| 1 | //! Protocol-recovery contract tests. |
| 2 | //! |
| 3 | //! These tests exist to keep the engine hostile to fake tool-call wrappers |
| 4 | //! (XML/Replit/markdown pseudo-calls in assistant text). Their job is to make |
| 5 | //! sure that: |
| 6 | //! |
| 7 | //! 1. The known wrapper markers are still present in `core/engine.rs` so the |
| 8 | //! streaming filter has something to scrub. |
| 9 | //! 2. The legacy text-based `tool_parser` flags fake wrappers for |
| 10 | //! stripping/status bookkeeping but does NOT treat the newer |
| 11 | //! `<function_calls>` wrapper as a real tool call — only the legacy |
| 12 | //! `[TOOL_CALL]` and `<invoke>` shapes ever produced structured calls, and |
| 13 | //! nothing should silently re-enable text-based execution. |
| 14 | //! 3. The closing-marker list stays the same length as the start-marker list, |
| 15 | //! so filter logic cannot get stuck in tool-call mode forever. |
| 16 | //! |
| 17 | //! The point is that protocol drift in the model output should be visible (we |
| 18 | //! still strip it and emit a status notice), not silently turned into tool |
| 19 | //! execution. |
| 20 | |
| 21 | use std::fs; |
| 22 | |
| 23 | // `engine.rs` was decomposed into submodules under `core/engine/`. The |
| 24 | // protocol-scrubbing strings the tests below assert on are now spread |
| 25 | // across `engine.rs` and several `engine/*.rs` files. We compile-time |
| 26 | // include each so a contributor moving a marker into a sibling submodule |
| 27 | // does not silently break these regression checks. |
| 28 | const ENGINE_SOURCES: &[&str] = &[ |
| 29 | include_str!("../../src/core/engine.rs"), |
| 30 | include_str!("../../src/core/engine/streaming.rs"), |
| 31 | include_str!("../../src/core/engine/turn_loop.rs"), |
| 32 | include_str!("../../src/core/engine/dispatch.rs"), |
| 33 | include_str!("../../src/core/engine/tool_setup.rs"), |
| 34 | include_str!("../../src/core/engine/tool_execution.rs"), |
| 35 | include_str!("../../src/core/engine/tool_catalog.rs"), |
| 36 | include_str!("../../src/core/engine/context.rs"), |
| 37 | include_str!("../../src/core/engine/approval.rs"), |
| 38 | include_str!("../../src/core/engine/lsp_hooks.rs"), |
| 39 | ]; |
| 40 | |
| 41 | fn any_engine_source_contains(needle: &str) -> bool { |
| 42 | ENGINE_SOURCES.iter().any(|src| src.contains(needle)) |
| 43 | } |
| 44 | |
| 45 | const EXPECTED_START_MARKERS: &[&str] = &[ |
| 46 | "[TOOL_CALL]", |
| 47 | "<codewhale:tool_call", |
| 48 | "<tool_call", |
| 49 | "<invoke ", |
| 50 | "<function_calls>", |
| 51 | ]; |
| 52 | |
| 53 | const EXPECTED_END_MARKERS: &[&str] = &[ |
| 54 | "[/TOOL_CALL]", |
| 55 | "</codewhale:tool_call>", |
| 56 | "</tool_call>", |
| 57 | "</invoke>", |
| 58 | "</function_calls>", |
| 59 | ]; |
| 60 | |
| 61 | #[test] |
| 62 | fn engine_keeps_known_fake_wrapper_start_markers() { |
| 63 | for marker in EXPECTED_START_MARKERS { |
| 64 | let needle = format!("\"{marker}\""); |
| 65 | assert!( |
| 66 | any_engine_source_contains(&needle), |
| 67 | "no engine source file still mentions start marker `{marker}` — \ |
| 68 | protocol scrubbing may have regressed. Searched for {needle:?} \ |
| 69 | across engine.rs and engine/* submodules." |
| 70 | ); |
| 71 | } |
| 72 | } |
| 73 | |
| 74 | #[test] |
| 75 | fn engine_keeps_known_fake_wrapper_end_markers() { |
| 76 | for marker in EXPECTED_END_MARKERS { |
| 77 | let needle = format!("\"{marker}\""); |
| 78 | assert!( |
| 79 | any_engine_source_contains(&needle), |
| 80 | "no engine source file still mentions end marker `{marker}` — \ |
| 81 | protocol scrubbing may have regressed. Searched for {needle:?} \ |
| 82 | across engine.rs and engine/* submodules." |
| 83 | ); |
| 84 | } |
| 85 | } |
| 86 | |
| 87 | #[test] |
| 88 | fn engine_marker_counts_stay_paired() { |
| 89 | // A future contributor could quietly drop a closing marker and leave the |
| 90 | // filter able to enter tool-call mode without ever leaving it. Lock the |
| 91 | // count to whatever the constants currently declare. |
| 92 | assert_eq!(EXPECTED_START_MARKERS.len(), EXPECTED_END_MARKERS.len()); |
| 93 | assert!(any_engine_source_contains("TOOL_CALL_START_MARKERS")); |
| 94 | assert!(any_engine_source_contains("TOOL_CALL_END_MARKERS")); |
| 95 | } |
| 96 | |
| 97 | #[test] |
| 98 | fn engine_emits_compact_fake_wrapper_notice() { |
| 99 | assert!( |
| 100 | any_engine_source_contains("FAKE_WRAPPER_NOTICE"), |
| 101 | "no engine source file references the protocol-recovery notice constant" |
| 102 | ); |
| 103 | assert!( |
| 104 | any_engine_source_contains("API tool channel"), |
| 105 | "the protocol-recovery notice should mention the API tool channel" |
| 106 | ); |
| 107 | } |
| 108 | |
| 109 | #[test] |
| 110 | fn legacy_parser_extracts_bracket_tool_call() { |
| 111 | let result = crate::tool_parser::parse_tool_calls( |
| 112 | "intro [TOOL_CALL]\n{\"tool\":\"x\",\"args\":{}}\n[/TOOL_CALL]", |
| 113 | ); |
| 114 | assert_eq!(result.tool_calls.len(), 1); |
| 115 | assert_eq!(result.tool_calls[0].name, "x"); |
| 116 | assert_eq!(result.clean_text, "intro"); |
| 117 | } |
| 118 | |
| 119 | #[test] |
| 120 | fn legacy_parser_extracts_invoke_block() { |
| 121 | let result = crate::tool_parser::parse_tool_calls( |
| 122 | "before <invoke name=\"do_thing\"><parameter name=\"k\">v</parameter></invoke> after", |
| 123 | ); |
| 124 | assert_eq!(result.tool_calls.len(), 1); |
| 125 | assert_eq!(result.tool_calls[0].name, "do_thing"); |
| 126 | } |
| 127 | |
| 128 | #[test] |
| 129 | fn legacy_parser_does_not_execute_function_calls_wrapper() { |
| 130 | // The newer `<function_calls>` wrapper is the kind of forged shape that |
| 131 | // shows up in non-DeepSeek tool-call leakage. The legacy text parser must |
| 132 | // NOT turn it into a structured tool call (the engine's filter still |
| 133 | // strips it from visible text and the model is expected to use the API |
| 134 | // tool channel instead). |
| 135 | let raw = "narrative <function_calls>\n{\"name\":\"x\",\"input\":{}}\n</function_calls> end"; |
| 136 | let result = crate::tool_parser::parse_tool_calls(raw); |
| 137 | assert!( |
| 138 | result.tool_calls.is_empty(), |
| 139 | "function_calls wrapper must not be parsed as a real tool call: {:?}", |
| 140 | result.tool_calls |
| 141 | ); |
| 142 | } |
| 143 | |
| 144 | #[test] |
| 145 | fn legacy_parser_marker_helper_flags_fake_wrappers_without_enabling_execution() { |
| 146 | // `has_tool_call_markers` now also flags forged wrappers so the engine can |
| 147 | // scrub them from visible text and keep reasoning-placeholder bookkeeping. |
| 148 | // The parser still must not turn those wrappers into executable calls. |
| 149 | assert!(crate::tool_parser::has_tool_call_markers( |
| 150 | "noise [TOOL_CALL]x[/TOOL_CALL]" |
| 151 | )); |
| 152 | assert!(crate::tool_parser::has_tool_call_markers( |
| 153 | "noise <invoke name=\"x\"></invoke>" |
| 154 | )); |
| 155 | assert!(crate::tool_parser::has_tool_call_markers( |
| 156 | "noise <function_calls>{}</function_calls>" |
| 157 | )); |
| 158 | assert!( |
| 159 | crate::tool_parser::parse_tool_calls("noise <function_calls>{}</function_calls>") |
| 160 | .tool_calls |
| 161 | .is_empty() |
| 162 | ); |
| 163 | } |
| 164 | |
| 165 | #[test] |
| 166 | fn engine_source_file_still_exists_and_is_non_trivial() { |
| 167 | // Sanity check so the `include_str!` above is meaningful — if the engine |
| 168 | // module ever moves, this test must be updated alongside it. |
| 169 | let metadata = fs::metadata("src/core/engine.rs").expect("engine.rs must exist next to tests"); |
| 170 | assert!( |
| 171 | metadata.len() > 10_000, |
| 172 | "engine.rs is unexpectedly small ({} bytes); did the file move?", |
| 173 | metadata.len() |
| 174 | ); |
| 175 | } |
| 176 |