RemoteFortressReader: GetCreatureRaws sends CP437 (not UTF-8) strings, which breaks protobuf decoding #6019

Disclaimer: this issue was drafted by GitHub Copilot (an AI assistant) under the direction of user @dwmuller. The test results below come from real runs against a live game, but the source-code analysis is AI-generated, so please verify it.

Summary

RemoteFortressReader.GetCreatureRaws returns some strings in the game's CP437 encoding instead of UTF-8. Protobuf string fields must be UTF-8, so strict clients (Rust prost, and likely others) reject the whole reply. A single bad string makes the 23 MB reply undecodable.

The 50.08-r2 changelog says "RemoteFortressReader: ensured names are transmitted in UTF-8 instead of CP437", but DF2UTF is applied selectively and was not applied here.

Environment

• Dwarf Fortress 53.16
• DFHack 53.16-r3
• Remote client: Rust dfhack-remote 0.12.0 (prost)
• Fortress with many generated creatures (raw indices ~2026-2085 are affected)

Observed

Decoding fails with:

failed to decode Protobuf message: CasteRaw.description: CreatureRaw.caste:
CreatureRawList.creature_raws: invalid string value: data is not UTF-8 encoded.

I walked the raw reply bytes without UTF-8 validation and found 74 non-UTF-8 strings. They fall into three fields:

0x86 is "å" in CP437, which fits these generated names ("Råsh").

Paging GetPartialCreatureRaws and bisecting shows that exactly 14 raws in the 2026-2085 range fail. Indices before 2026 decode fine.

Cause (from reading the source)

In plugins/remotefortressreader/remotefortressreader.cpp, the caste loop has:

send_caste->set_description(orig_caste->description);

There is no DF2UTF(...). The CreatureRaw/CasteRaw name fields (name, caste_name, baby/child names) show the same symptom in the data and should be checked as well. I only confirmed the unconverted set_description call in the source.

Not affected in this fortress

• GetUnitList and GetUnitListInside (305 units) have no invalid UTF-8.
• GetMaterialList, GetGrowthList, GetItemList and GetPlantRaws have no invalid UTF-8.

I looked at these specifically because they also carry caste strings. They're fine here, but it's worth auditing every string that reaches a protobuf message.

Suggested fix

Wrap all game-sourced strings in the creature-raw path with DF2UTF, and audit the other set_* string calls in the plugin for the same omission.

Reproduction

Any client works. Call GetCreatureRaws and validate the reply strictly as UTF-8. I wrote a small test harness that does this. It's in a project of mine, and everything is read-only:

https://github.com/dwmuller/DFCompanion/blob/experiment/dfhack-lua-http-spike/tests/live_game_tests.rs

Run it against a live game (DFHack remote on port 5000):

cargo test --test live_game_tests -- --ignored --nocapture --test-threads=1

• creature_raws and creature_raws_paged use the typed client, and the paged one bisects to the failing indices.
• raw_creature_raws_utf8 bypasses prost, reads the protobuf wire format directly, and lists every non-UTF-8 field by path with example bytes.
• The other raw_* tests do the same for the unit and material/item lists.