| 1 | //! Transcript history-cell tests. |
| 2 | //! |
| 3 | //! Rebuilt in v0.9.11 after declaring test bankruptcy on the previous suite |
| 4 | //! (123 tests / 3,964 lines). About a third of that file pinned glyph choices, |
| 5 | //! palette tokens, span indices and English label text — `spans[1] == "⣤"`, |
| 6 | //! `title_span.style.fg == theme.tool_title_color`, `visible[1] == "▏ done: |
| 7 | //! scan repo"`. Those assertions fail on every legitimate visual refactor and |
| 8 | //! catch nothing a user would notice, which is the liability `d64b9429b` |
| 9 | //! ("remove brittle visual test mass") named. |
| 10 | //! |
| 11 | //! What survives is named for the *property* it protects. Rules for additions: |
| 12 | //! |
| 13 | //! * Assert a property, not a token. `spans[1] == "⣤"` is a token; "the frame |
| 14 | //! does not change while motion is reduced" is the property, and it is |
| 15 | //! strictly stronger — it also catches an animation leak the constant missed. |
| 16 | //! * Where the value is a design choice (color, glyph, verb), assert the |
| 17 | //! *relationship* between cases instead: warning must not read as error. |
| 18 | //! * One test per property, with its cases in a table — not one test per case. |
| 19 | //! * Never assert `a || b` where `b` is trivially true of any English string. |
| 20 | |
| 21 | use super::constants::{TOOL_OUTPUT_HEAD_LINES, TOOL_OUTPUT_LINE_LIMIT, TOOL_OUTPUT_TAIL_LINES}; |
| 22 | use super::thinking::cached_color_depth; |
| 23 | use super::{ |
| 24 | ASSISTANT_GLYPH, ExecCell, ExecSource, GenericToolCell, HistoryCell, PlanUpdateCell, |
| 25 | REASONING_CURSOR, REASONING_OPENER, REASONING_RAIL, RenderMode, ThinkingFold, ToolCell, |
| 26 | ToolStatus, TranscriptRenderOptions, WebSearchCell, assistant_label_style_for, |
| 27 | extract_reasoning_summary, render_spillover_annotation, render_thinking, |
| 28 | render_thinking_with_analysis, running_status_label_with_elapsed, |
| 29 | }; |
| 30 | use crate::tools::plan::{PlanSnapshot, StepStatus}; |
| 31 | use crate::tui::motion::MotionMode; |
| 32 | use crate::tui::ui_text::{ |
| 33 | line_to_plain, slice_visible_columns, text_display_width, text_visible_width, |
| 34 | }; |
| 35 | use codewhale_models::{ContentBlock, Message, Role}; |
| 36 | use std::path::PathBuf; |
| 37 | use std::time::{Duration, Instant}; |
| 38 | |
| 39 | #[test] |
| 40 | fn extension_prompt_snapshots_and_withdrawal_show_the_complete_model_input() { |
| 41 | let instructions = format!("{}\nLast instruction.", "x".repeat(12_000)); |
| 42 | for block in [Some(instructions.as_str()), None] { |
| 43 | let message = crate::runtime_handoff::extension_prompt_contributions_runtime_message(block); |
| 44 | let cells = super::history_cells_from_message(&message); |
| 45 | let [HistoryCell::System { content }] = cells.as_slice() else { |
| 46 | panic!("runtime instructions must be shown as a system receipt"); |
| 47 | }; |
| 48 | match block { |
| 49 | Some(text) => assert!( |
| 50 | content.contains(text), |
| 51 | "the full model-visible text stays auditable" |
| 52 | ), |
| 53 | None => assert!(content.contains("withdrawn")), |
| 54 | } |
| 55 | } |
| 56 | } |
| 57 | |
| 58 | // --------------------------------------------------------------------------- |
| 59 | // Helpers |
| 60 | // --------------------------------------------------------------------------- |
| 61 | |
| 62 | fn line_text(line: &ratatui::text::Line<'static>) -> String { |
| 63 | line.spans |
| 64 | .iter() |
| 65 | .map(|span| span.content.as_ref()) |
| 66 | .collect() |
| 67 | } |
| 68 | |
| 69 | fn lines_text(lines: &[ratatui::text::Line<'static>]) -> String { |
| 70 | lines.iter().map(line_text).collect::<Vec<_>>().join("\n") |
| 71 | } |
| 72 | |
| 73 | fn generic_tool(name: &str, status: ToolStatus) -> GenericToolCell { |
| 74 | GenericToolCell { |
| 75 | name: name.to_string(), |
| 76 | status, |
| 77 | input_summary: None, |
| 78 | output: None, |
| 79 | prompts: None, |
| 80 | spillover_path: None, |
| 81 | output_summary: None, |
| 82 | is_diff: false, |
| 83 | } |
| 84 | } |
| 85 | |
| 86 | fn exec_tool(command: &str, status: ToolStatus) -> ExecCell { |
| 87 | ExecCell { |
| 88 | command: command.to_string(), |
| 89 | status, |
| 90 | output: None, |
| 91 | live_output: None, |
| 92 | shell_task_id: None, |
| 93 | owner_agent_id: None, |
| 94 | owner_agent_name: None, |
| 95 | started_at: None, |
| 96 | duration_ms: None, |
| 97 | stale_elapsed_since_output_ms: None, |
| 98 | source: ExecSource::Assistant, |
| 99 | interaction: None, |
| 100 | output_summary: None, |
| 101 | } |
| 102 | } |
| 103 | |
| 104 | fn numbered_output(count: usize) -> String { |
| 105 | (0..count) |
| 106 | .map(|i| format!("row {i:02} plain content")) |
| 107 | .collect::<Vec<_>>() |
| 108 | .join("\n") |
| 109 | } |
| 110 | |
| 111 | fn calm_options() -> TranscriptRenderOptions { |
| 112 | TranscriptRenderOptions { |
| 113 | low_motion: true, |
| 114 | ..TranscriptRenderOptions::default() |
| 115 | } |
| 116 | } |
| 117 | |
| 118 | // --------------------------------------------------------------------------- |
| 119 | // Leaks — a rendered cell never exposes something the user was not shown |
| 120 | // --------------------------------------------------------------------------- |
| 121 | |
| 122 | /// Spilled tool output lives in a file under the session directory. The path is |
| 123 | /// an internal storage detail: it names the user's home, their session id, and |
| 124 | /// a content hash, and it is useless to them because the affordance opens the |
| 125 | /// pager, not the file. No width, no render mode, and no standalone annotation |
| 126 | /// may print it. |
| 127 | /// |
| 128 | /// Replaces five separate tests that each checked one width or one mode. |
| 129 | #[test] |
| 130 | fn no_width_or_render_mode_leaks_a_spillover_storage_path() { |
| 131 | let secret = "/Users/private/.codewhale/sessions/session-a/artifacts/hash.txt"; |
| 132 | |
| 133 | for width in [18_u16, 40, 80, 120] { |
| 134 | for mode in [RenderMode::Live, RenderMode::Transcript] { |
| 135 | let mut cell = generic_tool("read_file", ToolStatus::Success); |
| 136 | cell.input_summary = Some("cmd: cargo build --release".to_string()); |
| 137 | cell.output = Some(numbered_output(40)); |
| 138 | cell.spillover_path = Some(PathBuf::from(secret)); |
| 139 | |
| 140 | let rendered = lines_text(&cell.lines_with_mode(width, true, mode)); |
| 141 | for fragment in ["/Users", ".codewhale", "sessions/", "hash.txt"] { |
| 142 | assert!( |
| 143 | !rendered.contains(fragment), |
| 144 | "storage path fragment {fragment:?} leaked at width {width} in {mode:?}: \ |
| 145 | {rendered:?}" |
| 146 | ); |
| 147 | } |
| 148 | } |
| 149 | |
| 150 | // The standalone affordance carries no path either, and it fits the |
| 151 | // width it was given — an affordance that overflows is a wrap artifact |
| 152 | // in the transcript. |
| 153 | let annotation = line_to_plain(&render_spillover_annotation(width)); |
| 154 | assert!( |
| 155 | text_display_width(&annotation) <= usize::from(width), |
| 156 | "affordance exceeds width {width}: {annotation:?}" |
| 157 | ); |
| 158 | for fragment in ["/Users", ".codewhale", "hash.txt"] { |
| 159 | assert!( |
| 160 | !annotation.contains(fragment), |
| 161 | "affordance leaked {fragment:?}: {annotation:?}" |
| 162 | ); |
| 163 | } |
| 164 | } |
| 165 | |
| 166 | // The common case — a result that never spilled — spends no row on an |
| 167 | // affordance that would open an empty pager. |
| 168 | let mut plain = generic_tool("read_file", ToolStatus::Success); |
| 169 | plain.output = Some("contents".to_string()); |
| 170 | let hint = crate::tui::key_shortcuts::tool_details_shortcut_action_hint("output"); |
| 171 | let rendered = lines_text(&plain.lines_with_mode(80, true, RenderMode::Live)); |
| 172 | assert!( |
| 173 | !rendered.contains(&hint), |
| 174 | "a result that did not spill must not advertise the spillover pager: {rendered:?}" |
| 175 | ); |
| 176 | } |
| 177 | |
| 178 | /// With reasoning display off, the model's chain of thought must not reach the |
| 179 | /// screen in any lifecycle state — not while streaming, not once complete. |
| 180 | /// The live case still needs a progress signal, so it renders one compact row. |
| 181 | #[test] |
| 182 | fn hidden_reasoning_never_renders_its_content_in_any_state() { |
| 183 | let secret = "private chain of thought that must not be shown"; |
| 184 | let hidden = TranscriptRenderOptions { |
| 185 | show_thinking: false, |
| 186 | low_motion: true, |
| 187 | ..TranscriptRenderOptions::default() |
| 188 | }; |
| 189 | |
| 190 | let streaming = HistoryCell::Thinking { |
| 191 | content: secret.to_string(), |
| 192 | streaming: true, |
| 193 | duration_secs: None, |
| 194 | }; |
| 195 | let live = streaming.lines_with_options(80, hidden); |
| 196 | let live_text = lines_text(&live); |
| 197 | assert!( |
| 198 | !live_text.contains(secret), |
| 199 | "hidden live reasoning revealed its body: {live_text}" |
| 200 | ); |
| 201 | assert_eq!( |
| 202 | live.len(), |
| 203 | 1, |
| 204 | "hidden reasoning is one compact progress row, not a stack of state \ |
| 205 | copy: {live_text}" |
| 206 | ); |
| 207 | |
| 208 | let complete = HistoryCell::Thinking { |
| 209 | content: secret.to_string(), |
| 210 | streaming: false, |
| 211 | duration_secs: Some(1.0), |
| 212 | }; |
| 213 | assert!( |
| 214 | complete.lines_with_options(80, hidden).is_empty(), |
| 215 | "completed hidden reasoning must leave the live transcript entirely" |
| 216 | ); |
| 217 | } |
| 218 | |
| 219 | /// A live card is a summary. It must name the tool that ran (so the row is |
| 220 | /// attributable) and must not spend its one line echoing arguments the caller |
| 221 | /// never chose — `max_count: 15` is a schema default, not a user intent. |
| 222 | /// Transcript replay is the record, so it keeps the exact tool id. |
| 223 | /// |
| 224 | /// Replaces six tests, one of which (`unknown_generic_tool_keeps_raw_name_in_ |
| 225 | /// live_mode`) asserted only `!text.is_empty()` and so could never fail. |
| 226 | #[test] |
| 227 | fn live_cards_name_their_tool_without_echoing_control_only_arguments() { |
| 228 | assert_eq!( |
| 229 | super::summarize_tool_args(&serde_json::json!({ |
| 230 | "max_count": 15, |
| 231 | "timeout_ms": 30_000 |
| 232 | })), |
| 233 | None, |
| 234 | "an argument set that is entirely control defaults summarizes to nothing" |
| 235 | ); |
| 236 | assert_eq!( |
| 237 | super::summarize_tool_args(&serde_json::json!({ |
| 238 | "max_count": 15, |
| 239 | "branch": "main" |
| 240 | })) |
| 241 | .as_deref(), |
| 242 | Some("branch: main"), |
| 243 | "the meaningful key is what the summary is for" |
| 244 | ); |
| 245 | |
| 246 | for name in ["git_log", "future_private_tool"] { |
| 247 | let mut cell = generic_tool(name, ToolStatus::Success); |
| 248 | cell.input_summary = Some("max_count: 15".to_string()); |
| 249 | let lines = cell.lines_with_mode(120, true, RenderMode::Live); |
| 250 | let joined = lines_text(&lines); |
| 251 | |
| 252 | assert_eq!(lines.len(), 1, "compact live row for {name}: {joined:?}"); |
| 253 | assert!( |
| 254 | joined.contains(name), |
| 255 | "the row must be attributable to {name}: {joined:?}" |
| 256 | ); |
| 257 | assert!( |
| 258 | !joined.contains("max_count"), |
| 259 | "control defaults must not become the visible summary for {name}: {joined:?}" |
| 260 | ); |
| 261 | } |
| 262 | |
| 263 | // A tool the UI has a family for is identified by that family live; the |
| 264 | // raw id would be a second, redundant name. Replay keeps it. |
| 265 | let mut known = generic_tool("run_verifiers", ToolStatus::Running); |
| 266 | known.input_summary = Some("profile: auto, level: quick".to_string()); |
| 267 | let known = HistoryCell::Tool(ToolCell::Generic(known)); |
| 268 | let live = lines_text(&known.lines(80)); |
| 269 | let transcript = lines_text(&known.transcript_lines(80)); |
| 270 | assert!( |
| 271 | !live.contains("run_verifiers"), |
| 272 | "a known tool id must not take a slot in the compact live card: {live}" |
| 273 | ); |
| 274 | assert!( |
| 275 | transcript.contains("run_verifiers"), |
| 276 | "transcript replay preserves the exact tool id: {transcript}" |
| 277 | ); |
| 278 | } |
| 279 | |
| 280 | // --------------------------------------------------------------------------- |
| 281 | // Budget — the card never lies about how much it is showing |
| 282 | // --------------------------------------------------------------------------- |
| 283 | |
| 284 | /// `selected_output_indices` fills head + tail and then tops up from lines that |
| 285 | /// look important (error / warning / path). Plain output — a list of names, a |
| 286 | /// clean build log — matches none of those, so the top-up found nothing and the |
| 287 | /// card silently forfeited the rest of its budget while still reporting the |
| 288 | /// remainder as omitted. A card that advertises N rows shows N rows. |
| 289 | #[test] |
| 290 | fn a_live_card_spends_the_whole_output_budget_it_advertises() { |
| 291 | let total = 40usize; |
| 292 | let cell = { |
| 293 | let mut exec = exec_tool("list_things", ToolStatus::Failed); |
| 294 | exec.output = Some(numbered_output(total)); |
| 295 | exec.duration_ms = Some(120); |
| 296 | HistoryCell::Tool(ToolCell::Exec(exec)) |
| 297 | }; |
| 298 | |
| 299 | let live_text = lines_text(&cell.lines_with_options(80, calm_options())); |
| 300 | let shown = (0..total) |
| 301 | .filter(|i| live_text.contains(&format!("row {i:02} plain content"))) |
| 302 | .count(); |
| 303 | |
| 304 | assert_eq!( |
| 305 | shown, TOOL_OUTPUT_LINE_LIMIT, |
| 306 | "a card promising {TOOL_OUTPUT_LINE_LIMIT} rows must show \ |
| 307 | {TOOL_OUTPUT_LINE_LIMIT}, not stop at head+tail: {live_text}" |
| 308 | ); |
| 309 | for i in 0..TOOL_OUTPUT_HEAD_LINES { |
| 310 | assert!( |
| 311 | live_text.contains(&format!("row {i:02} plain content")), |
| 312 | "head row {i} missing: {live_text}" |
| 313 | ); |
| 314 | } |
| 315 | for i in (total - TOOL_OUTPUT_TAIL_LINES)..total { |
| 316 | assert!( |
| 317 | live_text.contains(&format!("row {i:02} plain content")), |
| 318 | "tail row {i} missing: {live_text}" |
| 319 | ); |
| 320 | } |
| 321 | } |
| 322 | |
| 323 | /// Failure keeps the invocation plus head/tail evidence visible under every |
| 324 | /// density setting; the full record remains in the details view. |
| 325 | #[test] |
| 326 | fn calm1_failed_tool_keeps_context_and_result_under_every_density_setting() { |
| 327 | let total = 30usize; |
| 328 | let last = format!("row {:02} plain content", total - 1); |
| 329 | |
| 330 | for (label, options) in [ |
| 331 | ("default", TranscriptRenderOptions::default()), |
| 332 | ( |
| 333 | "tool details hidden", |
| 334 | TranscriptRenderOptions { |
| 335 | show_tool_details: false, |
| 336 | ..TranscriptRenderOptions::default() |
| 337 | }, |
| 338 | ), |
| 339 | ( |
| 340 | "calm mode", |
| 341 | TranscriptRenderOptions { |
| 342 | calm_mode: true, |
| 343 | ..TranscriptRenderOptions::default() |
| 344 | }, |
| 345 | ), |
| 346 | ] { |
| 347 | let cell = { |
| 348 | let mut cell = generic_tool("read_file", ToolStatus::Failed); |
| 349 | cell.input_summary = Some("command: noisy".to_string()); |
| 350 | cell.output = Some(numbered_output(total)); |
| 351 | HistoryCell::Tool(ToolCell::Generic(cell)) |
| 352 | }; |
| 353 | |
| 354 | let text = lines_text(&cell.lines_with_options(80, options)); |
| 355 | assert!( |
| 356 | text.contains("lines omitted"), |
| 357 | "[{label}] a bounded failure must advertise omitted output: {text}" |
| 358 | ); |
| 359 | assert!( |
| 360 | text.contains(&last), |
| 361 | "[{label}] failed output must stay expanded to its last row: {text}" |
| 362 | ); |
| 363 | assert!( |
| 364 | text.contains("command: noisy"), |
| 365 | "[{label}] the failing invocation must stay visible: {text}" |
| 366 | ); |
| 367 | } |
| 368 | } |
| 369 | |
| 370 | /// The live surface is a summary and the transcript is the record. The contract |
| 371 | /// is directional: anything the live view drops must still be in the |
| 372 | /// transcript, and the live view must say so when it drops something. A success |
| 373 | /// gets a bounded preview; a failure gets the full budget; neither may leave |
| 374 | /// the transcript short. |
| 375 | /// |
| 376 | /// Replaces four near-identical live/transcript comparison tests. |
| 377 | #[test] |
| 378 | fn whatever_live_truncates_the_transcript_still_holds() { |
| 379 | let total = 30usize; |
| 380 | let first = "row 00 plain content"; |
| 381 | let last = format!("row {:02} plain content", total - 1); |
| 382 | |
| 383 | // Failed exec: capped live with an honest marker, uncapped in transcript. |
| 384 | let failed = { |
| 385 | let mut exec = exec_tool("noisy_script.sh", ToolStatus::Failed); |
| 386 | exec.output = Some(numbered_output(total)); |
| 387 | exec.duration_ms = Some(120); |
| 388 | HistoryCell::Tool(ToolCell::Exec(exec)) |
| 389 | }; |
| 390 | let live = failed.lines_with_options(80, calm_options()); |
| 391 | let transcript = failed.transcript_lines(80); |
| 392 | let live_text = lines_text(&live); |
| 393 | let transcript_text = lines_text(&transcript); |
| 394 | assert!( |
| 395 | live.len() < transcript.len(), |
| 396 | "live must compress (live={}, transcript={})", |
| 397 | live.len(), |
| 398 | transcript.len() |
| 399 | ); |
| 400 | assert!( |
| 401 | live_text.contains("lines omitted"), |
| 402 | "a live view that drops rows must say so: {live_text}" |
| 403 | ); |
| 404 | assert!( |
| 405 | !transcript_text.contains("lines omitted"), |
| 406 | "the transcript drops nothing, so it claims nothing: {transcript_text}" |
| 407 | ); |
| 408 | assert!(transcript_text.contains(first) && transcript_text.contains(&last)); |
| 409 | assert!( |
| 410 | transcript_text.contains("row 15 plain content"), |
| 411 | "the transcript keeps the middle the live view skipped: {transcript_text}" |
| 412 | ); |
| 413 | |
| 414 | // Successful exec: a bounded head preview, never the whole body. |
| 415 | let success = { |
| 416 | let mut exec = exec_tool("noisy_script.sh", ToolStatus::Success); |
| 417 | exec.output = Some(numbered_output(total)); |
| 418 | exec.duration_ms = Some(120); |
| 419 | HistoryCell::Tool(ToolCell::Exec(exec)) |
| 420 | }; |
| 421 | let live_text = lines_text(&success.lines_with_options(80, calm_options())); |
| 422 | let transcript_text = lines_text(&success.transcript_lines(80)); |
| 423 | let previewed = (0..total) |
| 424 | .filter(|i| live_text.contains(&format!("row {i:02} plain content"))) |
| 425 | .count(); |
| 426 | assert_eq!( |
| 427 | previewed, 3, |
| 428 | "a successful exec previews only the opening and final result rows: {live_text}" |
| 429 | ); |
| 430 | assert!( |
| 431 | live_text.contains(first) && live_text.contains(&last), |
| 432 | "the bounded preview retains context and the final result: {live_text}" |
| 433 | ); |
| 434 | let output_hint = crate::tui::key_shortcuts::tool_details_shortcut_action_hint("output"); |
| 435 | assert!( |
| 436 | live_text.contains(&output_hint), |
| 437 | "a shortened success must point to its full output: {live_text}" |
| 438 | ); |
| 439 | assert!( |
| 440 | !live_text.contains("row 15 plain content"), |
| 441 | "the live preview should leave routine middle output for details: {live_text}" |
| 442 | ); |
| 443 | assert!(transcript_text.contains(first) && transcript_text.contains(&last)); |
| 444 | assert!( |
| 445 | transcript_text.contains("row 15 plain content"), |
| 446 | "the full transcript must retain output omitted from the live card: {transcript_text}" |
| 447 | ); |
| 448 | |
| 449 | // Successful generic tool: output collapses entirely live, and does so |
| 450 | // without spending a row telling the user it collapsed. |
| 451 | let quiet = { |
| 452 | let mut cell = generic_tool("read_file", ToolStatus::Success); |
| 453 | cell.input_summary = Some("path: crates/tui/src/main.rs".to_string()); |
| 454 | cell.output = Some(numbered_output(24)); |
| 455 | HistoryCell::Tool(ToolCell::Generic(cell)) |
| 456 | }; |
| 457 | let live_text = lines_text(&quiet.lines_with_options(80, TranscriptRenderOptions::default())); |
| 458 | let transcript_text = lines_text(&quiet.transcript_lines(80)); |
| 459 | assert!( |
| 460 | !live_text.contains(first) && !live_text.contains("lines omitted"), |
| 461 | "a quiet success collapses silently: {live_text}" |
| 462 | ); |
| 463 | assert!(transcript_text.contains(first)); |
| 464 | assert!(transcript_text.contains("row 23 plain content")); |
| 465 | } |
| 466 | |
| 467 | /// Repro for #80: a `git diff --stat`-shaped result must keep its newlines on |
| 468 | /// the transcript surface — one file per row, not squashed into one line. |
| 469 | #[test] |
| 470 | fn multi_line_tool_output_keeps_one_row_per_source_line() { |
| 471 | let diff_stat = "Cargo.lock | 1 +\n\ |
| 472 | crates/cli/Cargo.toml | 1 +\n\ |
| 473 | crates/cli/src/main.rs | 47 ++++++\n\ |
| 474 | crates/config/src/lib.rs | 27 ++++\n\ |
| 475 | crates/tui/src/mcp.rs | 384 +++++"; |
| 476 | |
| 477 | let cell = { |
| 478 | let mut cell = generic_tool("read_file", ToolStatus::Success); |
| 479 | cell.input_summary = Some("command: git diff --stat".to_string()); |
| 480 | cell.output = Some(diff_stat.to_string()); |
| 481 | HistoryCell::Tool(ToolCell::Generic(cell)) |
| 482 | }; |
| 483 | |
| 484 | let transcript_text = lines_text(&cell.transcript_lines(80)); |
| 485 | for needle in [ |
| 486 | "Cargo.lock", |
| 487 | "crates/cli/Cargo.toml", |
| 488 | "crates/cli/src/main.rs", |
| 489 | "crates/config/src/lib.rs", |
| 490 | "crates/tui/src/mcp.rs", |
| 491 | ] { |
| 492 | assert!( |
| 493 | transcript_text.contains(needle), |
| 494 | "transcript missing {needle:?}: {transcript_text}" |
| 495 | ); |
| 496 | } |
| 497 | let cargo_lock_row = transcript_text |
| 498 | .lines() |
| 499 | .find(|line| line.contains("Cargo.lock")) |
| 500 | .expect("Cargo.lock row must exist"); |
| 501 | assert!( |
| 502 | !cargo_lock_row.contains("crates/cli/Cargo.toml"), |
| 503 | "two files were joined onto one row: {cargo_lock_row}" |
| 504 | ); |
| 505 | } |
| 506 | |
| 507 | /// Reasoning folds in the live view and the fold is reversible: the collapsed |
| 508 | /// state must truncate a long body, the expanded state must restore every line |
| 509 | /// it dropped, and both must show the model's own identifiers verbatim (the |
| 510 | /// #4146/#4148 scrub rendered `refresh_catalog_cache` as `…` and protected |
| 511 | /// nothing, since the body was always one keypress away). The configured |
| 512 | /// default only inverts which state the toggle starts in. |
| 513 | /// |
| 514 | /// Replaces four separate fold tests. |
| 515 | #[test] |
| 516 | fn reasoning_folds_in_live_and_the_fold_is_reversible() { |
| 517 | let body = (1..=20) |
| 518 | .map(|i| format!("step {i:02}: refresh_catalog_cache iteration")) |
| 519 | .collect::<Vec<_>>() |
| 520 | .join("\n"); |
| 521 | let cell = HistoryCell::Thinking { |
| 522 | content: body, |
| 523 | streaming: false, |
| 524 | duration_secs: Some(1.0), |
| 525 | }; |
| 526 | |
| 527 | for default_expanded in [false, true] { |
| 528 | let options = TranscriptRenderOptions { |
| 529 | thinking_default_expanded: default_expanded, |
| 530 | low_motion: true, |
| 531 | ..TranscriptRenderOptions::default() |
| 532 | }; |
| 533 | // An explicit intent says expanded or collapsed outright, so the same |
| 534 | // two calls answer for either configured default. Running both proves |
| 535 | // the intent is not re-read through the preference. |
| 536 | let expanded = lines_text( |
| 537 | &cell |
| 538 | .lines_with_options_folded(80, options, Some(ThinkingFold::Expanded)) |
| 539 | .0, |
| 540 | ); |
| 541 | let collapsed = lines_text( |
| 542 | &cell |
| 543 | .lines_with_options_folded(80, options, Some(ThinkingFold::Collapsed)) |
| 544 | .0, |
| 545 | ); |
| 546 | |
| 547 | for i in 1..=20 { |
| 548 | assert!( |
| 549 | expanded.contains(&format!("step {i:02}: refresh_catalog_cache iteration")), |
| 550 | "[default_expanded={default_expanded}] expanded reasoning dropped line {i}: \ |
| 551 | {expanded}" |
| 552 | ); |
| 553 | } |
| 554 | assert!( |
| 555 | !collapsed.contains("step 20:"), |
| 556 | "[default_expanded={default_expanded}] the collapsed fold must truncate: {collapsed}" |
| 557 | ); |
| 558 | assert!( |
| 559 | collapsed.contains("refresh_catalog_cache"), |
| 560 | "[default_expanded={default_expanded}] the shown head keeps identifiers \ |
| 561 | verbatim: {collapsed}" |
| 562 | ); |
| 563 | assert!( |
| 564 | !collapsed.contains("Space:") && !collapsed.contains("Ctrl+O"), |
| 565 | "[default_expanded={default_expanded}] the per-cell renderer stays \ |
| 566 | target-neutral; the chord belongs to whoever owns focus: {collapsed}" |
| 567 | ); |
| 568 | } |
| 569 | } |
| 570 | |
| 571 | /// The preference baseline (verbose session or expanded default) decides only |
| 572 | /// the cells nobody has touched; an explicit intent decides its own cell in |
| 573 | /// both directions. Verbose plus an expanded default renders expanded — the |
| 574 | /// old triple-XOR collapsed exactly that cell — and an explicit expand stays |
| 575 | /// expanded under every preference combination, which the later relative bit |
| 576 | /// still got wrong (#5847). |
| 577 | #[test] |
| 578 | fn explicit_thinking_fold_outranks_every_preference_baseline() { |
| 579 | let body = (1..=20) |
| 580 | .map(|i| format!("step {i:02}: baseline check")) |
| 581 | .collect::<Vec<_>>() |
| 582 | .join("\n"); |
| 583 | let cell = HistoryCell::Thinking { |
| 584 | content: body, |
| 585 | streaming: false, |
| 586 | duration_secs: Some(1.0), |
| 587 | }; |
| 588 | // (fold, verbose, default_expanded, expect_expanded) |
| 589 | for (fold, verbose, default_expanded, expect_expanded) in [ |
| 590 | // No explicit intent: the preference baseline decides. |
| 591 | (None, false, false, false), |
| 592 | (None, false, true, true), |
| 593 | (None, true, false, true), |
| 594 | (None, true, true, true), |
| 595 | // An explicit expand renders expanded whatever the preferences say. |
| 596 | (Some(ThinkingFold::Expanded), false, false, true), |
| 597 | (Some(ThinkingFold::Expanded), false, true, true), |
| 598 | (Some(ThinkingFold::Expanded), true, false, true), |
| 599 | (Some(ThinkingFold::Expanded), true, true, true), |
| 600 | // And an explicit collapse renders collapsed whatever they say. |
| 601 | (Some(ThinkingFold::Collapsed), false, false, false), |
| 602 | (Some(ThinkingFold::Collapsed), false, true, false), |
| 603 | (Some(ThinkingFold::Collapsed), true, false, false), |
| 604 | (Some(ThinkingFold::Collapsed), true, true, false), |
| 605 | ] { |
| 606 | let options = TranscriptRenderOptions { |
| 607 | verbose, |
| 608 | thinking_default_expanded: default_expanded, |
| 609 | low_motion: true, |
| 610 | ..TranscriptRenderOptions::default() |
| 611 | }; |
| 612 | let text = lines_text(&cell.lines_with_options_folded(80, options, fold).0); |
| 613 | let expanded = text.contains("step 20: baseline check"); |
| 614 | assert_eq!( |
| 615 | expanded, expect_expanded, |
| 616 | "[fold={fold:?} verbose={verbose} default_expanded={default_expanded}]" |
| 617 | ); |
| 618 | } |
| 619 | } |
| 620 | |
| 621 | /// A completed reasoning cell short enough to fit needs no expand affordance, |
| 622 | /// and the live view must still show it — the alternative was a dead card that |
| 623 | /// said reasoning happened and nothing about what it was. |
| 624 | #[test] |
| 625 | fn short_completed_reasoning_is_shown_live_without_an_affordance() { |
| 626 | let cell = HistoryCell::Thinking { |
| 627 | content: "One brief reasoning step.".to_string(), |
| 628 | streaming: false, |
| 629 | duration_secs: Some(0.4), |
| 630 | }; |
| 631 | |
| 632 | let live_text = lines_text(&cell.lines_with_options(80, calm_options())); |
| 633 | let transcript_text = lines_text(&cell.transcript_lines(80)); |
| 634 | |
| 635 | assert!( |
| 636 | live_text.contains("One brief reasoning step."), |
| 637 | "short completed reasoning belongs inline: {live_text}" |
| 638 | ); |
| 639 | assert!(transcript_text.contains("One brief reasoning step.")); |
| 640 | assert!( |
| 641 | !live_text.contains("Ctrl+O") && !live_text.contains("Space:"), |
| 642 | "a body that fits needs no affordance: {live_text}" |
| 643 | ); |
| 644 | } |
| 645 | |
| 646 | /// A live reasoning block must show what the model is thinking right now — the |
| 647 | /// old behavior stalled on a `thinking...` placeholder until the block closed, |
| 648 | /// and a long body must keep the newest line rather than the oldest. |
| 649 | #[test] |
| 650 | fn streaming_reasoning_shows_its_newest_line_not_a_placeholder() { |
| 651 | let short = render_thinking( |
| 652 | "Step 1: read the code\nStep 2: trace the call\nStep 3: form a hypothesis", |
| 653 | 80, |
| 654 | true, |
| 655 | None, |
| 656 | true, |
| 657 | true, |
| 658 | ); |
| 659 | let short_text = lines_text(&short); |
| 660 | assert!( |
| 661 | short_text.contains("Step 3: form a hypothesis"), |
| 662 | "the newest reasoning line must be visible while streaming: {short_text}" |
| 663 | ); |
| 664 | assert!( |
| 665 | !short_text.contains("thinking..."), |
| 666 | "real content means the placeholder must not be drawn: {short_text}" |
| 667 | ); |
| 668 | |
| 669 | let long = (1..=16) |
| 670 | .map(|i| format!("Reasoning line {i}")) |
| 671 | .collect::<Vec<_>>() |
| 672 | .join("\n"); |
| 673 | let long_text = lines_text(&render_thinking(&long, 80, true, None, true, true)); |
| 674 | assert!( |
| 675 | long_text.contains("Reasoning line 16"), |
| 676 | "the tail is what is live: {long_text}" |
| 677 | ); |
| 678 | assert!( |
| 679 | !long_text.contains("Reasoning line 1\n"), |
| 680 | "the head is what gets clipped: {long_text}" |
| 681 | ); |
| 682 | } |
| 683 | |
| 684 | /// A foreground shell wait blocks the turn. The card's job is to tell the user |
| 685 | /// how to take the terminal back, not to re-print the command they just watched |
| 686 | /// the model type, and not to duplicate the sidebar's live tail in the |
| 687 | /// transcript. Once the command finishes, the final output supersedes any stale |
| 688 | /// live tail. |
| 689 | /// |
| 690 | /// Replaces three tests. |
| 691 | #[test] |
| 692 | fn a_foreground_shell_wait_offers_the_escape_hatch_not_the_command_echo() { |
| 693 | let command = "cargo test --workspace --all-features"; |
| 694 | let running = { |
| 695 | let mut exec = exec_tool(command, ToolStatus::Running); |
| 696 | exec.live_output = Some("running line 1\nrunning line 2".to_string()); |
| 697 | exec.shell_task_id = Some("shell_live".to_string()); |
| 698 | exec |
| 699 | }; |
| 700 | |
| 701 | for (label, text) in [ |
| 702 | ("live", lines_text(&running.lines_with_motion(80, true))), |
| 703 | ( |
| 704 | "transcript", |
| 705 | lines_text(&HistoryCell::Tool(ToolCell::Exec(running.clone())).transcript_lines(80)), |
| 706 | ), |
| 707 | ] { |
| 708 | assert!( |
| 709 | text.contains("Ctrl+B"), |
| 710 | "[{label}] the backgrounding chord is the point of the card: {text}" |
| 711 | ); |
| 712 | assert!( |
| 713 | !text.contains("running line 1"), |
| 714 | "[{label}] the live tail belongs to the sidebar and /jobs: {text}" |
| 715 | ); |
| 716 | assert!( |
| 717 | !text.contains(command), |
| 718 | "[{label}] the header already carries the summary; do not echo the \ |
| 719 | command target: {text}" |
| 720 | ); |
| 721 | assert!(!text.contains("command:"), "[{label}] {text}"); |
| 722 | } |
| 723 | |
| 724 | let mut finished = exec_tool(command, ToolStatus::Success); |
| 725 | finished.output = Some("final output".to_string()); |
| 726 | finished.live_output = Some("stale live tail".to_string()); |
| 727 | finished.shell_task_id = Some("shell_live".to_string()); |
| 728 | let text = lines_text(&finished.lines_with_motion(80, true)); |
| 729 | assert!( |
| 730 | !text.contains("stale live tail"), |
| 731 | "a finished command must not show the tail it already superseded: {text}" |
| 732 | ); |
| 733 | } |
| 734 | |
| 735 | // --------------------------------------------------------------------------- |
| 736 | // Clipboard — what you copy is what was authored |
| 737 | // --------------------------------------------------------------------------- |
| 738 | |
| 739 | /// Every rendered line carries a `copy_prefix_width`: the display columns of |
| 740 | /// decoration the clipboard must skip. The property is that slicing a line at |
| 741 | /// that width yields the payload and nothing decorative — for role markers, |
| 742 | /// status chrome, and the two-column continuation prefix on wrapped fenced code |
| 743 | /// (which must be counted in display columns, not bytes, or CJK shifts it). |
| 744 | /// |
| 745 | /// Replaces three tests that each covered one cell kind. |
| 746 | #[test] |
| 747 | fn the_copy_prefix_skips_every_decoration_and_keeps_the_payload() { |
| 748 | let decorations = ['╎', '▎', '●', '│', '┃', '✓', '▏']; |
| 749 | |
| 750 | let copied_line = |cell: &HistoryCell, width: u16, needle: &str| -> (String, usize) { |
| 751 | let rendered = cell.lines_with_copy_metadata(width, TranscriptRenderOptions::default()); |
| 752 | let target = rendered |
| 753 | .iter() |
| 754 | .find(|entry| { |
| 755 | entry |
| 756 | .line |
| 757 | .spans |
| 758 | .iter() |
| 759 | .any(|span| span.content.contains(needle)) |
| 760 | }) |
| 761 | .unwrap_or_else(|| panic!("no rendered line contains {needle:?}")); |
| 762 | let text = line_to_plain(&target.line); |
| 763 | ( |
| 764 | slice_visible_columns(&text, target.copy_prefix_width, text_visible_width(&text)), |
| 765 | target.copy_prefix_width, |
| 766 | ) |
| 767 | }; |
| 768 | |
| 769 | // Fenced code: indentation survives, decoration does not. |
| 770 | let rust_fence = HistoryCell::Assistant { |
| 771 | content: "```rust\n let answer = 42;\n```".to_string(), |
| 772 | streaming: false, |
| 773 | }; |
| 774 | let (copied, _) = copied_line(&rust_fence, 40, "answer"); |
| 775 | assert!( |
| 776 | copied.contains(" let answer = 42;"), |
| 777 | "code indentation was not preserved: {copied:?}" |
| 778 | ); |
| 779 | for glyph in decorations { |
| 780 | assert!( |
| 781 | !copied.contains(glyph), |
| 782 | "decorative glyph {glyph:?} leaked into copied code: {copied:?}" |
| 783 | ); |
| 784 | } |
| 785 | |
| 786 | // Wrapped CJK code: the prefix is two *display* columns, not two bytes. |
| 787 | let cjk_fence = HistoryCell::Assistant { |
| 788 | content: "```text\n 中文 = 1\n```".to_string(), |
| 789 | streaming: false, |
| 790 | }; |
| 791 | let (copied, prefix) = copied_line(&cjk_fence, 24, "中文"); |
| 792 | assert_eq!( |
| 793 | prefix, 2, |
| 794 | "the continuation prefix is the role marker's two display columns" |
| 795 | ); |
| 796 | assert!( |
| 797 | copied.starts_with(" 中文"), |
| 798 | "wide-character indentation was mis-sliced: {copied:?}" |
| 799 | ); |
| 800 | |
| 801 | // Tool receipt: status and family chrome are prefix, the receipt text is |
| 802 | // payload. |
| 803 | let receipt = { |
| 804 | let mut exec = exec_tool("printf 'receipt'", ToolStatus::Success); |
| 805 | exec.output = Some("receipt".to_string()); |
| 806 | HistoryCell::Tool(ToolCell::Exec(exec)) |
| 807 | }; |
| 808 | let rendered = receipt.lines_with_copy_metadata(80, TranscriptRenderOptions::default()); |
| 809 | let header = rendered.first().expect("tool receipt header"); |
| 810 | assert!( |
| 811 | header.copy_prefix_width >= 4, |
| 812 | "status and family chrome should be measured as prefix, got {}", |
| 813 | header.copy_prefix_width |
| 814 | ); |
| 815 | let body = line_to_plain(&ratatui::text::Line::from( |
| 816 | header |
| 817 | .line |
| 818 | .spans |
| 819 | .iter() |
| 820 | .skip(1) |
| 821 | .cloned() |
| 822 | .collect::<Vec<_>>(), |
| 823 | )); |
| 824 | let copied = slice_visible_columns(&body, header.copy_prefix_width, text_visible_width(&body)); |
| 825 | assert!( |
| 826 | copied.contains("run Done"), |
| 827 | "receipt text was clipped away: {copied:?}" |
| 828 | ); |
| 829 | for glyph in decorations { |
| 830 | assert!( |
| 831 | !copied.contains(glyph), |
| 832 | "decorative glyph {glyph:?} leaked into the copied receipt: {copied:?}" |
| 833 | ); |
| 834 | } |
| 835 | } |
| 836 | |
| 837 | /// Issue #1212: the transcript rail (`▏`) marks prose continuation. Inside a |
| 838 | /// fence it corrupts anything the user copies, so no line of a code block may |
| 839 | /// carry it — not the first, not a blank line in the middle, not a wrapped |
| 840 | /// continuation of an over-long source line. |
| 841 | /// |
| 842 | /// Replaces four tests that each covered one fence shape. |
| 843 | #[test] |
| 844 | fn no_line_inside_a_fence_carries_the_transcript_rail() { |
| 845 | let long_source = "let x = ".to_string() + &"abcdef ".repeat(40); |
| 846 | |
| 847 | for (label, content, width) in [ |
| 848 | ( |
| 849 | "short fence", |
| 850 | "SQL:\n```sql\nSELECT\nFROM customers\n```".to_string(), |
| 851 | 80u16, |
| 852 | ), |
| 853 | ( |
| 854 | "multi-line fence", |
| 855 | "Here's the query:\n```sql\nSELECT\n c.customer_id,\n c.name,\n \ |
| 856 | COUNT(o.order_id) AS order_count\nFROM customers c\nJOIN orders o ON \ |
| 857 | c.customer_id = o.customer_id;\n```" |
| 858 | .to_string(), |
| 859 | 80, |
| 860 | ), |
| 861 | ( |
| 862 | "fence with a blank line", |
| 863 | "```\nfn one() {}\n\nfn two() {}\n```".to_string(), |
| 864 | 80, |
| 865 | ), |
| 866 | ("wrapped fence", format!("```\n{long_source}\n```"), 40), |
| 867 | ] { |
| 868 | let cell = HistoryCell::Assistant { |
| 869 | content, |
| 870 | streaming: false, |
| 871 | }; |
| 872 | // Line 0 is the intro paragraph (or the fence opener); every line |
| 873 | // after it belongs to the code block. |
| 874 | for line in cell.lines(width).iter().skip(1) { |
| 875 | let text = line_text(line); |
| 876 | assert!( |
| 877 | !text.contains('\u{258F}'), |
| 878 | "[{label}] code line took the transcript rail: {text:?}" |
| 879 | ); |
| 880 | } |
| 881 | } |
| 882 | } |
| 883 | |
| 884 | /// Whose text gets interpreted is a trust boundary. The model's markdown is |
| 885 | /// rendered; the user's prompt is shown exactly as typed, including leading |
| 886 | /// hashes, dashes and runs of spaces. A cell holding only whitespace renders |
| 887 | /// nothing at all rather than an orphaned role glyph. |
| 888 | /// |
| 889 | /// Replaces three tests. |
| 890 | #[test] |
| 891 | fn authored_text_keeps_its_shape_on_both_sides_of_the_turn() { |
| 892 | let user = HistoryCell::User { |
| 893 | content: " # heading\n- item\n \nhello world".to_string(), |
| 894 | }; |
| 895 | let visible: Vec<String> = user.lines(80).iter().map(line_text).collect(); |
| 896 | assert!( |
| 897 | visible[0].trim_end().ends_with("# heading"), |
| 898 | "a user's literal `#` must not become a rendered heading: {visible:?}" |
| 899 | ); |
| 900 | assert!( |
| 901 | visible[1].trim_end().ends_with("- item"), |
| 902 | "dash-prefixed user text stays literal: {visible:?}" |
| 903 | ); |
| 904 | assert!( |
| 905 | visible[2].ends_with(" "), |
| 906 | "whitespace-only user lines survive: {visible:?}" |
| 907 | ); |
| 908 | assert!( |
| 909 | visible[3].trim_end().ends_with("hello world"), |
| 910 | "internal spacing stays literal: {visible:?}" |
| 911 | ); |
| 912 | assert!( |
| 913 | !visible.iter().any(|line| line.contains('\u{2500}')), |
| 914 | "user text must not gain a markdown heading rule: {visible:?}" |
| 915 | ); |
| 916 | |
| 917 | let assistant = HistoryCell::Assistant { |
| 918 | content: "# Heading\n\n- item".to_string(), |
| 919 | streaming: false, |
| 920 | }; |
| 921 | let visible: Vec<String> = assistant.lines(80).iter().map(line_text).collect(); |
| 922 | assert!( |
| 923 | visible[0].contains("Heading") && !visible[0].contains("# Heading"), |
| 924 | "the model's markdown is still parsed: {visible:?}" |
| 925 | ); |
| 926 | assert!( |
| 927 | visible.iter().any(|line| line.contains('\u{2500}')), |
| 928 | "an assistant h1 still draws its rule: {visible:?}" |
| 929 | ); |
| 930 | |
| 931 | // A stray newline streamed between reasoning and a tool call used to render |
| 932 | // as a bare role glyph with nothing after it. |
| 933 | for content in ["", " ", "\n", "\n\n", " \t \n"] { |
| 934 | for streaming in [false, true] { |
| 935 | let cell = HistoryCell::Assistant { |
| 936 | content: content.to_string(), |
| 937 | streaming, |
| 938 | }; |
| 939 | assert!( |
| 940 | cell.lines(80).is_empty(), |
| 941 | "whitespace-only assistant content {content:?} (streaming={streaming}) \ |
| 942 | must render nothing" |
| 943 | ); |
| 944 | } |
| 945 | } |
| 946 | let real = HistoryCell::Assistant { |
| 947 | content: "hi".to_string(), |
| 948 | streaming: false, |
| 949 | }; |
| 950 | assert_eq!( |
| 951 | real.lines(80)[0].spans[0].content.as_ref(), |
| 952 | ASSISTANT_GLYPH, |
| 953 | "real content still gets its role marker" |
| 954 | ); |
| 955 | } |
| 956 | |
| 957 | /// Reasoning is neither the user's prompt nor the model's answer, and a reader |
| 958 | /// scanning the transcript has to be able to skip it. The markers below are |
| 959 | /// referenced as named constants rather than literal glyphs on purpose: this |
| 960 | /// protects the *distinction*, so a redesign that restyles reasoning stays |
| 961 | /// green while one that stops marking it at all fails. |
| 962 | #[test] |
| 963 | fn reasoning_is_marked_apart_from_both_the_prompt_and_the_answer() { |
| 964 | let body_text = "concrete reasoning content"; |
| 965 | let reasoning = render_thinking(body_text, 80, false, Some(1.0), false, true); |
| 966 | assert!(reasoning.len() >= 2, "expected a header and a body line"); |
| 967 | |
| 968 | let header = line_text(&reasoning[0]); |
| 969 | assert!( |
| 970 | header.starts_with(REASONING_OPENER), |
| 971 | "the reasoning header opens with its own marker: {header:?}" |
| 972 | ); |
| 973 | let body = line_text(&reasoning[1]); |
| 974 | assert!( |
| 975 | body.starts_with(REASONING_RAIL), |
| 976 | "the reasoning body carries its own rail: {body:?}" |
| 977 | ); |
| 978 | |
| 979 | let rail = REASONING_RAIL.trim(); |
| 980 | for cell in [ |
| 981 | HistoryCell::User { |
| 982 | content: body_text.to_string(), |
| 983 | }, |
| 984 | HistoryCell::Assistant { |
| 985 | content: body_text.to_string(), |
| 986 | streaming: false, |
| 987 | }, |
| 988 | ] { |
| 989 | let rendered = lines_text(&cell.lines(80)); |
| 990 | assert!( |
| 991 | rendered.contains(body_text), |
| 992 | "sanity: the cell rendered its content: {rendered}" |
| 993 | ); |
| 994 | assert!( |
| 995 | !rendered.contains(rail), |
| 996 | "only reasoning may wear the reasoning rail: {rendered}" |
| 997 | ); |
| 998 | } |
| 999 | } |
| 1000 | |
| 1001 | /// A filled background behind reasoning is unreadable on a transparent or |
| 1002 | /// light terminal, so the highlight is configurable. With it off, not one span |
| 1003 | /// may carry a background — a single tinted span is the bug. The enabled case |
| 1004 | /// follows the terminal's actual color depth: capable terminals tint the body, |
| 1005 | /// while ANSI-16 intentionally stays untinted because it cannot render the |
| 1006 | /// subtle surface faithfully. |
| 1007 | #[test] |
| 1008 | fn disabling_the_reasoning_highlight_leaves_no_span_with_a_background() { |
| 1009 | let render = |highlight: bool| { |
| 1010 | render_thinking_with_analysis( |
| 1011 | "reasoning without a filled surface", |
| 1012 | 80, |
| 1013 | false, |
| 1014 | Some(1.0), |
| 1015 | false, |
| 1016 | true, |
| 1017 | highlight, |
| 1018 | ) |
| 1019 | .0 |
| 1020 | }; |
| 1021 | |
| 1022 | assert!( |
| 1023 | render(false) |
| 1024 | .iter() |
| 1025 | .flat_map(|line| line.spans.iter()) |
| 1026 | .all(|span| span.style.bg.is_none()), |
| 1027 | "a disabled highlight must not tint any span" |
| 1028 | ); |
| 1029 | let enabled_has_background = render(true) |
| 1030 | .iter() |
| 1031 | .flat_map(|line| line.spans.iter()) |
| 1032 | .any(|span| span.style.bg.is_some()); |
| 1033 | assert_eq!( |
| 1034 | enabled_has_background, |
| 1035 | codewhale_palette::reasoning_surface_tint(cached_color_depth()).is_some(), |
| 1036 | "the enabled highlight must follow the terminal color-depth contract" |
| 1037 | ); |
| 1038 | } |
| 1039 | |
| 1040 | // --------------------------------------------------------------------------- |
| 1041 | // Motion — reduced motion is actually still |
| 1042 | // --------------------------------------------------------------------------- |
| 1043 | |
| 1044 | /// The deleted tests pinned the frozen glyphs (`assert_eq!(spans[1], "⣤")`). |
| 1045 | /// That breaks on a skin change and passes on the bug that matters: a marker |
| 1046 | /// that keeps animating for a user who asked it to stop. The property is |
| 1047 | /// stillness — a running cell's rendered frame must not depend on how long it |
| 1048 | /// has been running once motion is reduced — and the full-motion case is |
| 1049 | /// asserted alongside it so a renderer that froze everything could not make |
| 1050 | /// this test vacuously true. |
| 1051 | /// |
| 1052 | /// Stillness is not enough on its own. Animation frame 0 is U+2800 BRAILLE |
| 1053 | /// PATTERN BLANK, an invisible cell. Freezing there (or on the Still path) |
| 1054 | /// looks like a missing marker, which is why reduced motion must freeze on a |
| 1055 | /// filled, legible bubble rather than the blank the spinner starts on. |
| 1056 | #[test] |
| 1057 | fn reduced_and_still_motion_render_a_frame_that_does_not_move() { |
| 1058 | let frame_symbols = super::TOOL_RUNNING_SYMBOLS.len() as u64; |
| 1059 | let frame_at = |elapsed_ms: u64, low_motion: bool, motion: MotionMode| { |
| 1060 | let mut exec = exec_tool("echo hi", ToolStatus::Running); |
| 1061 | exec.started_at = Some(Instant::now() - Duration::from_millis(elapsed_ms)); |
| 1062 | let cell = HistoryCell::Tool(ToolCell::Exec(exec)); |
| 1063 | lines_text(&cell.lines_with_options( |
| 1064 | 80, |
| 1065 | TranscriptRenderOptions { |
| 1066 | low_motion, |
| 1067 | motion_mode: motion, |
| 1068 | ..TranscriptRenderOptions::default() |
| 1069 | }, |
| 1070 | )) |
| 1071 | }; |
| 1072 | |
| 1073 | // Half a spinner cycle apart, and both well under the 3s elapsed-badge |
| 1074 | // threshold so the badge itself cannot be the thing that differs. |
| 1075 | let early = crate::tui::spinner::LIVE_MARKER_DELAY_MS; |
| 1076 | let late = early + super::TOOL_STATUS_SYMBOL_MS * (frame_symbols / 2); |
| 1077 | assert!( |
| 1078 | late < 3_000, |
| 1079 | "both samples must stay under the elapsed badge" |
| 1080 | ); |
| 1081 | |
| 1082 | // Two independent mechanisms are supposed to produce stillness — the |
| 1083 | // `low_motion` flag and the resolved `motion_mode`. Each is asserted on its |
| 1084 | // own so losing either one fails here, rather than only losing both. |
| 1085 | for (low_motion, motion) in [ |
| 1086 | (true, MotionMode::Reduced), |
| 1087 | (true, MotionMode::Still), |
| 1088 | (false, MotionMode::Reduced), |
| 1089 | (false, MotionMode::Still), |
| 1090 | ] { |
| 1091 | let frozen = frame_at(early, low_motion, motion); |
| 1092 | assert_eq!( |
| 1093 | frozen, |
| 1094 | frame_at(late, low_motion, motion), |
| 1095 | "low_motion={low_motion} / {motion:?} must not animate the live marker" |
| 1096 | ); |
| 1097 | assert!( |
| 1098 | !frozen.contains('\u{2800}'), |
| 1099 | "a frozen marker must still be visible: low_motion={low_motion} / {motion:?}: {frozen:?}" |
| 1100 | ); |
| 1101 | } |
| 1102 | assert_ne!( |
| 1103 | frame_at(early, false, MotionMode::Full), |
| 1104 | frame_at(late, false, MotionMode::Full), |
| 1105 | "full motion must actually animate, or the stillness assertions above \ |
| 1106 | prove nothing" |
| 1107 | ); |
| 1108 | |
| 1109 | // The same contract for the two other animated surfaces: the streaming |
| 1110 | // reasoning cursor and the assistant role marker's pulse. |
| 1111 | let cursor_off = lines_text(&render_thinking( |
| 1112 | "ongoing reasoning...", |
| 1113 | 80, |
| 1114 | true, |
| 1115 | None, |
| 1116 | false, |
| 1117 | true, |
| 1118 | )); |
| 1119 | assert!( |
| 1120 | !cursor_off.contains(REASONING_CURSOR), |
| 1121 | "low motion must suppress the streaming reasoning cursor: {cursor_off}" |
| 1122 | ); |
| 1123 | assert_eq!( |
| 1124 | assistant_label_style_for(true, true).fg, |
| 1125 | assistant_label_style_for(false, false).fg, |
| 1126 | "a streaming assistant marker under low motion must look exactly like an \ |
| 1127 | idle one — no pulse" |
| 1128 | ); |
| 1129 | } |
| 1130 | |
| 1131 | /// Dual of the low-motion freeze above: when the cell is streaming and |
| 1132 | /// motion is allowed, the assistant marker must actually pulse. The deleted |
| 1133 | /// test slept up to 1s sampling `SystemTime` until the 2s sine dipped; the |
| 1134 | /// property is that the streaming+motion color is `pulse_brightness` of the |
| 1135 | /// idle source, which we can check without waiting on the wall clock. |
| 1136 | /// |
| 1137 | /// Around the sine crest, `pulse_brightness` rounds back to the source |
| 1138 | /// (~70ms of a 2s cycle). Matching the current instant would then also pass |
| 1139 | /// a renderer that never pulsed, so we only compare once the pure function |
| 1140 | /// itself is off the crest — a busy wait, not a sleep. |
| 1141 | #[test] |
| 1142 | fn assistant_marker_pulses_when_streaming_and_motion_is_allowed() { |
| 1143 | use codewhale_palette::{self as palette, pulse_brightness}; |
| 1144 | |
| 1145 | let idle = assistant_label_style_for(false, false).fg; |
| 1146 | assert_eq!( |
| 1147 | idle, |
| 1148 | Some(palette::WHALE_ACTION), |
| 1149 | "the idle marker is the unpulsed source; pulsing everything would make \ |
| 1150 | the streaming assertion vacuously true" |
| 1151 | ); |
| 1152 | assert_eq!( |
| 1153 | assistant_label_style_for(true, true).fg, |
| 1154 | idle, |
| 1155 | "low motion must keep the streaming marker at the unpulsed source" |
| 1156 | ); |
| 1157 | |
| 1158 | let epoch_ms = || { |
| 1159 | std::time::SystemTime::now() |
| 1160 | .duration_since(std::time::UNIX_EPOCH) |
| 1161 | .map(|d| d.as_millis() as u64) |
| 1162 | .unwrap_or(0) |
| 1163 | }; |
| 1164 | let deadline = Instant::now() + Duration::from_millis(250); |
| 1165 | let (t0, actual) = loop { |
| 1166 | assert!( |
| 1167 | Instant::now() < deadline, |
| 1168 | "pulse_brightness stayed at the source color through a 250ms spin; \ |
| 1169 | the 2s cycle leaves the crest in ~70ms" |
| 1170 | ); |
| 1171 | let t0 = epoch_ms(); |
| 1172 | // Skip the crest and a few ms of margin so the product read of |
| 1173 | // SystemTime cannot land back on identity between this sample and |
| 1174 | // the call under test. |
| 1175 | let near_crest = (t0.saturating_sub(8)..=t0.saturating_add(8)) |
| 1176 | .any(|ms| pulse_brightness(palette::WHALE_ACTION, ms) == palette::WHALE_ACTION); |
| 1177 | if near_crest { |
| 1178 | continue; |
| 1179 | } |
| 1180 | break (t0, assistant_label_style_for(true, false).fg); |
| 1181 | }; |
| 1182 | let t1 = epoch_ms(); |
| 1183 | let matches_pulse = |
| 1184 | (t0..=t1.max(t0)).any(|ms| actual == Some(pulse_brightness(palette::WHALE_ACTION, ms))); |
| 1185 | assert!( |
| 1186 | matches_pulse, |
| 1187 | "streaming + motion must apply pulse_brightness to the assistant \ |
| 1188 | marker, got {actual:?}" |
| 1189 | ); |
| 1190 | assert_ne!( |
| 1191 | actual, idle, |
| 1192 | "streaming + motion must not sit at the idle color once the pulse \ |
| 1193 | is off its crest" |
| 1194 | ); |
| 1195 | } |
| 1196 | |
| 1197 | /// The still-motion path rewrites the leading status marker in place. It once |
| 1198 | /// rewrote any braille cell it found, which silently ate braille that was part |
| 1199 | /// of the tool's own output. |
| 1200 | #[test] |
| 1201 | fn the_still_marker_rewrite_never_consumes_braille_tool_output() { |
| 1202 | let mut cell = generic_tool("read_file", ToolStatus::Running); |
| 1203 | cell.output = Some("⣿".to_string()); |
| 1204 | let cell = HistoryCell::Tool(ToolCell::Generic(cell)); |
| 1205 | |
| 1206 | let lines = cell.lines_with_options( |
| 1207 | 80, |
| 1208 | TranscriptRenderOptions { |
| 1209 | low_motion: true, |
| 1210 | motion_mode: MotionMode::Still, |
| 1211 | ..TranscriptRenderOptions::default() |
| 1212 | }, |
| 1213 | ); |
| 1214 | |
| 1215 | assert!( |
| 1216 | lines |
| 1217 | .iter() |
| 1218 | .flat_map(|line| line.spans.iter()) |
| 1219 | .any(|span| span.content.as_ref() == "⣿"), |
| 1220 | "tool output must survive the typed-header marker pass: {lines:?}" |
| 1221 | ); |
| 1222 | } |
| 1223 | |
| 1224 | // --------------------------------------------------------------------------- |
| 1225 | // Identity — a card never names a verb or a tool it did not run |
| 1226 | // --------------------------------------------------------------------------- |
| 1227 | |
| 1228 | /// #4145: a completed grep grouped under the exploration card rendered |
| 1229 | /// `read done · Searching …` — the header verb contradicted the label directly |
| 1230 | /// under it. The verb must agree with the work, in every locale, and a locale |
| 1231 | /// must not fall back to the English status word. |
| 1232 | /// |
| 1233 | /// Replaces two tests, each of which hard-coded one direction. |
| 1234 | #[test] |
| 1235 | fn a_card_verb_agrees_with_its_own_label_in_every_locale() { |
| 1236 | use codewhale_localization::Locale; |
| 1237 | |
| 1238 | for (label, expected_en, expected_zh, forbidden_en) in [ |
| 1239 | ( |
| 1240 | "Searching for `TranscriptScroll`", |
| 1241 | "find Done", |
| 1242 | "find 完成", |
| 1243 | "read Done", |
| 1244 | ), |
| 1245 | ("Reading src/foo.rs", "read Done", "read 完成", "find Done"), |
| 1246 | ] { |
| 1247 | let cell = super::ExploringCell { |
| 1248 | entries: vec![super::ExploringEntry { |
| 1249 | label: label.to_string(), |
| 1250 | status: ToolStatus::Success, |
| 1251 | }], |
| 1252 | }; |
| 1253 | |
| 1254 | let header_en = line_text(&cell.lines_with_motion(80, true)[0]); |
| 1255 | assert!( |
| 1256 | header_en.contains(expected_en), |
| 1257 | "{label:?} should read {expected_en:?}: {header_en:?}" |
| 1258 | ); |
| 1259 | assert!( |
| 1260 | !header_en.contains(forbidden_en), |
| 1261 | "{label:?} must not be paired with {forbidden_en:?}: {header_en:?}" |
| 1262 | ); |
| 1263 | assert!( |
| 1264 | header_en.contains(label), |
| 1265 | "the label itself must survive: {header_en:?}" |
| 1266 | ); |
| 1267 | |
| 1268 | let header_zh = line_text(&cell.lines_with_motion_and_locale(80, true, Locale::ZhHans)[0]); |
| 1269 | assert!( |
| 1270 | header_zh.contains(expected_zh), |
| 1271 | "{label:?} should read {expected_zh:?} in zh-Hans: {header_zh:?}" |
| 1272 | ); |
| 1273 | assert!( |
| 1274 | !header_zh.to_lowercase().contains("done"), |
| 1275 | "zh-Hans must not leak the English status word: {header_zh:?}" |
| 1276 | ); |
| 1277 | assert!( |
| 1278 | header_zh.contains(label), |
| 1279 | "the label itself must survive localization: {header_zh:?}" |
| 1280 | ); |
| 1281 | } |
| 1282 | } |
| 1283 | |
| 1284 | /// A read/find receipt reports a line count, so the count has to be real — |
| 1285 | /// including the singular/plural and the localized unit. A run receipt reports |
| 1286 | /// no count at all: inferring "3 lines" from rendered text that happens to |
| 1287 | /// contain `stdout:` would be inventing a number the shell never reported. |
| 1288 | #[test] |
| 1289 | fn receipts_count_only_what_they_actually_counted() { |
| 1290 | use crate::tui::widgets::tool_card::ToolFamily; |
| 1291 | use codewhale_localization::Locale; |
| 1292 | |
| 1293 | for (locale, done, unit) in [(Locale::En, "Done", "line"), (Locale::ZhHans, "完成", "行")] { |
| 1294 | let label = |family, status, output| { |
| 1295 | super::tool_receipt_label(family, status, Some(output), locale) |
| 1296 | }; |
| 1297 | |
| 1298 | assert_eq!(label(ToolFamily::Read, ToolStatus::Success, ""), done); |
| 1299 | assert_eq!( |
| 1300 | label(ToolFamily::Read, ToolStatus::Success, "hello\n"), |
| 1301 | if locale == Locale::En { |
| 1302 | "1 line".to_string() |
| 1303 | } else { |
| 1304 | format!("1 {unit}") |
| 1305 | } |
| 1306 | ); |
| 1307 | assert_eq!( |
| 1308 | label(ToolFamily::Read, ToolStatus::Success, "a\nb\nc\n"), |
| 1309 | if locale == Locale::En { |
| 1310 | "3 lines".to_string() |
| 1311 | } else { |
| 1312 | format!("3 {unit}") |
| 1313 | } |
| 1314 | ); |
| 1315 | assert_eq!( |
| 1316 | label(ToolFamily::Find, ToolStatus::Success, "match 1\nmatch 2\n"), |
| 1317 | if locale == Locale::En { |
| 1318 | "2 lines".to_string() |
| 1319 | } else { |
| 1320 | format!("2 {unit}") |
| 1321 | } |
| 1322 | ); |
| 1323 | |
| 1324 | // Run never counts, whatever the body looks like. |
| 1325 | for body in [ |
| 1326 | "stdout:\nok\nmore\nstderr:\nbad\n", |
| 1327 | "line 1\nline 2\nline 3\n", |
| 1328 | ] { |
| 1329 | assert_eq!( |
| 1330 | label(ToolFamily::Run, ToolStatus::Success, body), |
| 1331 | done, |
| 1332 | "a run receipt must not infer counts from {body:?}" |
| 1333 | ); |
| 1334 | } |
| 1335 | } |
| 1336 | |
| 1337 | assert_eq!( |
| 1338 | super::tool_receipt_label( |
| 1339 | ToolFamily::Read, |
| 1340 | ToolStatus::Running, |
| 1341 | Some("a\nb"), |
| 1342 | Locale::En |
| 1343 | ), |
| 1344 | "running", |
| 1345 | "an unfinished read has nothing to count yet" |
| 1346 | ); |
| 1347 | } |
| 1348 | |
| 1349 | /// The same truthfulness contract through the real shell render path, where a |
| 1350 | /// formatter has already rewritten the output: the header still reports a plain |
| 1351 | /// localized completion and never a fabricated line count or stream name. |
| 1352 | #[test] |
| 1353 | fn shell_headers_stay_truthful_through_the_output_formatters() { |
| 1354 | use codewhale_localization::Locale; |
| 1355 | |
| 1356 | let cases = [ |
| 1357 | ( |
| 1358 | "printf redirect", |
| 1359 | "printf '%s\\n' 'hello' 'world' > src/main.rs", |
| 1360 | "printf > src/main.rs\nhello\nworld\n", |
| 1361 | ), |
| 1362 | ( |
| 1363 | "logical-or fallback", |
| 1364 | "cargo build || echo fallback", |
| 1365 | " Compiling pkg v0.1.0\n Finished dev [unoptimized + debuginfo]\n", |
| 1366 | ), |
| 1367 | ]; |
| 1368 | |
| 1369 | for (label, command, output) in cases { |
| 1370 | let mut cell = exec_tool(command, ToolStatus::Success); |
| 1371 | cell.output = Some(output.to_string()); |
| 1372 | cell.duration_ms = Some(42); |
| 1373 | |
| 1374 | for (locale, done, unit) in [ |
| 1375 | (Locale::En, "Done", "lines"), |
| 1376 | (Locale::ZhHans, "完成", "行"), |
| 1377 | ] { |
| 1378 | let header = line_text(&cell.render_with_locale(80, true, RenderMode::Live, locale)[0]); |
| 1379 | assert!( |
| 1380 | header.contains(done), |
| 1381 | "[{label}] header must carry the localized completion: {header}" |
| 1382 | ); |
| 1383 | assert!( |
| 1384 | !header.contains(unit) && !header.contains("stdout") && !header.contains("stderr"), |
| 1385 | "[{label}] header must not invent counts or stream names: {header}" |
| 1386 | ); |
| 1387 | } |
| 1388 | } |
| 1389 | } |
| 1390 | |
| 1391 | /// #4133 / #4148: a spawn yields its card entirely to the DelegateCard, and an |
| 1392 | /// inspection (`peek` / `wait` / `status`) is a one-line check in every render |
| 1393 | /// mode. It must name the child it checked, must not read as a completed |
| 1394 | /// delegation, and must not leak the internal "unknown child" placeholder or |
| 1395 | /// echo the verb twice when the resolved identity collapses onto it. |
| 1396 | /// |
| 1397 | /// Replaces eight tests. |
| 1398 | #[test] |
| 1399 | fn agent_cards_stay_one_line_and_spawn_cards_yield_to_the_delegate_card() { |
| 1400 | let agent = |summary: &str, output: Option<&str>| { |
| 1401 | let mut cell = generic_tool("agent", ToolStatus::Success); |
| 1402 | cell.input_summary = Some(summary.to_string()); |
| 1403 | cell.output = output.map(str::to_string); |
| 1404 | cell |
| 1405 | }; |
| 1406 | |
| 1407 | for mode in [RenderMode::Live, RenderMode::Transcript] { |
| 1408 | let spawn = agent( |
| 1409 | "prompt: map the repo", |
| 1410 | Some(r#"{"agent_id":"agent_scout_1","status":"running"}"#), |
| 1411 | ); |
| 1412 | assert!( |
| 1413 | spawn.lines_with_mode(120, true, mode).is_empty(), |
| 1414 | "a spawn must not draw a generic card beside the DelegateCard in {mode:?}" |
| 1415 | ); |
| 1416 | |
| 1417 | for (summary, output, expected) in [ |
| 1418 | ( |
| 1419 | "action: peek agent_id: agent_scout_1", |
| 1420 | Some(r#"{"agent_id":"agent_scout_1","status":"running"}"#), |
| 1421 | "checked", |
| 1422 | ), |
| 1423 | ( |
| 1424 | "action: wait", |
| 1425 | Some(r#"{"action":"wait","settled":[{"agent_id":"agent_scout_1"}]}"#), |
| 1426 | "waited", |
| 1427 | ), |
| 1428 | ( |
| 1429 | "action: status agent_id: agent_scout_1", |
| 1430 | Some(r#"{"agent_id":"agent_scout_1","status":"running","terminal":false}"#), |
| 1431 | "checked", |
| 1432 | ), |
| 1433 | ] { |
| 1434 | let cell = agent(summary, output); |
| 1435 | let lines = cell.lines_with_mode(120, true, mode); |
| 1436 | let text = lines_text(&lines); |
| 1437 | assert_eq!( |
| 1438 | lines.len(), |
| 1439 | 1, |
| 1440 | "{summary:?} must stay one line in {mode:?}: {lines:?}" |
| 1441 | ); |
| 1442 | assert!( |
| 1443 | text.contains(expected), |
| 1444 | "{summary:?} should read as {expected:?}: {text:?}" |
| 1445 | ); |
| 1446 | assert!( |
| 1447 | !text.to_lowercase().contains("delegate done"), |
| 1448 | "an inspection must not read as a finished delegation: {text:?}" |
| 1449 | ); |
| 1450 | } |
| 1451 | } |
| 1452 | |
| 1453 | // Identity fallbacks: no raw placeholder, no doubled verb. |
| 1454 | let unresolved = agent("action: peek agent_type: delegate", None); |
| 1455 | let text = lines_text(&unresolved.lines_with_mode(80, true, RenderMode::Live)); |
| 1456 | assert!( |
| 1457 | !text.contains("unknown child"), |
| 1458 | "the internal fallback token must not reach the transcript: {text:?}" |
| 1459 | ); |
| 1460 | |
| 1461 | let collapsing = agent("action: peek role: delegate", None); |
| 1462 | let text = lines_text(&collapsing.lines_with_mode(80, true, RenderMode::Live)); |
| 1463 | assert!( |
| 1464 | text.contains(" agent "), |
| 1465 | "the agent family label should appear once in the summary: {text:?}" |
| 1466 | ); |
| 1467 | } |
| 1468 | |
| 1469 | /// A tool the catalog does not have produces one useful sentence — the catalog |
| 1470 | /// error — and nothing else. The old rendering spent a `name:` / `args:` / |
| 1471 | /// `result:` block restating a call that never happened. |
| 1472 | #[test] |
| 1473 | fn an_unknown_tool_failure_shows_only_the_catalog_error() { |
| 1474 | let mut cell = generic_tool("item", ToolStatus::Failed); |
| 1475 | cell.input_summary = Some("status: pending".to_string()); |
| 1476 | cell.output = Some( |
| 1477 | "Tool 'item' is not available in the current tool catalog. \ |
| 1478 | Checklist entries are not separate tool calls." |
| 1479 | .to_string(), |
| 1480 | ); |
| 1481 | |
| 1482 | for mode in [RenderMode::Live, RenderMode::Transcript] { |
| 1483 | let lines = cell.lines_with_mode(120, true, mode); |
| 1484 | let text = lines_text(&lines); |
| 1485 | assert_eq!(lines.len(), 1, "single header line in {mode:?}: {lines:?}"); |
| 1486 | assert!( |
| 1487 | text.contains("Tool 'item' is not available"), |
| 1488 | "the catalog error is the useful part: {text:?}" |
| 1489 | ); |
| 1490 | assert!( |
| 1491 | !text.contains("name: item"), |
| 1492 | "no name/args/result block for a call that did not happen: {text:?}" |
| 1493 | ); |
| 1494 | } |
| 1495 | } |
| 1496 | |
| 1497 | // --------------------------------------------------------------------------- |
| 1498 | // Severity — the ranks stay distinguishable |
| 1499 | // --------------------------------------------------------------------------- |
| 1500 | |
| 1501 | /// The deleted tests pinned each severity to a named palette constant, so a |
| 1502 | /// theme change broke four tests and a severity collapse broke none. What has |
| 1503 | /// to hold is the *relationship*: `Critical` reads exactly as loud as `Error`, |
| 1504 | /// `Warning` is visibly not an error, and `Info` is quieter than both — so a |
| 1505 | /// transient retry cannot be mistaken for a hard failure sitting next to it. |
| 1506 | #[test] |
| 1507 | fn error_severity_ranks_stay_visually_distinguishable() { |
| 1508 | use crate::error_taxonomy::ErrorSeverity; |
| 1509 | |
| 1510 | let rank = |severity| { |
| 1511 | let cell = HistoryCell::Error { |
| 1512 | message: "Authentication failed: invalid API key".to_string(), |
| 1513 | severity, |
| 1514 | }; |
| 1515 | let lines = cell.lines(80); |
| 1516 | assert!(!lines.is_empty(), "{severity:?} must render a line"); |
| 1517 | let label = &lines[0].spans[0]; |
| 1518 | (label.content.to_string(), label.style.fg) |
| 1519 | }; |
| 1520 | |
| 1521 | let (error_label, error_fg) = rank(ErrorSeverity::Error); |
| 1522 | let (critical_label, critical_fg) = rank(ErrorSeverity::Critical); |
| 1523 | let (warning_label, warning_fg) = rank(ErrorSeverity::Warning); |
| 1524 | let (info_label, info_fg) = rank(ErrorSeverity::Info); |
| 1525 | |
| 1526 | assert_eq!( |
| 1527 | (critical_label, critical_fg), |
| 1528 | (error_label.clone(), error_fg), |
| 1529 | "Critical and Error both flip offline mode; they must read identically" |
| 1530 | ); |
| 1531 | assert_ne!( |
| 1532 | warning_fg, error_fg, |
| 1533 | "a warning that reads as an error is the whole bug this guards" |
| 1534 | ); |
| 1535 | assert_ne!(warning_label, error_label, "and the labels must differ too"); |
| 1536 | assert_ne!(info_fg, error_fg, "info must not shout"); |
| 1537 | assert_ne!(info_fg, warning_fg, "info must not read as a warning"); |
| 1538 | assert_ne!(info_label, warning_label); |
| 1539 | |
| 1540 | // The body inherits the label's rank rather than staying neutral, or the |
| 1541 | // colour would carry no information past the first word. |
| 1542 | let cell = HistoryCell::Error { |
| 1543 | message: "Authentication failed: invalid API key".to_string(), |
| 1544 | severity: ErrorSeverity::Error, |
| 1545 | }; |
| 1546 | let body_fg = cell |
| 1547 | .lines(80) |
| 1548 | .iter() |
| 1549 | .flat_map(|line| line.spans.iter()) |
| 1550 | .find(|span| span.content.contains("Authentication")) |
| 1551 | .expect("error body span") |
| 1552 | .style |
| 1553 | .fg; |
| 1554 | assert_eq!(body_fg, error_fg); |
| 1555 | } |
| 1556 | |
| 1557 | #[test] |
| 1558 | fn error_guidance_keeps_recovery_commands_on_their_own_line() { |
| 1559 | let cell = HistoryCell::Error { |
| 1560 | message: "DeepSeek API key not found.\nSave it:\n codewhale auth set --provider deepseek" |
| 1561 | .to_string(), |
| 1562 | severity: crate::error_taxonomy::ErrorSeverity::Error, |
| 1563 | }; |
| 1564 | for width in [80, 140] { |
| 1565 | let lines = cell.lines(width); |
| 1566 | let command_line = lines |
| 1567 | .iter() |
| 1568 | .map(|line| { |
| 1569 | line.spans |
| 1570 | .iter() |
| 1571 | .map(|span| span.content.as_ref()) |
| 1572 | .collect::<String>() |
| 1573 | }) |
| 1574 | .find(|line| line.contains("codewhale auth set --provider deepseek")) |
| 1575 | .expect("recovery command stays intact"); |
| 1576 | assert!(!command_line.contains('\n'), "{command_line:?}"); |
| 1577 | assert!(!command_line.contains("Save it:"), "{command_line:?}"); |
| 1578 | } |
| 1579 | } |
| 1580 | |
| 1581 | /// A multiline failure can run past the bottom of the terminal while its full |
| 1582 | /// text stays in history. The live cell advertises the pager; the pager and the |
| 1583 | /// transcript must carry the recovery instruction verbatim and must not |
| 1584 | /// recursively advertise themselves. |
| 1585 | #[test] |
| 1586 | fn an_error_cell_advertises_the_pager_live_and_never_inside_it() { |
| 1587 | let recovery = "Refusing insecure base URL 'http://192.168.1.25:8000/v1'.\n\ |
| 1588 | Loopback hosts (localhost, 127.0.0.1, [::1]) are auto-allowed.\n\ |
| 1589 | Set CODEWHALE_ALLOW_INSECURE_HTTP=1 only for a trusted LAN host."; |
| 1590 | let cell = HistoryCell::Error { |
| 1591 | message: recovery.to_string(), |
| 1592 | severity: crate::error_taxonomy::ErrorSeverity::Error, |
| 1593 | }; |
| 1594 | |
| 1595 | let live_text = lines_text(&cell.lines(48)); |
| 1596 | let transcript_text = lines_text(&cell.transcript_lines(200)); |
| 1597 | let hint = crate::tui::key_shortcuts::tool_details_shortcut_action_hint("full error"); |
| 1598 | |
| 1599 | assert!(live_text.contains(&hint), "{live_text}"); |
| 1600 | assert!(!transcript_text.contains(&hint), "{transcript_text}"); |
| 1601 | assert!( |
| 1602 | transcript_text.contains("CODEWHALE_ALLOW_INSECURE_HTTP=1"), |
| 1603 | "the actionable instruction must survive verbatim: {transcript_text}" |
| 1604 | ); |
| 1605 | assert!( |
| 1606 | transcript_text.contains("192.168.1.25"), |
| 1607 | "the offending host must survive verbatim: {transcript_text}" |
| 1608 | ); |
| 1609 | } |
| 1610 | |
| 1611 | // --------------------------------------------------------------------------- |
| 1612 | // Cards with a content contract |
| 1613 | // --------------------------------------------------------------------------- |
| 1614 | |
| 1615 | /// A search receipt has to name where the answer came from and whether the |
| 1616 | /// provider it claimed was the provider it used. |
| 1617 | #[test] |
| 1618 | fn a_web_search_receipt_names_its_source_and_any_degradation() { |
| 1619 | let cell = WebSearchCell { |
| 1620 | query: "current release".to_string(), |
| 1621 | status: ToolStatus::Success, |
| 1622 | summary: Some("Found 2 results".to_string()), |
| 1623 | source: Some("provider-native/xai/grok-4.5".to_string()), |
| 1624 | degraded: Some("provider_native -> duckduckgo".to_string()), |
| 1625 | ref_count: 2, |
| 1626 | }; |
| 1627 | |
| 1628 | let rendered = lines_text(&cell.lines_with_motion(120, true)); |
| 1629 | |
| 1630 | for needle in [ |
| 1631 | "source", |
| 1632 | "provider-native/xai/grok-4.5", |
| 1633 | "degraded", |
| 1634 | "provider_native -> duckduckgo", |
| 1635 | "citations", |
| 1636 | ] { |
| 1637 | assert!(rendered.contains(needle), "missing {needle:?}: {rendered}"); |
| 1638 | } |
| 1639 | } |
| 1640 | |
| 1641 | /// A workflow's transcript is one row per run; live progress is the |
| 1642 | /// workbar's. A foreground `run` card that returned its settled record says |
| 1643 | /// only the finish — the final state replaces `started` — without repeating the header |
| 1644 | /// in a body; the expanded card adds the goal, the child labels, the final |
| 1645 | /// result and the error; the status card lists the runs it found. |
| 1646 | /// |
| 1647 | /// Replaces three tests, and drops assertions of the form |
| 1648 | /// `contains('s') || contains('m')` — true of essentially any English string. |
| 1649 | #[test] |
| 1650 | fn workflow_cards_report_lifecycle_children_phases_and_failures() { |
| 1651 | let run_output = serde_json::json!({ |
| 1652 | "run_id": "workflow_2400c600", |
| 1653 | "status": "completed", |
| 1654 | "workflow_goal": "audit the FLEET and WORKFLOW docs", |
| 1655 | "child_ids": ["a1", "a2", "a3"], |
| 1656 | "progress": ["phase: Scan", "log: 3 findings"], |
| 1657 | "events": [ |
| 1658 | {"type": "task_started", "task_id": "a1", "label": "scan-docs", |
| 1659 | "workflow_run_id": "workflow_2400c600", "workflow_phase_id": "Scan", |
| 1660 | "workflow_task_label": "scan-docs", "workflow_child_index": 0}, |
| 1661 | {"type": "task_started", "task_id": "a2", "workflow_task_label": "check-fleet", |
| 1662 | "workflow_run_id": "workflow_2400c600", "workflow_child_index": 1}, |
| 1663 | {"type": "task_started", "task_id": "a3", "label": "summarize", |
| 1664 | "workflow_run_id": "workflow_2400c600", "workflow_child_index": 2}, |
| 1665 | ], |
| 1666 | "schema_errors": [], |
| 1667 | }) |
| 1668 | .to_string(); |
| 1669 | let mut run = generic_tool("workflow", ToolStatus::Success); |
| 1670 | run.input_summary = Some("action: run".to_string()); |
| 1671 | run.output = Some(run_output); |
| 1672 | let text = lines_text(&run.lines_with_mode(120, true, RenderMode::Live)); |
| 1673 | assert!( |
| 1674 | !text.contains("started"), |
| 1675 | "the settled record replaces the start line: {text:?}" |
| 1676 | ); |
| 1677 | assert!( |
| 1678 | text.contains("finished") && text.contains("/3 done"), |
| 1679 | "the finish line with done/total agents: {text:?}" |
| 1680 | ); |
| 1681 | assert_eq!(text.lines().count(), 1, "one row for the run: {text:?}"); |
| 1682 | // #6503: a run with no failures does not announce `0 fail`. |
| 1683 | assert!(!text.contains("fail"), "no zero failure count: {text:?}"); |
| 1684 | assert!( |
| 1685 | !text.contains("status:"), |
| 1686 | "the body must not repeat the header lifecycle: {text:?}" |
| 1687 | ); |
| 1688 | |
| 1689 | let failed_output = serde_json::json!({ |
| 1690 | "run_id": "workflow_exp", |
| 1691 | "status": "failed", |
| 1692 | "workflow_goal": "ship v0.8.68", |
| 1693 | "started_at_ms": 1000, |
| 1694 | "completed_at_ms": 5000, |
| 1695 | "source_path": "workflows/demo.workflow.js", |
| 1696 | "error": "phase Verify failed", |
| 1697 | "result": {"summary": "2 of 3 children ok"}, |
| 1698 | "events": [ |
| 1699 | {"type": "run_started", "at_ms": 1000, "run_id": "workflow_exp", |
| 1700 | "workflow_goal": "ship v0.8.68"}, |
| 1701 | {"type": "phase_started", "at_ms": 1100, "title": "Verify"}, |
| 1702 | {"type": "task_started", "at_ms": 1200, "task_id": "t1", "label": "run tests", |
| 1703 | "workflow_task_label": "run tests", "profile": "implementer"}, |
| 1704 | {"type": "task_completed", "at_ms": 4000, "task_id": "t1", "status": "failed"}, |
| 1705 | {"type": "run_completed", "at_ms": 5000, "status": "failed", |
| 1706 | "error": "phase Verify failed"} |
| 1707 | ] |
| 1708 | }) |
| 1709 | .to_string(); |
| 1710 | let mut failed = generic_tool("workflow", ToolStatus::Failed); |
| 1711 | failed.input_summary = Some("action: run".to_string()); |
| 1712 | failed.output = Some(failed_output); |
| 1713 | failed.spillover_path = Some(PathBuf::from("/tmp/wf-artifact.json")); |
| 1714 | let text = lines_text(&failed.lines_with_mode(140, true, RenderMode::Transcript)); |
| 1715 | for needle in [ |
| 1716 | "ship v0.8.68", |
| 1717 | "Verify", |
| 1718 | "run tests", |
| 1719 | "2 of 3", |
| 1720 | "phase Verify failed", |
| 1721 | ] { |
| 1722 | assert!( |
| 1723 | text.contains(needle), |
| 1724 | "the expanded card must carry {needle:?}: {text}" |
| 1725 | ); |
| 1726 | } |
| 1727 | |
| 1728 | let status_output = serde_json::json!({ |
| 1729 | "action": "status", |
| 1730 | "count": 2, |
| 1731 | "runs": [ |
| 1732 | {"run_id": "workflow_aaa", "status": "running", "child_count": 4}, |
| 1733 | {"run_id": "workflow_bbb", "status": "completed", "child_count": 1}, |
| 1734 | ], |
| 1735 | }) |
| 1736 | .to_string(); |
| 1737 | let mut status = generic_tool("workflow", ToolStatus::Success); |
| 1738 | status.input_summary = Some("action: status".to_string()); |
| 1739 | status.output = Some(status_output); |
| 1740 | let text = lines_text(&status.lines_with_mode(120, true, RenderMode::Live)); |
| 1741 | for needle in ["2 run(s)", "workflow_aaa", "running", "workflow_bbb"] { |
| 1742 | assert!( |
| 1743 | text.contains(needle), |
| 1744 | "the status card must list {needle:?}: {text:?}" |
| 1745 | ); |
| 1746 | } |
| 1747 | } |
| 1748 | |
| 1749 | /// The founder's transcript (2026-09-28): a refused `start` echoed |
| 1750 | /// `action: start` and hid its reason; the settled run's reason was cut |
| 1751 | /// mid-word. Each is one row that says why. |
| 1752 | #[test] |
| 1753 | fn workflow_rows_say_why_without_raw_action_or_mid_word_cuts() { |
| 1754 | let mut refused = generic_tool("workflow", ToolStatus::Failed); |
| 1755 | refused.input_summary = Some("action: start".to_string()); |
| 1756 | refused.output = Some( |
| 1757 | "Error: Invalid input for tool 'workflow': Workflow leaf 'engine-readiness': \ |
| 1758 | task(): cwd entries must be bounded repo-relative paths\n\ |
| 1759 | Tool validation feedback: {\"category\":\"invalid_input\"}" |
| 1760 | .to_string(), |
| 1761 | ); |
| 1762 | let text = lines_text(&refused.lines_with_mode(100, true, RenderMode::Live)); |
| 1763 | assert!(!text.contains("action: start"), "{text}"); |
| 1764 | assert!( |
| 1765 | text.contains("cwd entries must be bounded repo-relative paths"), |
| 1766 | "{text}" |
| 1767 | ); |
| 1768 | assert!(!text.contains("Invalid input for tool"), "{text}"); |
| 1769 | |
| 1770 | let reason = "[auth] Authorization failed: You have run out of credits or need a Grok \ |
| 1771 | subscription. Add credits at https://grok.com/?_s=usage."; |
| 1772 | let failed = serde_json::json!({ |
| 1773 | "run_id": "workflow_6409ebe6", |
| 1774 | "status": "failed", |
| 1775 | "workflow_goal": "Read-only release-readiness audit for Codewhale v0.10.1. Determine blockers.", |
| 1776 | "started_at_ms": 1_000, |
| 1777 | "completed_at_ms": 1_355, |
| 1778 | "error": "no task produced a result: all 2 task(s) failed and 1 fan-out(s) lost every slot (no work survived them); the recorded result reflects no completed work", |
| 1779 | "transcript_line": "finished", |
| 1780 | "events": [ |
| 1781 | {"type": "run_started", "at_ms": 1_000, "workflow_goal": "Read-only release-readiness audit for Codewhale v0.10.1. Determine blockers."}, |
| 1782 | {"type": "task_started", "at_ms": 1_080, "task_id": "a", "workflow_task_label": "engine-readiness"}, |
| 1783 | {"type": "task_started", "at_ms": 1_117, "task_id": "b", "workflow_task_label": "desktop-readiness"}, |
| 1784 | {"type": "task_completed", "at_ms": 1_329, "task_id": "a", "status": "failed", "reason": reason}, |
| 1785 | {"type": "task_completed", "at_ms": 1_338, "task_id": "b", "status": "failed", "reason": reason}, |
| 1786 | {"type": "run_completed", "at_ms": 1_355, "status": "failed"}, |
| 1787 | ], |
| 1788 | }) |
| 1789 | .to_string(); |
| 1790 | let mut finish = generic_tool("workflow", ToolStatus::Failed); |
| 1791 | finish.output = Some(failed); |
| 1792 | let text = lines_text(&finish.lines_with_mode(60, true, RenderMode::Live)); |
| 1793 | assert!(!text.contains("started"), "{text}"); |
| 1794 | assert!(text.contains("0/2 done · 2 failed"), "{text}"); |
| 1795 | assert!(text.contains("355ms"), "{text}"); |
| 1796 | // The whole first sentence, wrapped, never cut. |
| 1797 | let flat = text |
| 1798 | .split_whitespace() |
| 1799 | .filter(|word| { |
| 1800 | !word |
| 1801 | .chars() |
| 1802 | .all(|ch| ('\u{2500}'..='\u{259F}').contains(&ch)) |
| 1803 | }) |
| 1804 | .collect::<Vec<_>>() |
| 1805 | .join(" "); |
| 1806 | assert!( |
| 1807 | flat.contains( |
| 1808 | "Authorization failed: You have run out of credits or need a Grok subscription" |
| 1809 | ), |
| 1810 | "{text}" |
| 1811 | ); |
| 1812 | assert!( |
| 1813 | !flat.contains("grok.com"), |
| 1814 | "only the first sentence: {text}" |
| 1815 | ); |
| 1816 | assert!(!text.contains("..."), "{text}"); |
| 1817 | } |
| 1818 | |
| 1819 | #[test] |
| 1820 | fn degraded_workflow_receipt_is_terminal_warning_not_running_or_success() { |
| 1821 | let output = serde_json::json!({ |
| 1822 | "run_id": "workflow_partial", |
| 1823 | "status": "degraded", |
| 1824 | "workflow_goal": "review the release", |
| 1825 | "started_at_ms": 1_000, |
| 1826 | "completed_at_ms": 2_000, |
| 1827 | "dispatch_failure_count": 1, |
| 1828 | "dispatch_failures": [{ |
| 1829 | "label": "review docs", |
| 1830 | "message": "profile unavailable", |
| 1831 | "at_ms": 1_500, |
| 1832 | }], |
| 1833 | }) |
| 1834 | .to_string(); |
| 1835 | let mut run = generic_tool("workflow", ToolStatus::Success); |
| 1836 | run.output = Some(output); |
| 1837 | |
| 1838 | let lines = run.lines_with_mode(120, false, RenderMode::Live); |
| 1839 | let text = lines_text(&lines); |
| 1840 | assert!( |
| 1841 | text.contains("finished with gaps"), |
| 1842 | "warning receipt missing: {text:?}" |
| 1843 | ); |
| 1844 | assert!( |
| 1845 | !text.to_lowercase().contains(" done"), |
| 1846 | "must not read as success: {text:?}" |
| 1847 | ); |
| 1848 | assert!( |
| 1849 | !text.contains(" running"), |
| 1850 | "must not read as live: {text:?}" |
| 1851 | ); |
| 1852 | |
| 1853 | let warning = lines |
| 1854 | .iter() |
| 1855 | .flat_map(|line| line.spans.iter()) |
| 1856 | .find(|span| span.content.as_ref() == "finished with gaps") |
| 1857 | .expect("terminal warning status span"); |
| 1858 | assert_eq!( |
| 1859 | warning.style.fg, |
| 1860 | Some(super::tool_rail_color(ToolStatus::Warning)), |
| 1861 | "degraded receipt must use the terminal warning accent" |
| 1862 | ); |
| 1863 | assert!( |
| 1864 | lines |
| 1865 | .iter() |
| 1866 | .flat_map(|line| line.spans.iter()) |
| 1867 | .all(|span| !span |
| 1868 | .content |
| 1869 | .chars() |
| 1870 | .any(|ch| ('\u{2800}'..='\u{28ff}').contains(&ch))), |
| 1871 | "terminal receipt must not retain a spinner: {text:?}" |
| 1872 | ); |
| 1873 | } |
| 1874 | |
| 1875 | /// A checklist update names one item. Showing the rest would make every |
| 1876 | /// single-item edit cost the height of the whole list; showing none would make |
| 1877 | /// the row unreadable. An id past the end of the list falls back to a |
| 1878 | /// placeholder instead of panicking. |
| 1879 | #[test] |
| 1880 | fn a_checklist_update_shows_only_the_item_that_changed() { |
| 1881 | let snapshot = super::ChecklistSnapshot { |
| 1882 | items: vec![ |
| 1883 | super::ChecklistItemSnapshot { |
| 1884 | content: "Read the spec".to_string(), |
| 1885 | status: "completed".to_string(), |
| 1886 | }, |
| 1887 | super::ChecklistItemSnapshot { |
| 1888 | content: "Write the test".to_string(), |
| 1889 | status: "in_progress".to_string(), |
| 1890 | }, |
| 1891 | super::ChecklistItemSnapshot { |
| 1892 | content: "Land the PR".to_string(), |
| 1893 | status: "pending".to_string(), |
| 1894 | }, |
| 1895 | ], |
| 1896 | completion_pct: 33, |
| 1897 | completed: 1, |
| 1898 | total: 3, |
| 1899 | }; |
| 1900 | let lines = super::render_checklist_change_card( |
| 1901 | "todo_update", |
| 1902 | ToolStatus::Success, |
| 1903 | &snapshot, |
| 1904 | &super::ChecklistChange { |
| 1905 | id: 2, |
| 1906 | status: "in_progress".to_string(), |
| 1907 | }, |
| 1908 | 80, |
| 1909 | true, |
| 1910 | ); |
| 1911 | assert!(lines.len() >= 3, "header, change, summary: {}", lines.len()); |
| 1912 | |
| 1913 | let change = line_text(&lines[1]); |
| 1914 | for needle in ["#2", "Write the test", "in_progress"] { |
| 1915 | assert!(change.contains(needle), "missing {needle:?}: {change:?}"); |
| 1916 | } |
| 1917 | for other in ["Land the PR", "Read the spec"] { |
| 1918 | assert!( |
| 1919 | !change.contains(other), |
| 1920 | "an update must not redraw the whole list: {change:?}" |
| 1921 | ); |
| 1922 | } |
| 1923 | |
| 1924 | let summary = line_text(lines.last().expect("summary row")); |
| 1925 | assert!(summary.contains("3 items"), "{summary:?}"); |
| 1926 | assert!( |
| 1927 | summary.contains(&crate::tui::key_shortcuts::tool_details_shortcut_action_hint("list")), |
| 1928 | "the full list stays one keypress away: {summary:?}" |
| 1929 | ); |
| 1930 | |
| 1931 | let single = super::ChecklistSnapshot { |
| 1932 | items: vec![super::ChecklistItemSnapshot { |
| 1933 | content: "only item".to_string(), |
| 1934 | status: "pending".to_string(), |
| 1935 | }], |
| 1936 | completion_pct: 0, |
| 1937 | completed: 0, |
| 1938 | total: 1, |
| 1939 | }; |
| 1940 | let lines = super::render_checklist_change_card( |
| 1941 | "todo_update", |
| 1942 | ToolStatus::Success, |
| 1943 | &single, |
| 1944 | &super::ChecklistChange { |
| 1945 | id: 99, |
| 1946 | status: "completed".to_string(), |
| 1947 | }, |
| 1948 | 80, |
| 1949 | true, |
| 1950 | ); |
| 1951 | let change = line_text(&lines[1]); |
| 1952 | assert!(change.contains("#99") && change.contains("(missing title)")); |
| 1953 | } |
| 1954 | |
| 1955 | /// The plan card is the only place a plan's supporting artifact is visible. |
| 1956 | /// Every populated section has to reach the surface, or the model can record |
| 1957 | /// context the user never sees. |
| 1958 | #[test] |
| 1959 | fn a_plan_card_surfaces_every_populated_artifact_section() { |
| 1960 | let cell = PlanUpdateCell { |
| 1961 | snapshot: PlanSnapshot { |
| 1962 | objective: Some("Make Plan mode reviewable".to_string()), |
| 1963 | context_summary: Some("Grounded in issue #2691".to_string()), |
| 1964 | sources_used: vec!["gh issue view 2691".to_string()], |
| 1965 | critical_files: vec!["crates/tui/src/tools/plan.rs".to_string()], |
| 1966 | constraints: vec!["Keep To-do primary".to_string()], |
| 1967 | recommended_approach: Some( |
| 1968 | "Enrich update_plan without breaking legacy calls".to_string(), |
| 1969 | ), |
| 1970 | verification_plan: Some("Run focused renderer tests".to_string()), |
| 1971 | risks_and_unknowns: Some("Metadata-only plans can disappear".to_string()), |
| 1972 | handoff_packet: Some("Next agent should inspect relay output".to_string()), |
| 1973 | items: vec![crate::tools::plan::PlanItemArg { |
| 1974 | step: "Render artifact sections".to_string(), |
| 1975 | status: StepStatus::InProgress, |
| 1976 | }], |
| 1977 | ..PlanSnapshot::default() |
| 1978 | }, |
| 1979 | status: ToolStatus::Success, |
| 1980 | }; |
| 1981 | |
| 1982 | let visible = lines_text(&cell.lines_with_motion(120, true)); |
| 1983 | |
| 1984 | for needle in [ |
| 1985 | "objective:", |
| 1986 | "Make Plan mode reviewable", |
| 1987 | "source:", |
| 1988 | "gh issue view 2691", |
| 1989 | "file:", |
| 1990 | "verify:", |
| 1991 | "handoff:", |
| 1992 | "Render artifact sections", |
| 1993 | ] { |
| 1994 | assert!(visible.contains(needle), "missing {needle:?}: {visible}"); |
| 1995 | } |
| 1996 | } |
| 1997 | |
| 1998 | /// A fan-out tool's per-child prompts get one row each so the user can read |
| 1999 | /// what each child was asked; the inline `args:` summary that would otherwise |
| 2000 | /// say `prompts: <3 items>` is suppressed rather than printed alongside them. |
| 2001 | #[test] |
| 2002 | fn fan_out_prompts_replace_the_inline_argument_summary() { |
| 2003 | let mut cell = generic_tool("read_file", ToolStatus::Running); |
| 2004 | cell.input_summary = Some("prompts: <3 items>".to_string()); |
| 2005 | cell.prompts = Some(vec![ |
| 2006 | "Summarize the README".to_string(), |
| 2007 | "List the public types in client.rs".to_string(), |
| 2008 | "Diff this commit against main".to_string(), |
| 2009 | ]); |
| 2010 | let text = lines_text(&HistoryCell::Tool(ToolCell::Generic(cell)).lines(80)); |
| 2011 | |
| 2012 | assert!(text.contains("[0] Summarize the README")); |
| 2013 | assert!(text.contains("[1] List the public types in client.rs")); |
| 2014 | assert!(text.contains("[2] Diff this commit against main")); |
| 2015 | assert!( |
| 2016 | !text.contains("args: prompts:"), |
| 2017 | "the summary the rows replaced must not also render: {text}" |
| 2018 | ); |
| 2019 | |
| 2020 | let mut plain = generic_tool("file_search", ToolStatus::Running); |
| 2021 | plain.input_summary = Some("query: foo".to_string()); |
| 2022 | let text = lines_text(&HistoryCell::Tool(ToolCell::Generic(plain)).lines(80)); |
| 2023 | assert!( |
| 2024 | text.contains("query: foo"), |
| 2025 | "a non-fan-out tool keeps its argument summary: {text}" |
| 2026 | ); |
| 2027 | } |
| 2028 | |
| 2029 | /// A grouped activity row is metadata, not a tool card: exactly one line, and |
| 2030 | /// the synthetic tool name that carries it never reaches the screen. |
| 2031 | #[test] |
| 2032 | fn an_activity_group_renders_as_a_single_metadata_line() { |
| 2033 | let mut cell = generic_tool("activity_group", ToolStatus::Success); |
| 2034 | cell.input_summary = Some("Explored 2 files, 1 search".to_string()); |
| 2035 | |
| 2036 | let lines = cell.lines_with_mode(120, true, RenderMode::Live); |
| 2037 | |
| 2038 | assert_eq!(lines.len(), 1); |
| 2039 | assert_eq!(lines_text(&lines), "Explored 2 files, 1 search ›"); |
| 2040 | assert!(!lines_text(&lines).contains("activity_group")); |
| 2041 | } |
| 2042 | |
| 2043 | // --------------------------------------------------------------------------- |
| 2044 | // Replay — wire messages project to the right typed cell |
| 2045 | // --------------------------------------------------------------------------- |
| 2046 | |
| 2047 | /// The wire carries a `(reasoning omitted)` placeholder for turns whose |
| 2048 | /// reasoning the provider did not return. Replaying it as a reasoning cell |
| 2049 | /// would put words in the model's mouth. |
| 2050 | #[test] |
| 2051 | fn restored_history_drops_the_wire_reasoning_placeholder() { |
| 2052 | let message = Message { |
| 2053 | role: Role::Assistant, |
| 2054 | content: vec![ |
| 2055 | ContentBlock::Thinking { |
| 2056 | thinking: "(reasoning omitted)".to_string(), |
| 2057 | signature: None, |
| 2058 | state: None, |
| 2059 | }, |
| 2060 | ContentBlock::Thinking { |
| 2061 | thinking: "Actual model reasoning".to_string(), |
| 2062 | signature: None, |
| 2063 | state: None, |
| 2064 | }, |
| 2065 | ], |
| 2066 | }; |
| 2067 | |
| 2068 | let cells = super::history_cells_from_message(&message); |
| 2069 | assert_eq!(cells.len(), 1); |
| 2070 | assert!(matches!( |
| 2071 | &cells[0], |
| 2072 | HistoryCell::Thinking { content, .. } if content == "Actual model reasoning" |
| 2073 | )); |
| 2074 | } |
| 2075 | |
| 2076 | /// Compaction writes an `<archived_context>` envelope whose attributes are the |
| 2077 | /// only record of what was dropped. Attribute parsing must survive spaces and |
| 2078 | /// punctuation inside the values, or the summary reads with a mangled range. |
| 2079 | #[test] |
| 2080 | fn archived_context_metadata_survives_spaces_inside_attribute_values() { |
| 2081 | let msg = Message { |
| 2082 | role: Role::Assistant, |
| 2083 | content: vec![ContentBlock::Text { |
| 2084 | text: "<archived_context level=\"1\" range=\"msg 0-128\" tokens=\"2499\" \ |
| 2085 | density=\"~2,500 tokens\" model=\"deepseek-v4-flash\" \ |
| 2086 | timestamp=\"2026-04-28T00:00:00Z\">\nSummary body\n</archived_context>" |
| 2087 | .to_string(), |
| 2088 | cache_control: None, |
| 2089 | }], |
| 2090 | }; |
| 2091 | |
| 2092 | let cells = super::history_cells_from_message(&msg); |
| 2093 | assert_eq!(cells.len(), 1); |
| 2094 | let HistoryCell::ArchivedContext { |
| 2095 | level, |
| 2096 | range, |
| 2097 | tokens, |
| 2098 | density, |
| 2099 | model, |
| 2100 | timestamp, |
| 2101 | summary, |
| 2102 | } = &cells[0] |
| 2103 | else { |
| 2104 | panic!("expected archived context cell, got {:?}", cells[0]); |
| 2105 | }; |
| 2106 | |
| 2107 | assert_eq!(*level, 1); |
| 2108 | assert_eq!(range, "msg 0-128"); |
| 2109 | assert_eq!(tokens, "2499"); |
| 2110 | assert_eq!(density, "~2,500 tokens"); |
| 2111 | assert_eq!(model, "deepseek-v4-flash"); |
| 2112 | assert_eq!(timestamp, "2026-04-28T00:00:00Z"); |
| 2113 | assert_eq!(summary, "Summary body"); |
| 2114 | } |
| 2115 | |
| 2116 | /// Two projections that must not become generic assistant prose: a repair |
| 2117 | /// receipt is a system note, and a replayed `update_plan` call rebuilds the |
| 2118 | /// typed plan cell with its snapshot intact. |
| 2119 | #[test] |
| 2120 | fn replay_routes_repair_receipts_and_plan_calls_to_typed_cells() { |
| 2121 | let repair = Message { |
| 2122 | role: Role::Assistant, |
| 2123 | content: vec![ContentBlock::Text { |
| 2124 | text: "[tool_history_repair] Repaired 1 crashed tool call(s); quarantined 0 \ |
| 2125 | duplicate and 0 orphan terminal result(s)." |
| 2126 | .to_string(), |
| 2127 | cache_control: None, |
| 2128 | }], |
| 2129 | }; |
| 2130 | assert!(matches!( |
| 2131 | super::history_cells_from_message(&repair).as_slice(), |
| 2132 | [HistoryCell::System { content }] if content.starts_with("[tool_history_repair]") |
| 2133 | )); |
| 2134 | |
| 2135 | let plan = Message { |
| 2136 | role: Role::Assistant, |
| 2137 | content: vec![ContentBlock::ToolUse { |
| 2138 | execution_id: None, |
| 2139 | id: "plan-1".to_string(), |
| 2140 | name: "update_plan".to_string(), |
| 2141 | input: serde_json::json!({ |
| 2142 | "objective": "Make Plan mode reviewable", |
| 2143 | "sources_used": ["gh issue view 2691"], |
| 2144 | "critical_files": ["crates/tui/src/tools/plan.rs"], |
| 2145 | "plan": [ |
| 2146 | { "step": "render replay card", "status": "completed" } |
| 2147 | ] |
| 2148 | }), |
| 2149 | caller: None, |
| 2150 | thought_signature: None, |
| 2151 | }], |
| 2152 | }; |
| 2153 | let cells = super::history_cells_from_message(&plan); |
| 2154 | assert_eq!(cells.len(), 1); |
| 2155 | let HistoryCell::Tool(ToolCell::PlanUpdate(cell)) = &cells[0] else { |
| 2156 | panic!("expected update_plan replay cell"); |
| 2157 | }; |
| 2158 | assert_eq!(cell.status, ToolStatus::Success); |
| 2159 | assert_eq!( |
| 2160 | cell.snapshot.objective.as_deref(), |
| 2161 | Some("Make Plan mode reviewable") |
| 2162 | ); |
| 2163 | assert_eq!(cell.snapshot.sources_used, vec!["gh issue view 2691"]); |
| 2164 | assert_eq!(cell.snapshot.items[0].status, StepStatus::Completed); |
| 2165 | } |
| 2166 | |
| 2167 | /// The runtime appends a `<turn_meta>` block to the user's message. It is |
| 2168 | /// scaffolding and must be hidden — but only when it is the trailing block the |
| 2169 | /// runtime appended. A user who types the same tag is quoting, not injecting, |
| 2170 | /// and their text must survive verbatim. |
| 2171 | #[test] |
| 2172 | fn user_history_hides_only_the_trailing_turn_metadata_block() { |
| 2173 | let visible = "Explain this literal: <turn_meta>example</turn_meta>"; |
| 2174 | let turn_meta = concat!( |
| 2175 | "<turn_meta>\n", |
| 2176 | "Current local date: 2026-07-22\n", |
| 2177 | "Input provenance: external_user\n", |
| 2178 | "Input authority: external_current_turn\n", |
| 2179 | "</turn_meta>", |
| 2180 | ); |
| 2181 | let msg = Message { |
| 2182 | role: Role::User, |
| 2183 | content: vec![ |
| 2184 | ContentBlock::Text { |
| 2185 | text: visible.to_string(), |
| 2186 | cache_control: None, |
| 2187 | }, |
| 2188 | ContentBlock::Text { |
| 2189 | text: turn_meta.to_string(), |
| 2190 | cache_control: None, |
| 2191 | }, |
| 2192 | ], |
| 2193 | }; |
| 2194 | assert!(matches!( |
| 2195 | super::history_cells_from_message(&msg).as_slice(), |
| 2196 | [HistoryCell::User { content }] if content == visible |
| 2197 | )); |
| 2198 | |
| 2199 | let literal_only = Message { |
| 2200 | role: Role::User, |
| 2201 | content: vec![ContentBlock::Text { |
| 2202 | text: "<turn_meta>user-authored example</turn_meta>".to_string(), |
| 2203 | cache_control: None, |
| 2204 | }], |
| 2205 | }; |
| 2206 | assert!(matches!( |
| 2207 | super::history_cells_from_message(&literal_only).as_slice(), |
| 2208 | [HistoryCell::User { content }] |
| 2209 | if content == "<turn_meta>user-authored example</turn_meta>" |
| 2210 | )); |
| 2211 | } |
| 2212 | |
| 2213 | /// "Copy answer" must select the last completed assistant cell and serialize |
| 2214 | /// exactly its authored text — no reasoning, no tool bodies, no runtime status, |
| 2215 | /// no role marker, and never a half-streamed cell. |
| 2216 | #[test] |
| 2217 | fn the_answer_projection_copies_authored_text_and_nothing_else() { |
| 2218 | use crate::tui::ui_text::history_cell_to_clipboard_text; |
| 2219 | |
| 2220 | let cells = [ |
| 2221 | HistoryCell::User { |
| 2222 | content: "please summarize".to_string(), |
| 2223 | }, |
| 2224 | HistoryCell::Thinking { |
| 2225 | content: "private reasoning trace".to_string(), |
| 2226 | streaming: false, |
| 2227 | duration_secs: Some(1.0), |
| 2228 | }, |
| 2229 | HistoryCell::Tool(ToolCell::Generic({ |
| 2230 | let mut cell = generic_tool("read_file", ToolStatus::Success); |
| 2231 | cell.input_summary = Some("src/lib.rs".to_string()); |
| 2232 | cell.output = Some("raw tool result body".to_string()); |
| 2233 | cell |
| 2234 | })), |
| 2235 | HistoryCell::System { |
| 2236 | content: "runtime status note".to_string(), |
| 2237 | }, |
| 2238 | HistoryCell::Assistant { |
| 2239 | content: "still streaming partial".to_string(), |
| 2240 | streaming: true, |
| 2241 | }, |
| 2242 | HistoryCell::Assistant { |
| 2243 | content: "## Final answer\nauthored markdown".to_string(), |
| 2244 | streaming: false, |
| 2245 | }, |
| 2246 | ]; |
| 2247 | |
| 2248 | let answer = cells |
| 2249 | .iter() |
| 2250 | .rev() |
| 2251 | .find(|cell| cell.is_completed_assistant_answer()) |
| 2252 | .expect("the completed assistant cell must qualify"); |
| 2253 | let copied = history_cell_to_clipboard_text(answer, 80); |
| 2254 | |
| 2255 | assert_eq!(copied, "## Final answer\nauthored markdown"); |
| 2256 | for excluded in [ |
| 2257 | "please summarize", |
| 2258 | "private reasoning trace", |
| 2259 | "raw tool result body", |
| 2260 | "runtime status note", |
| 2261 | "still streaming partial", |
| 2262 | ASSISTANT_GLYPH, |
| 2263 | ] { |
| 2264 | assert!( |
| 2265 | !copied.contains(excluded), |
| 2266 | "answer copy leaked {excluded:?}" |
| 2267 | ); |
| 2268 | } |
| 2269 | } |
| 2270 | |
| 2271 | // --------------------------------------------------------------------------- |
| 2272 | // Grouping and small parsers |
| 2273 | // --------------------------------------------------------------------------- |
| 2274 | |
| 2275 | fn tool_cell(name: &str, status: ToolStatus) -> HistoryCell { |
| 2276 | let mut cell = generic_tool(name, status); |
| 2277 | cell.input_summary = Some(format!("args for {name}")); |
| 2278 | cell.output = Some(format!("output for {name}")); |
| 2279 | HistoryCell::Tool(ToolCell::Generic(cell)) |
| 2280 | } |
| 2281 | |
| 2282 | /// Collapsing a run of tool cards is only safe when nothing in the run needs |
| 2283 | /// the user's eyes: a failure, an in-flight call, or a shell command must all |
| 2284 | /// break the group and stay individually visible, and a run shorter than the |
| 2285 | /// threshold is not a run at all. |
| 2286 | /// |
| 2287 | /// Replaces three tests. |
| 2288 | #[test] |
| 2289 | fn only_contiguous_finished_safe_tool_calls_collapse_into_a_run() { |
| 2290 | let history = vec![ |
| 2291 | HistoryCell::User { |
| 2292 | content: "go".to_string(), |
| 2293 | }, |
| 2294 | tool_cell("read_file", ToolStatus::Success), |
| 2295 | tool_cell("list_dir", ToolStatus::Success), |
| 2296 | tool_cell("web_search", ToolStatus::Success), |
| 2297 | HistoryCell::Assistant { |
| 2298 | content: "done".to_string(), |
| 2299 | streaming: false, |
| 2300 | }, |
| 2301 | ]; |
| 2302 | let runs = super::detect_tool_runs(&history, 3); |
| 2303 | assert_eq!(runs.len(), 1); |
| 2304 | assert_eq!((runs[0].start, runs[0].count), (1, 3)); |
| 2305 | assert_eq!( |
| 2306 | runs[0].tool_families, |
| 2307 | vec!["read_file", "list_dir", "web_search"] |
| 2308 | ); |
| 2309 | assert_eq!(runs[0].activity.files, 2); |
| 2310 | assert_eq!(runs[0].activity.searches, 1); |
| 2311 | |
| 2312 | assert!( |
| 2313 | super::detect_tool_runs( |
| 2314 | &[ |
| 2315 | tool_cell("read_file", ToolStatus::Success), |
| 2316 | tool_cell("list_dir", ToolStatus::Success), |
| 2317 | ], |
| 2318 | 3 |
| 2319 | ) |
| 2320 | .is_empty(), |
| 2321 | "a run below the threshold is not collapsed" |
| 2322 | ); |
| 2323 | |
| 2324 | assert!( |
| 2325 | super::detect_tool_runs( |
| 2326 | &[ |
| 2327 | tool_cell("read_file", ToolStatus::Success), |
| 2328 | HistoryCell::Assistant { |
| 2329 | content: "pause".to_string(), |
| 2330 | streaming: false, |
| 2331 | }, |
| 2332 | tool_cell("list_dir", ToolStatus::Success), |
| 2333 | tool_cell("web_search", ToolStatus::Success), |
| 2334 | ], |
| 2335 | 3 |
| 2336 | ) |
| 2337 | .is_empty(), |
| 2338 | "assistant prose breaks the run" |
| 2339 | ); |
| 2340 | |
| 2341 | // Each of failure, in-flight and shell breaks the group; only the clean |
| 2342 | // trailing triple survives. |
| 2343 | let mut mixed = Vec::new(); |
| 2344 | for breaker in [ |
| 2345 | tool_cell("web_search", ToolStatus::Failed), |
| 2346 | tool_cell("web_search", ToolStatus::Running), |
| 2347 | HistoryCell::Tool(ToolCell::Exec({ |
| 2348 | let mut exec = exec_tool("rm -rf target", ToolStatus::Success); |
| 2349 | exec.output = Some("ok".to_string()); |
| 2350 | exec |
| 2351 | })), |
| 2352 | ] { |
| 2353 | mixed.push(tool_cell("read_file", ToolStatus::Success)); |
| 2354 | mixed.push(tool_cell("list_dir", ToolStatus::Success)); |
| 2355 | mixed.push(breaker); |
| 2356 | } |
| 2357 | let tail_start = mixed.len(); |
| 2358 | mixed.push(tool_cell("read_file", ToolStatus::Success)); |
| 2359 | mixed.push(tool_cell("list_dir", ToolStatus::Success)); |
| 2360 | mixed.push(tool_cell("web_search", ToolStatus::Success)); |
| 2361 | |
| 2362 | let runs = super::detect_tool_runs(&mixed, 3); |
| 2363 | assert_eq!(runs.len(), 1, "only the clean tail collapses: {runs:?}"); |
| 2364 | assert_eq!((runs[0].start, runs[0].count), (tail_start, 3)); |
| 2365 | } |
| 2366 | |
| 2367 | /// The one-line summary that replaces a collapsed run is the user's only |
| 2368 | /// record of it, so it must name what actually happened — the right verb, the |
| 2369 | /// right counts, and only the tool families that belong to each clause. |
| 2370 | /// |
| 2371 | /// Replaces four tests. |
| 2372 | #[test] |
| 2373 | fn a_collapsed_run_summary_names_what_actually_happened() { |
| 2374 | let run = |families: &[&str], activity: super::ToolRunActivitySummary| super::ToolRun { |
| 2375 | start: 4, |
| 2376 | count: families.len(), |
| 2377 | tool_families: families.iter().map(|f| f.to_string()).collect(), |
| 2378 | activity, |
| 2379 | }; |
| 2380 | |
| 2381 | assert_eq!( |
| 2382 | super::tool_run_summary(&run( |
| 2383 | &["read_file", "list_dir"], |
| 2384 | super::ToolRunActivitySummary { |
| 2385 | files: 4, |
| 2386 | searches: 1, |
| 2387 | ..Default::default() |
| 2388 | } |
| 2389 | )), |
| 2390 | "Explored 4 files, 1 search: read_file, list_dir" |
| 2391 | ); |
| 2392 | |
| 2393 | assert_eq!( |
| 2394 | super::tool_run_summary(&run( |
| 2395 | &["read_file", "run_tests", "validate_data"], |
| 2396 | super::ToolRunActivitySummary { |
| 2397 | files: 2, |
| 2398 | commands: 2, |
| 2399 | ..Default::default() |
| 2400 | } |
| 2401 | )), |
| 2402 | "Explored 2 files: read_file, ran 2 commands: run_tests, validate_data", |
| 2403 | "each clause lists only its own families" |
| 2404 | ); |
| 2405 | |
| 2406 | assert_eq!( |
| 2407 | super::tool_run_summary(&run( |
| 2408 | &["session_sync"], |
| 2409 | super::ToolRunActivitySummary { |
| 2410 | other: 2, |
| 2411 | ..Default::default() |
| 2412 | } |
| 2413 | )), |
| 2414 | "Updated metadata", |
| 2415 | "a run of tools with no user-facing family falls back to a plain note" |
| 2416 | ); |
| 2417 | |
| 2418 | // Classification is derived from the real cells, not hand-set counters: |
| 2419 | // command tools count as commands, git history tools count as files. |
| 2420 | let commands = super::detect_tool_runs( |
| 2421 | &[ |
| 2422 | tool_cell("run_tests", ToolStatus::Success), |
| 2423 | tool_cell("run_verifiers", ToolStatus::Success), |
| 2424 | tool_cell("validate_data", ToolStatus::Success), |
| 2425 | ], |
| 2426 | 3, |
| 2427 | ); |
| 2428 | assert_eq!(commands[0].activity.commands, 3); |
| 2429 | assert_eq!( |
| 2430 | super::tool_run_summary(&commands[0]), |
| 2431 | "Ran 3 commands: run_tests, run_verifiers, validate_data" |
| 2432 | ); |
| 2433 | |
| 2434 | let git = super::detect_tool_runs( |
| 2435 | &[ |
| 2436 | tool_cell("git_log", ToolStatus::Success), |
| 2437 | tool_cell("git_show", ToolStatus::Success), |
| 2438 | tool_cell("git_blame", ToolStatus::Success), |
| 2439 | ], |
| 2440 | 3, |
| 2441 | ); |
| 2442 | assert_eq!(git[0].activity.files, 3); |
| 2443 | assert_eq!( |
| 2444 | super::tool_run_summary(&git[0]), |
| 2445 | "Explored 3 files: git_log, git_show, git_blame" |
| 2446 | ); |
| 2447 | } |
| 2448 | |
| 2449 | /// The small pure helpers behind the cards, as one table each. Every row is a |
| 2450 | /// documented input shape or a documented rejection; a helper that guesses on |
| 2451 | /// malformed input is worse than one that declines. |
| 2452 | /// |
| 2453 | /// Replaces ten single-case tests. |
| 2454 | #[test] |
| 2455 | fn the_card_helpers_accept_their_documented_forms_and_decline_the_rest() { |
| 2456 | // Agent ids come out of a JSON body that the renderer must not fully parse. |
| 2457 | for (input, expected) in [ |
| 2458 | ( |
| 2459 | r#"{"agent_id": "agent-abc12", "nickname": "Beluga"}"#, |
| 2460 | Some("agent-abc12"), |
| 2461 | ), |
| 2462 | ( |
| 2463 | "{\n \"agent_id\" : \"agent-xyz\",\n \"model\": \"x\"\n}", |
| 2464 | Some("agent-xyz"), |
| 2465 | ), |
| 2466 | (r#"{"nickname": "Orca", "model": "x"}"#, None), |
| 2467 | (r#"{"agent_id": "", "model": "x"}"#, None), |
| 2468 | ("(not json)", None), |
| 2469 | ("", None), |
| 2470 | ] { |
| 2471 | assert_eq!( |
| 2472 | super::extract_agent_id(input), |
| 2473 | expected, |
| 2474 | "extract_agent_id({input:?})" |
| 2475 | ); |
| 2476 | } |
| 2477 | |
| 2478 | // Checklist update prefixes: both vocabularies, and no guessing. |
| 2479 | for (input, expected) in [ |
| 2480 | ( |
| 2481 | "Updated todo #3 to in_progress\n{ \"items\": [...] }", |
| 2482 | Some(super::ChecklistChange { |
| 2483 | id: 3, |
| 2484 | status: "in_progress".to_string(), |
| 2485 | }), |
| 2486 | ), |
| 2487 | ( |
| 2488 | "Updated checklist #7 to completed\n{ \"items\": [] }", |
| 2489 | Some(super::ChecklistChange { |
| 2490 | id: 7, |
| 2491 | status: "completed".to_string(), |
| 2492 | }), |
| 2493 | ), |
| 2494 | ("{ \"items\": [] }", None), |
| 2495 | ("Wrote 5 todos\n{}", None), |
| 2496 | ("Updated todo #3\n", None), |
| 2497 | ("Updated todo #foo to done\n", None), |
| 2498 | ] { |
| 2499 | assert_eq!( |
| 2500 | super::parse_update_prefix(input), |
| 2501 | expected, |
| 2502 | "parse_update_prefix({input:?})" |
| 2503 | ); |
| 2504 | } |
| 2505 | |
| 2506 | // The elapsed badge appears at three seconds and not before, so quick |
| 2507 | // reads and greps do not visually churn. |
| 2508 | for secs in [0, 1, 2] { |
| 2509 | assert_eq!(running_status_label_with_elapsed(secs), "running"); |
| 2510 | } |
| 2511 | for secs in [3u64, 7, 120] { |
| 2512 | assert_eq!( |
| 2513 | running_status_label_with_elapsed(secs), |
| 2514 | format!("running ({secs}s)") |
| 2515 | ); |
| 2516 | } |
| 2517 | |
| 2518 | // A reasoning summary prefers an explicit Summary block, and otherwise is |
| 2519 | // the reasoning itself rather than nothing. |
| 2520 | assert_eq!( |
| 2521 | extract_reasoning_summary("Thinking...\nSummary: First line\nSecond line\n\nTail") |
| 2522 | .expect("summary"), |
| 2523 | "First line\nSecond line" |
| 2524 | ); |
| 2525 | assert_eq!( |
| 2526 | extract_reasoning_summary("Line one\nLine two").expect("summary"), |
| 2527 | "Line one\nLine two" |
| 2528 | ); |
| 2529 | } |
| 2530 | |
| 2531 | /// The card rail is the block's border, so it carries the cell's own state. |
| 2532 | /// Property, not token: every status paints a distinct rail, the rail is never |
| 2533 | /// left unstyled — an unstyled rail is what shipped before, and it made a |
| 2534 | /// failed card look exactly like a finished one from the border in — and it |
| 2535 | /// agrees with the glyph beside it on everything that needs attention while |
| 2536 | /// receding on a settled card, which is the whole point of the split. |
| 2537 | #[test] |
| 2538 | fn card_rail_carries_the_cell_status() { |
| 2539 | let mut rails = Vec::new(); |
| 2540 | for status in [ |
| 2541 | ToolStatus::Running, |
| 2542 | ToolStatus::Success, |
| 2543 | ToolStatus::Hydrated, |
| 2544 | ToolStatus::Warning, |
| 2545 | ToolStatus::Failed, |
| 2546 | ] { |
| 2547 | let cell = exec_tool("cargo test", status); |
| 2548 | let lines = cell.render(80, /*low_motion*/ true, RenderMode::Live); |
| 2549 | let first = &lines[0]; |
| 2550 | |
| 2551 | let rail = first.spans.first().expect("card rail span"); |
| 2552 | assert!( |
| 2553 | matches!(rail.content.as_ref(), "─ " | "╭ "), |
| 2554 | "{status:?} lost its card rail: {:?}", |
| 2555 | rail.content |
| 2556 | ); |
| 2557 | let rail_color = rail.style.fg.unwrap_or_else(|| { |
| 2558 | panic!("{status:?} left the card rail unstyled — the border must follow the state") |
| 2559 | }); |
| 2560 | |
| 2561 | let glyph_color = first.spans[1] |
| 2562 | .style |
| 2563 | .fg |
| 2564 | .expect("header status glyph must be styled"); |
| 2565 | assert_eq!( |
| 2566 | rail_color, glyph_color, |
| 2567 | "{status:?} must read the same on the rail and the glyph" |
| 2568 | ); |
| 2569 | |
| 2570 | rails.push((status, rail_color)); |
| 2571 | } |
| 2572 | |
| 2573 | for (i, (status, color)) in rails.iter().enumerate() { |
| 2574 | for (other_status, other_color) in &rails[i + 1..] { |
| 2575 | assert_ne!( |
| 2576 | color, other_color, |
| 2577 | "{status:?} and {other_status:?} draw the same rail" |
| 2578 | ); |
| 2579 | } |
| 2580 | } |
| 2581 | } |
| 2582 | |
| 2583 | /// Finished work shares quiet ink while its glyph shape preserves identity: |
| 2584 | /// a passed verify and a completed read must still be distinguishable. |
| 2585 | #[test] |
| 2586 | fn a_settled_verify_glyph_does_not_read_as_a_settled_read() { |
| 2587 | let verify = generic_tool("run_tests", ToolStatus::Success); |
| 2588 | let read = generic_tool("read_file", ToolStatus::Success); |
| 2589 | |
| 2590 | let glyph = |cell: &GenericToolCell| { |
| 2591 | cell.lines_with_mode_and_locale( |
| 2592 | 80, |
| 2593 | /*low_motion*/ true, |
| 2594 | RenderMode::Live, |
| 2595 | codewhale_localization::Locale::En, |
| 2596 | )[0] |
| 2597 | // Rail, shared status mark, then the tool-family identity glyph. |
| 2598 | .spans[2] |
| 2599 | .clone() |
| 2600 | }; |
| 2601 | let verify = glyph(&verify); |
| 2602 | let read = glyph(&read); |
| 2603 | assert_ne!(verify.content, read.content, "tool identity stays visible"); |
| 2604 | assert_eq!( |
| 2605 | verify.style.fg, read.style.fg, |
| 2606 | "settled work shares quiet ink" |
| 2607 | ); |
| 2608 | } |
| 2609 | |
| 2610 | /// An exploring cell rolls its parallel entries up into one state, and a |
| 2611 | /// single failure is never averaged away by its successful siblings. This is |
| 2612 | /// the fold that used to exist in three places and could not produce `Failed` |
| 2613 | /// in the one that painted the header. |
| 2614 | #[test] |
| 2615 | fn exploring_cell_status_keeps_the_loudest_terminal_state() { |
| 2616 | use super::{ExploringCell, ExploringEntry}; |
| 2617 | |
| 2618 | let cell = |statuses: &[ToolStatus]| ExploringCell { |
| 2619 | entries: statuses |
| 2620 | .iter() |
| 2621 | .map(|status| ExploringEntry { |
| 2622 | label: "Reading src/foo.rs".to_string(), |
| 2623 | status: *status, |
| 2624 | }) |
| 2625 | .collect(), |
| 2626 | }; |
| 2627 | |
| 2628 | for (statuses, expected) in [ |
| 2629 | ( |
| 2630 | vec![ToolStatus::Success, ToolStatus::Success], |
| 2631 | ToolStatus::Success, |
| 2632 | ), |
| 2633 | ( |
| 2634 | vec![ToolStatus::Success, ToolStatus::Running], |
| 2635 | ToolStatus::Running, |
| 2636 | ), |
| 2637 | ( |
| 2638 | vec![ToolStatus::Failed, ToolStatus::Running], |
| 2639 | ToolStatus::Running, |
| 2640 | ), |
| 2641 | ( |
| 2642 | vec![ToolStatus::Success, ToolStatus::Failed], |
| 2643 | ToolStatus::Failed, |
| 2644 | ), |
| 2645 | ( |
| 2646 | vec![ToolStatus::Warning, ToolStatus::Failed], |
| 2647 | ToolStatus::Failed, |
| 2648 | ), |
| 2649 | ( |
| 2650 | vec![ToolStatus::Success, ToolStatus::Warning], |
| 2651 | ToolStatus::Warning, |
| 2652 | ), |
| 2653 | ( |
| 2654 | vec![ToolStatus::Success, ToolStatus::Hydrated], |
| 2655 | ToolStatus::Hydrated, |
| 2656 | ), |
| 2657 | ] { |
| 2658 | assert_eq!( |
| 2659 | cell(&statuses).status(), |
| 2660 | expected, |
| 2661 | "{statuses:?} should roll up to {expected:?}" |
| 2662 | ); |
| 2663 | } |
| 2664 | |
| 2665 | // And the cell reports that state through the ToolCell it lives in. |
| 2666 | assert_eq!( |
| 2667 | ToolCell::Exploring(cell(&[ToolStatus::Success, ToolStatus::Failed])).status(), |
| 2668 | Some(ToolStatus::Failed) |
| 2669 | ); |
| 2670 | } |
| 2671 | |
| 2672 | // --------------------------------------------------------------------------- |
| 2673 | // Tool-card ink: rail vs glyph |
| 2674 | // --------------------------------------------------------------------------- |
| 2675 | |
| 2676 | /// The rail is the card border and follows OMP's rule: every status draws a |
| 2677 | /// distinct border. Two statuses sharing a rail is the failure this mapping |
| 2678 | /// exists to prevent — it is how `Hydrated` once sat on the running accent and |
| 2679 | /// read as live work. |
| 2680 | #[test] |
| 2681 | fn tool_rail_is_distinct_for_every_status() { |
| 2682 | let statuses = [ |
| 2683 | ToolStatus::Running, |
| 2684 | ToolStatus::Success, |
| 2685 | ToolStatus::Hydrated, |
| 2686 | ToolStatus::Warning, |
| 2687 | ToolStatus::Failed, |
| 2688 | ]; |
| 2689 | for (i, status) in statuses.iter().enumerate() { |
| 2690 | for other in &statuses[i + 1..] { |
| 2691 | assert_ne!( |
| 2692 | super::tool_rail_color(*status), |
| 2693 | super::tool_rail_color(*other), |
| 2694 | "{status:?} and {other:?} draw the same rail" |
| 2695 | ); |
| 2696 | } |
| 2697 | } |
| 2698 | } |
| 2699 | |
| 2700 | /// Finished rows recede while failures and warnings retain attention ink. |
| 2701 | #[test] |
| 2702 | fn calm1_settled_headers_are_muted_without_hiding_failures() { |
| 2703 | use crate::tui::widgets::tool_card::ToolFamily; |
| 2704 | use ratatui::style::Modifier; |
| 2705 | for family in [ToolFamily::Read, ToolFamily::Verify] { |
| 2706 | for status in [ |
| 2707 | ToolStatus::Running, |
| 2708 | ToolStatus::Success, |
| 2709 | ToolStatus::Hydrated, |
| 2710 | ToolStatus::Warning, |
| 2711 | ToolStatus::Failed, |
| 2712 | ] { |
| 2713 | assert_eq!( |
| 2714 | super::tool_rail_color(status), |
| 2715 | super::tool_glyph_color(status, family) |
| 2716 | ); |
| 2717 | } |
| 2718 | assert_ne!( |
| 2719 | super::tool_glyph_color(ToolStatus::Success, family), |
| 2720 | super::tool_glyph_color(ToolStatus::Failed, family) |
| 2721 | ); |
| 2722 | } |
| 2723 | assert!( |
| 2724 | !super::tool_title_style(ToolStatus::Success) |
| 2725 | .add_modifier |
| 2726 | .contains(Modifier::BOLD) |
| 2727 | ); |
| 2728 | assert!( |
| 2729 | super::tool_title_style(ToolStatus::Failed) |
| 2730 | .add_modifier |
| 2731 | .contains(Modifier::BOLD) |
| 2732 | ); |
| 2733 | } |
| 2734 | |
| 2735 | /// Issue #5871: `todo_write` replaces the whole list on every call, so a long |
| 2736 | /// session stacked full checklist cards that could only be cleared by `/clear` |
| 2737 | /// or `/new` — and both of those also drop `api_messages` and the compaction |
| 2738 | /// summary. Only the newest snapshot keeps its card; the ones it replaced keep |
| 2739 | /// their header (the progress reading) and a details affordance. |
| 2740 | #[test] |
| 2741 | fn superseded_todo_snapshots_collapse_to_their_header() { |
| 2742 | let snapshot = |done: usize| { |
| 2743 | let items: Vec<String> = (0..3) |
| 2744 | .map(|i| { |
| 2745 | let status = if i < done { "completed" } else { "pending" }; |
| 2746 | format!(r#"{{"content":"step {i}","status":"{status}"}}"#) |
| 2747 | }) |
| 2748 | .collect(); |
| 2749 | let mut cell = generic_tool("todo_write", ToolStatus::Success); |
| 2750 | cell.output = Some(format!(r#"{{"items":[{}]}}"#, items.join(","))); |
| 2751 | HistoryCell::Tool(ToolCell::Generic(cell)) |
| 2752 | }; |
| 2753 | |
| 2754 | let mut options = TranscriptRenderOptions { |
| 2755 | show_tool_details: true, |
| 2756 | ..Default::default() |
| 2757 | }; |
| 2758 | let older = snapshot(1); |
| 2759 | |
| 2760 | let expanded = older.lines_with_options(120, options); |
| 2761 | assert!( |
| 2762 | expanded.len() > 2, |
| 2763 | "the newest snapshot renders its full card: {expanded:?}" |
| 2764 | ); |
| 2765 | |
| 2766 | options.superseded_work_receipt = true; |
| 2767 | let collapsed = older.lines_with_options(120, options); |
| 2768 | assert_eq!( |
| 2769 | collapsed.len(), |
| 2770 | 2, |
| 2771 | "a replaced snapshot keeps its header plus the details affordance" |
| 2772 | ); |
| 2773 | let header: String = collapsed[0] |
| 2774 | .spans |
| 2775 | .iter() |
| 2776 | .map(|span| span.content.as_ref()) |
| 2777 | .collect(); |
| 2778 | assert!( |
| 2779 | header.contains("1/3"), |
| 2780 | "the collapsed row keeps the progress reading: {header}" |
| 2781 | ); |
| 2782 | } |
| 2783 | |
| 2784 | /// One click is one request to open one file. |
| 2785 | /// |
| 2786 | /// The old `try_open_file_at_line` looped over every line of the cell and |
| 2787 | /// spawned a detached editor per match, so a stack trace or a grep result could |
| 2788 | /// launch several at once, all fighting the still-raw-mode TUI for the tty |
| 2789 | /// (#6235). The parser now returns the first resolvable reference and nothing |
| 2790 | /// else; spawning belongs to `external_editor`. |
| 2791 | #[test] |
| 2792 | fn first_file_line_reference_returns_one_match_and_resolves_it() { |
| 2793 | let dir = tempfile::tempdir().unwrap(); |
| 2794 | let workspace = dir.path(); |
| 2795 | std::fs::create_dir_all(workspace.join("src")).unwrap(); |
| 2796 | std::fs::write(workspace.join("src/first.rs"), "fn a() {}\n").unwrap(); |
| 2797 | std::fs::write(workspace.join("src/second.rs"), "fn b() {}\n").unwrap(); |
| 2798 | |
| 2799 | let text = "note: two frames below\n src/first.rs:12\n src/second.rs:34\n"; |
| 2800 | let (path, line) = super::first_file_line_reference(text, workspace) |
| 2801 | .expect("the first resolvable reference is returned"); |
| 2802 | assert_eq!(path, workspace.join("src/first.rs")); |
| 2803 | assert_eq!(line, 12); |
| 2804 | } |
| 2805 | |
| 2806 | /// The forms tools and models print: rustc's `-->` locator with a column, |
| 2807 | /// a backticked reference, and one inside a sentence with punctuation. |
| 2808 | #[test] |
| 2809 | fn file_line_reference_reads_common_forms_on_one_line() { |
| 2810 | let dir = tempfile::tempdir().unwrap(); |
| 2811 | let workspace = dir.path(); |
| 2812 | std::fs::create_dir_all(workspace.join("src")).unwrap(); |
| 2813 | std::fs::write(workspace.join("src/a.rs"), "fn a() {}\n").unwrap(); |
| 2814 | let expected = Some((workspace.join("src/a.rs"), 12)); |
| 2815 | |
| 2816 | for line in [ |
| 2817 | " --> src/a.rs:12:5", |
| 2818 | "see `src/a.rs:12` for the loop", |
| 2819 | "the bug is in (src/a.rs:12), again.", |
| 2820 | "src/a.rs:12: error: mismatched types", |
| 2821 | "./src/a.rs:12", |
| 2822 | ] { |
| 2823 | assert_eq!( |
| 2824 | super::file_line_reference(line, workspace), |
| 2825 | expected, |
| 2826 | "{line:?}" |
| 2827 | ); |
| 2828 | } |
| 2829 | assert_eq!(super::file_line_reference("src/a.rs:0", workspace), None); |
| 2830 | } |
| 2831 | |
| 2832 | /// Model output is not trusted to name a file: an absolute path outside the |
| 2833 | /// workspace and a `../` escape both used to open in `$EDITOR`. |
| 2834 | #[test] |
| 2835 | fn file_line_reference_refuses_paths_outside_the_workspace() { |
| 2836 | let root = tempfile::tempdir().unwrap(); |
| 2837 | let workspace = root.path().join("ws"); |
| 2838 | std::fs::create_dir_all(workspace.join("src")).unwrap(); |
| 2839 | std::fs::write(workspace.join("src/in.rs"), "fn a() {}\n").unwrap(); |
| 2840 | let outside = root.path().join("outside.rs"); |
| 2841 | std::fs::write(&outside, "secret\n").unwrap(); |
| 2842 | |
| 2843 | let absolute_outside = format!("{}:3", outside.display()); |
| 2844 | assert_eq!( |
| 2845 | super::file_line_reference(&absolute_outside, &workspace), |
| 2846 | None |
| 2847 | ); |
| 2848 | assert_eq!( |
| 2849 | super::file_line_reference("../outside.rs:3", &workspace), |
| 2850 | None |
| 2851 | ); |
| 2852 | assert_eq!( |
| 2853 | super::first_file_line_reference( |
| 2854 | &format!("{absolute_outside}\n../outside.rs:1\n"), |
| 2855 | &workspace |
| 2856 | ), |
| 2857 | None |
| 2858 | ); |
| 2859 | |
| 2860 | let absolute_inside = format!("{}:4", workspace.join("src/in.rs").display()); |
| 2861 | assert_eq!( |
| 2862 | super::file_line_reference(&absolute_inside, &workspace), |
| 2863 | Some((workspace.join("src/in.rs"), 4)), |
| 2864 | "an absolute path inside the workspace still opens" |
| 2865 | ); |
| 2866 | } |
| 2867 | |
| 2868 | /// A link inside the workspace passed the text-only check and `is_file()` |
| 2869 | /// followed it, so `vendor -> <outside>` or `notes.md -> <outside file>` in |
| 2870 | /// model output offered "Open in editor" on a file outside the workspace. |
| 2871 | #[cfg(unix)] |
| 2872 | #[test] |
| 2873 | fn file_line_reference_refuses_links_out_of_the_workspace() { |
| 2874 | let root = tempfile::tempdir().unwrap(); |
| 2875 | let workspace = root.path().join("ws"); |
| 2876 | let outside = root.path().join("outside"); |
| 2877 | std::fs::create_dir_all(workspace.join("src")).unwrap(); |
| 2878 | std::fs::create_dir_all(&outside).unwrap(); |
| 2879 | std::fs::write(outside.join("secret.rs"), "secret\n").unwrap(); |
| 2880 | std::fs::write(workspace.join("src/in.rs"), "fn a() {}\n").unwrap(); |
| 2881 | std::os::unix::fs::symlink(&outside, workspace.join("vendor")).unwrap(); |
| 2882 | std::os::unix::fs::symlink(outside.join("secret.rs"), workspace.join("notes.rs")).unwrap(); |
| 2883 | // A link that stays inside is still a link: refused, not followed. |
| 2884 | std::os::unix::fs::symlink(workspace.join("src"), workspace.join("alias")).unwrap(); |
| 2885 | |
| 2886 | for line in ["vendor/secret.rs:1", "notes.rs:1", "alias/in.rs:1"] { |
| 2887 | assert_eq!( |
| 2888 | super::file_line_reference(line, &workspace), |
| 2889 | None, |
| 2890 | "{line:?}" |
| 2891 | ); |
| 2892 | } |
| 2893 | assert_eq!( |
| 2894 | super::workspace_file(&workspace, "src"), |
| 2895 | None, |
| 2896 | "a directory" |
| 2897 | ); |
| 2898 | assert_eq!( |
| 2899 | super::file_line_reference("src/in.rs:2", &workspace), |
| 2900 | Some((workspace.join("src/in.rs"), 2)) |
| 2901 | ); |
| 2902 | } |
| 2903 | |
| 2904 | /// The two escapes the review named: a directory link to `/` and links into |
| 2905 | /// an `.ssh` directory. The `.ssh` here is one the test creates outside the |
| 2906 | /// workspace with real files in it, so a follow-the-link check would find |
| 2907 | /// them and resolve; the test does not depend on the host's own keys. |
| 2908 | #[cfg(unix)] |
| 2909 | #[test] |
| 2910 | fn file_line_reference_refuses_links_to_root_and_ssh() { |
| 2911 | let dir = tempfile::tempdir().unwrap(); |
| 2912 | let workspace = &dir.path().join("ws"); |
| 2913 | std::fs::create_dir_all(workspace).unwrap(); |
| 2914 | std::os::unix::fs::symlink("/", workspace.join("rootfs")).unwrap(); |
| 2915 | let ssh = dir.path().join("home/.ssh"); |
| 2916 | std::fs::create_dir_all(&ssh).unwrap(); |
| 2917 | std::fs::write(ssh.join("id_ed25519"), "PRIVATE KEY\n").unwrap(); |
| 2918 | std::fs::write(ssh.join("config"), "Host *\n").unwrap(); |
| 2919 | std::os::unix::fs::symlink(ssh.join("id_ed25519"), workspace.join("key.rs")).unwrap(); |
| 2920 | std::os::unix::fs::symlink(&ssh, workspace.join("ssh")).unwrap(); |
| 2921 | assert!(workspace.join("key.rs").is_file(), "the file link resolves"); |
| 2922 | assert!( |
| 2923 | workspace.join("ssh/config").is_file(), |
| 2924 | "the dir link resolves" |
| 2925 | ); |
| 2926 | |
| 2927 | for line in [ |
| 2928 | "rootfs/etc/hosts:1", |
| 2929 | "./rootfs/etc/hosts:1", |
| 2930 | "key.rs:1", |
| 2931 | "ssh/config:1", |
| 2932 | "ssh/id_ed25519:1", |
| 2933 | ] { |
| 2934 | assert_eq!( |
| 2935 | super::file_line_reference(line, workspace), |
| 2936 | None, |
| 2937 | "{line:?}" |
| 2938 | ); |
| 2939 | } |
| 2940 | let absolute = workspace.join("rootfs/etc/hosts"); |
| 2941 | assert_eq!( |
| 2942 | super::workspace_file(workspace, absolute.to_str().unwrap()), |
| 2943 | None, |
| 2944 | "an absolute path through the link" |
| 2945 | ); |
| 2946 | } |
| 2947 | |
| 2948 | #[test] |
| 2949 | fn first_file_line_reference_skips_unresolvable_and_malformed_rows() { |
| 2950 | let dir = tempfile::tempdir().unwrap(); |
| 2951 | let workspace = dir.path(); |
| 2952 | std::fs::create_dir_all(workspace.join("src")).unwrap(); |
| 2953 | std::fs::write(workspace.join("src/real.rs"), "fn a() {}\n").unwrap(); |
| 2954 | |
| 2955 | // A path that does not exist, a bare word, a non-numeric suffix and an |
| 2956 | // empty suffix all fall through to the one row that resolves. |
| 2957 | let text = concat!( |
| 2958 | " src/missing.rs:9\n", |
| 2959 | " notafile:12\n", |
| 2960 | " src/real.rs:abc\n", |
| 2961 | " src/real.rs:\n", |
| 2962 | " src/real.rs:7\n", |
| 2963 | ); |
| 2964 | let (path, line) = |
| 2965 | super::first_file_line_reference(text, workspace).expect("the only resolvable row wins"); |
| 2966 | assert_eq!(path, workspace.join("src/real.rs")); |
| 2967 | assert_eq!(line, 7); |
| 2968 | |
| 2969 | assert!( |
| 2970 | super::first_file_line_reference("no references here\n", workspace).is_none(), |
| 2971 | "a cell with nothing to open must report nothing, not a default" |
| 2972 | ); |
| 2973 | } |
| 2974 | |
| 2975 | /// #6601: the project-trust warning is a runtime-owned internal message; the |
| 2976 | /// model reads it, the transcript never shows it as the user's words. |
| 2977 | #[test] |
| 2978 | fn workspace_trust_warning_renders_no_transcript_cell() { |
| 2979 | for warning in [Some("untrusted project skills were skipped"), None] { |
| 2980 | let message = crate::runtime_handoff::workspace_trust_runtime_message(warning); |
| 2981 | assert!(crate::runtime_handoff::is_internal_runtime_handoff( |
| 2982 | &message |
| 2983 | )); |
| 2984 | assert!( |
| 2985 | super::history_cells_from_message(&message).is_empty(), |
| 2986 | "{warning:?}" |
| 2987 | ); |
| 2988 | } |
| 2989 | } |
| 2990 | |
| 2991 | #[test] |
| 2992 | fn calm1_failed_tool_preview_matrix_preserves_full_details_and_mcp_identity() { |
| 2993 | for width in [40, 60, 80, 140] { |
| 2994 | for status in [ToolStatus::Running, ToolStatus::Success, ToolStatus::Failed] { |
| 2995 | let output = (0..30) |
| 2996 | .map(|i| format!("row {i:02}")) |
| 2997 | .collect::<Vec<_>>() |
| 2998 | .join("\n"); |
| 2999 | let mcp = HistoryCell::Tool(ToolCell::Mcp(super::McpToolCell { |
| 3000 | tool: "linear_get_issue".into(), |
| 3001 | status, |
| 3002 | content: Some(output.clone()), |
| 3003 | is_image: false, |
| 3004 | })); |
| 3005 | let mut generic = generic_tool("read_file", status); |
| 3006 | generic.output = Some(output); |
| 3007 | for (cell, is_mcp) in [ |
| 3008 | (mcp, true), |
| 3009 | (HistoryCell::Tool(ToolCell::Generic(generic)), false), |
| 3010 | ] { |
| 3011 | let options = TranscriptRenderOptions { |
| 3012 | calm_mode: true, |
| 3013 | show_tool_details: false, |
| 3014 | low_motion: true, |
| 3015 | ..Default::default() |
| 3016 | }; |
| 3017 | let live = cell.lines_with_options(width, options); |
| 3018 | let text = lines_text(&live); |
| 3019 | let full = lines_text(&cell.transcript_lines(width)); |
| 3020 | assert!( |
| 3021 | full.contains("row 15"), |
| 3022 | "full details must preserve omitted content" |
| 3023 | ); |
| 3024 | if is_mcp { |
| 3025 | assert_eq!(text.matches("linear_get_issue").count(), 1, "{text}"); |
| 3026 | } |
| 3027 | if status == ToolStatus::Failed { |
| 3028 | assert!(text.contains("row 00") && text.contains("row 29"), "{text}"); |
| 3029 | assert!(!text.contains("row 15"), "{text}"); |
| 3030 | assert!(text.contains("lines omitted"), "{text}"); |
| 3031 | assert_eq!( |
| 3032 | (0..30) |
| 3033 | .filter(|i| text.contains(&format!("row {i:02}"))) |
| 3034 | .count(), |
| 3035 | 6 |
| 3036 | ); |
| 3037 | if is_mcp { |
| 3038 | assert!(live.len() <= 8, "{text}"); |
| 3039 | } |
| 3040 | } |
| 3041 | } |
| 3042 | } |
| 3043 | } |
| 3044 | } |
| 3045 | |
| 3046 | #[test] |
| 3047 | fn calm1_settled_reasoning_is_one_localized_row_and_stays_expandable() { |
| 3048 | for (locale, expected) in [ |
| 3049 | (codewhale_localization::Locale::En, "Thought for 12s"), |
| 3050 | (codewhale_localization::Locale::ZhHans, "思考用时 12s"), |
| 3051 | ] { |
| 3052 | let cell = HistoryCell::Thinking { |
| 3053 | content: "private reasoning body\nlast step".into(), |
| 3054 | streaming: false, |
| 3055 | duration_secs: Some(12.0), |
| 3056 | }; |
| 3057 | let options = TranscriptRenderOptions { |
| 3058 | calm_mode: true, |
| 3059 | locale, |
| 3060 | ..Default::default() |
| 3061 | }; |
| 3062 | let (lines, action) = cell.lines_with_options_folded(80, options, None); |
| 3063 | assert_eq!(lines.len(), 1); |
| 3064 | assert!( |
| 3065 | lines_text(&lines).contains(expected), |
| 3066 | "{}", |
| 3067 | lines_text(&lines) |
| 3068 | ); |
| 3069 | assert_eq!(action, Some(super::ReasoningAction::Expand)); |
| 3070 | let expanded = cell |
| 3071 | .lines_with_options_folded(80, options, Some(ThinkingFold::Expanded)) |
| 3072 | .0; |
| 3073 | assert!(lines_text(&expanded).contains("private reasoning body")); |
| 3074 | } |
| 3075 | } |
| 3076 | |
| 3077 | #[test] |
| 3078 | fn calm1_calm_and_hidden_details_share_one_card_budget() { |
| 3079 | let mut exec = exec_tool("command", ToolStatus::Success); |
| 3080 | exec.output = Some(numbered_output(40)); |
| 3081 | let cell = HistoryCell::Tool(ToolCell::Exec(exec)); |
| 3082 | let mut rendered = Vec::new(); |
| 3083 | for (calm_mode, show_tool_details) in [(true, false), (true, true), (false, false)] { |
| 3084 | let options = TranscriptRenderOptions { |
| 3085 | calm_mode, |
| 3086 | show_tool_details, |
| 3087 | low_motion: true, |
| 3088 | ..Default::default() |
| 3089 | }; |
| 3090 | let lines = cell.lines_with_options(80, options); |
| 3091 | assert!(lines.len() <= super::constants::TOOL_SUMMARY_CARD_LINES); |
| 3092 | rendered.push(lines_text(&lines)); |
| 3093 | } |
| 3094 | assert!(rendered.windows(2).all(|pair| pair[0] == pair[1])); |
| 3095 | } |
| 3096 |