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