disagree, and it may be worth a look.

CI

The suite is red on master before this change: four failures around Simnet: Command AP NoDrift and Simnet: Command AP Change Course, which reproduce on an untouched c584d70 with the @canboat/ts-pgns version that ^1.11.9 currently resolves to. This branch goes from 188 passing / 4 failing to 189 passing / 4 failing — the same four. prettier --check lib test and eslint are both clean.

The new test was mutation-checked with clean builds: reverting the STRING_LAU line to 'ascii' makes it fail, restoring 'latin1' makes it pass. (Worth noting for anyone verifying locally: tsc -b is incremental and will not rebuild the file on a revert, which makes the fix look like it has no effect.)

Co-Authored by Claude Code Opus 5.

Summary by CodeRabbit

• Bug Fixes

  • Improved decoding of Latin-1 text in navigation data, including route and waypoint names.
  • Preserved correct handling of null-terminated and fixed-length strings.

• Tests

  • Added regression coverage for packets containing accented characters and navigation metadata.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: c0c41984-bea8-475f-a1e9-ba539891f15c

📥 Commits

Reviewing files that changed from the base of the PR and between c584d70 and 3cd7961.

📒 Files selected for processing (2)

• lib/fromPgn.ts
• test/pgns/129285.js

---

📝 Walkthrough

Walkthrough

PGN string decoding now uses Latin-1 for four string formats. A PGN 129285 fixture verifies decoding of Latin-1 route and waypoint names, including Hättan-Askim.

Changes

Latin-1 decoding

Layer / File(s)	Summary
Update PGN string decoding lib/fromPgn.ts	STRING_LAU, start/stop-byte, length-prefixed, and fixed-length strings now decode with Latin-1.
Add PGN 129285 regression coverage test/pgns/129285.js	The captured B&G Vulcan 7 message verifies Latin-1 names, NUL-terminated strings, navigation values, nullable fields, and waypoint coordinates.

Estimated code review effort: 2 (Simple) | ~10 minutes

Suggested labels: fix, test

Suggested reviewers: sbender9, dirkwa

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely describes the main change from ASCII to Latin-1 decoding for 8-bit string fields.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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.}

[johansolve]

Please label as Fix