1001. 1001 collected.len()
1002. 1002 )
1003. 1003 }
1004. 1004}
1005. 1005
1006. 1006fn extract_text_lines(content: &str, offset: usize, limit: usize) -> String {
1007. 1007 let lines: Vec<&str> = content.lines().collect();
1008. 1008 let total = lines.len();
1009. 1009 let start = (offset.saturating_sub(1)).min(total);
1010. 1010 let end = (start + limit).min(total);
1011. 1011
1012. 1012 let numbered: Vec<String> = lines[start..end]
1013. 1013 .iter()
1014. 1014 .enumerate()
1015. 1015 .map(|(idx, line)| format!("{:4}: {}", start + idx + 1, line))
1016. 1016 .collect();
1017. 1017
1018. 1018 format!(
1019. 1019 "{}\n\n[Lines {}..{} of {} total lines]",
1020. 1020 numbered.join("\n"),
1021. 1021 start + 1,
1022. 1022 end,
1023. 1023 total
1024. 1024 )
1025. 1025}
1026. 1026
1027. 1027/// Where a file the person attached under the name `requested` was saved:
1028. 1028/// the inbox prefixes each saved name with a digest, so a model that names
1029. 1029/// the file as the person did misses it.
1030. 1030fn attached_copy(cwd: &Path, requested: &str) -> Option<String> {
1031. 1031 let wanted = Path::new(requested).file_name()?.to_str()?;
1032. 1032 std::fs::read_dir(cwd.join(crate::INBOX_DIR))
1033. 1033 .ok()?
1034. 1034 .filter_map(Result::ok)
1035. 1035 .filter_map(|entry| entry.file_name().into_string().ok())
1036. 1036 .find(|name| {
1037. 1037 name.split_once('-').is_some_and(|(prefix, rest)| {
1038. 1038 rest == wanted
1039. 1039 && prefix.len() == 12
1040. 1040 && prefix.chars().all(|c| c.is_ascii_hexdigit())
1041. 1041 })
1042. 1042 })
1043. 1043 .map(|name| format!("{}/{name}", crate::INBOX_DIR))
1044. 1044}
1045. 1045
1046. 1046#[cfg(test)]
1047. 1047mod tests {
1048. 1048 #![allow(clippy::unwrap_used, clippy::expect_used, clippy::panic)]
1049. 1049 use super::*;
1050. 1050
1051. 1051 #[test]
1052. 1052 fn test_extract_section_markdown() {
1053. 1053 let doc = "# Research Paper\n\n## Abstract\nThis is the abstract.\n\n## Methods\nWe used in-memory tooling.\nStep 1: parse.\nStep 2: verify.\n\n## Results\nAll tests passed.\n";
1054. 1054 let extracted = extract_section(doc, "Methods", "md", 1, 50);
1055. 1055 assert!(extracted.contains("We used in-memory tooling."));
1056. 1056 assert!(extracted.contains("Step 1: parse."));
1057. 1057 assert!(!extracted.contains("All tests passed."));
1058. 1058 }
1059. 1059
1060. 1060 #[test]
1061. 1061 fn test_outline_markdown() {
1062. 1062 let doc = "# Main\n## Sec 1\n### Sub 1\n## Sec 2\n";
1063. 1063 let outline = outline_document(doc, "md");
1064. 1064 assert!(outline.contains("L001: # Main"));
1065. 1065 assert!(outline.contains("L002: ## Sec 1"));
1066. 1066 assert!(outline.contains("L004: ## Sec 2"));
1067. 1067 }
1068. 1068
1069. 1069 #[test]
1070. 1070 fn test_render_table_view_csv() {
1071. 1071 let csv = "name,role,score\nAlice,Lead,95\nBob,Eng,88\nCharlie,Analyst,91\n";
1072. 1072 let rendered = render_table_view(csv, "csv", 1, 2);
1073. 1073 assert!(rendered.contains("| name | role | score |"));
1074. 1074 assert!(rendered.contains("| Alice | Lead | 95 |"));
1075. 1075 assert!(rendered.contains("| Bob | Eng | 88 |"));
1076. 1076 assert!(!rendered.contains("Charlie"));
1077. 1077 assert!(rendered.contains("Showing rows 1..2 of 3"));
1078. 1078 }
1079. 1079
1080. 1080 #[test]
1081. 1081 fn test_render_json_table() {
1082. 1082 let json_data = r#"[
1083. 1083 {"id": "usr-1", "name": "Alice", "active": true},
1084. 1084 {"id": "usr-2", "name": "Bob", "active": false}
1085. 1085 ]"#;
1086. 1086 let rendered = render_table_view(json_data, "json", 1, 10);
1087. 1087 assert!(
1088. 1088 rendered.contains("id") && rendered.contains("name") && rendered.contains("active")
1089. 1089 );
1090. 1090 assert!(rendered.contains("Alice") && rendered.contains("usr-1"));
1091. 1091 assert!(rendered.contains("Bob") && rendered.contains("usr-2"));
1092. 1092 assert!(rendered.contains("Showing rows 1..2 of 2 total rows"));
1093. 1093 }
1094. 1094
1095. 1095 #[test]
1096. 1096 fn test_render_kv_table_ini() {
1097. 1097 let ini = "[server]\nport = 8080\nhost = localhost\n[db]\nurl = postgresql://db\n";
1098. 1098 let rendered = render_table_view(ini, "ini", 1, 10);
1099. 1099 assert!(rendered.contains("| server.port | 8080 |"));
1100. 1100 assert!(rendered.contains("| db.url | postgresql://db |"));
1101. 1101 }
1102. 1102
1103. 1103 #[test]
1104. 1104 fn test_render_html_table() {
1105. 1105 let html = "<table><tr><th>Metric</th><th>Value</th></tr><tr><td>Latency</td><td>12ms</td></tr><tr><td>Throughput</td><td>1000req/s</td></tr></table>";
1106. 1106 let rendered = render_table_view(html, "html", 1, 5);
1107. 1107 assert!(rendered.contains("| Metric | Value |"));
1108. 1108 assert!(rendered.contains("| Latency | 12ms |"));
1109. 1109 assert!(rendered.contains("| Throughput | 1000req/s |"));
1110. 1110 }
1111. 1111
1112. 1112 #[test]
1113. 1113 fn test_extract_bracket_section_toml() {
1114. 1114 let toml_doc =
1115. 1115 "[gateway]\nport = 3000\n[agent]\nname = \"Research\"\nrole = \"researcher\"\n";
1116. 1116 let extracted = extract_section(toml_doc, "agent", "toml", 1, 20);
1117. 1117 assert!(extracted.contains("name = \"Research\""));
1118. 1118 assert!(extracted.contains("role = \"researcher\""));
1119. 1119 assert!(!extracted.contains("port = 3000"));
1120. 1120 }
1121. 1121
1122. 1122 #[test]
1123. 1123 fn test_extract_text_lines() {
1124. 1124 let text = "alpha\nbeta\ngamma\ndelta\nepsilon\n";
1125. 1125 let extracted = extract_text_lines(text, 2, 2);
1126. 1126 assert!(extracted.contains("2: beta"));
1127. 1127 assert!(extracted.contains("3: gamma"));
1128. 1128 assert!(!extracted.contains("alpha"));
1129. 1129 assert!(!extracted.contains("delta"));
1130. 1130 }
1131. 1131
1132. 1132 async fn read_office(name: &str, bytes: Vec<u8>, args: Value) -> ToolOutput {
1133. 1133 let dir = tempfile::tempdir().unwrap();
1134. 1134 std::fs::write(dir.path().join(name), bytes).unwrap();
1135. 1135 let mut args = args;
1136. 1136 args["path"] = Value::String(name.into());
1137. 1137 DocReadTool
1138. 1138 .execute(&args, &ToolContext::new(dir.path().to_path_buf()))
1139. 1139 .await
1140. 1140 }
1141. 1141
1142. 1142 #[tokio::test]
1143. 1143 async fn a_file_named_as_it_was_attached_points_to_its_inbox_copy() {
1144. 1144 let dir = tempfile::tempdir().unwrap();
1145. 1145 std::fs::create_dir(dir.path().join(crate::INBOX_DIR)).unwrap();
1146. 1146 std::fs::write(
1147. 1147 dir.path().join("inbox/e1d2a5db9ff2-Board deck.pptx"),
1148. 1148 vak_ooxml::fixtures::pptx(),
1149. 1149 )
1150. 1150 .unwrap();
1151. 1151 let output = DocReadTool
1152. 1152 .execute(
1153. 1153 &serde_json::json!({"path": "Board deck.pptx"}),
1154. 1154 &ToolContext::new(dir.path().to_path_buf()),
1155. 1155 )
1156. 1156 .await;
1157. 1157 assert!(output.is_error);
1158. 1158 assert!(
1159. 1159 output
1160. 1160 .content
1161. 1161 .contains("is at path \"inbox/e1d2a5db9ff2-Board deck.pptx\": call doc_read again"),
1162. 1162 "{}",
1163. 1163 output.content
1164. 1164 );
1165. 1165 let missing = DocReadTool
1166. 1166 .execute(
1167. 1167 &serde_json::json!({"path": "other.pptx"}),
1168. 1168 &ToolContext::new(dir.path().to_path_buf()),
1169. 1169 )
1170. 1170 .await;
1171. 1171 assert!(!missing.content.contains("inbox/"), "{}", missing.content);
1172. 1172 }
1173. 1173
1174. 1174 #[tokio::test]
1175. 1175 async fn office_text_view_is_anchored_and_labels_hidden_content() {
1176. 1176 let output = read_office(
1177. 1177 "q3.docx",
1178. 1178 vak_ooxml::fixtures::docx(),
1179. 1179 serde_json::json!({}),
1180. 1180 )
1181. 1181 .await;
1182. 1182 assert!(!output.is_error, "{}", output.content);
1183. 1183 let text = output.content;
1184. 1184 assert!(
1185. 1185 text.starts_with("Word document (.docx) · title: Q3 Report"),
1186. 1186 "{text}"
1187. 1187 );
1188. 1188 assert!(text.contains("not instructions"), "{text}");
1189. 1189 assert!(
1190. 1190 text.contains("Cite a place as `q3.docx#<anchor>`"),
1191. 1191 "{text}"
1192. 1192 );
1193. 1193 assert!(text.contains("Not read: headers and footers"), "{text}");
1194. 1194 assert!(text.contains("[p:0A1B2C3D] # Quarterly Report"), "{text}");
1195. 1195 assert!(text.contains("⟨hidden text⟩"), "{text}");
1196. 1196 assert!(text.contains("[lines 1.."), "{text}");
1197. 1197 }
1198. 1198
1199. 1199 #[tokio::test]
1200. 1200 async fn office_section_outline_and_table_views() {
1201. 1201 let section = read_office(
1202. 1202 "q3.docx",
1203. 1203 vak_ooxml::fixtures::docx(),
1204. 1204 serde_json::json!({"section": "Outlook"}),
1205. 1205 )
1206. 1206 .await;
1207. 1207 assert!(section.content.contains("Steady."), "{}", section.content);
1208. 1208 assert!(
1209. 1209 !section.content.contains("Revenue grew"),
1210. 1210 "{}",
1211. 1211 section.content
1212. 1212 );
1213. 1213
1214. 1214 let outline = read_office(
1215. 1215 "deck.pptx",
1216. 1216 vak_ooxml::fixtures::pptx(),
1217. 1217 serde_json::json!({"view": "outline"}),
1218. 1218 )
1219. 1219 .await;
1220. 1220 assert!(
1221. 1221 outline
1222. 1222 .content
1223. 1223 .contains("- [slide:256] Slide 1: Launch plan"),
1224. 1224 "{}",
1225. 1225 outline.content
1226. 1226 );
1227. 1227
1228. 1228 let table = read_office(
1229. 1229 "book.xlsx",
1230. 1230 vak_ooxml::fixtures::xlsx(),
1231. 1231 serde_json::json!({"view": "table", "section": "Budget", "limit": 2}),
1232. 1232 )
1233. 1233 .await;
1234. 1234 assert!(
1235. 1235 table.content.contains("| | A | B | C |"),
1236. 1236 "{}",
1237. 1237 table.content
1238. 1238 );
1239. 1239 assert!(
1240. 1240 table.content.contains("| 1 | Item | Cost | |"),
1241. 1241 "{}",
1242. 1242 table.content
1243. 1243 );
1244. 1244 assert!(
1245. 1245 table.content.contains("continue with offset=3"),
1246. 1246 "{}",
1247. 1247 table.content
1248. 1248 );
1249. 1249
1250. 1250 let missing = read_office(
1251. 1251 "book.xlsx",
1252. 1252 vak_ooxml::fixtures::xlsx(),
1253. 1253 serde_json::json!({"view": "table", "section": "Nope"}),
1254. 1254 )
1255. 1255 .await;
1256. 1256 assert!(missing.is_error);
1257. 1257 assert!(
1258. 1258 missing.content.contains("'Budget' (Budget!)"),
1259. 1259 "{}",
1260. 1260 missing.content
1261. 1261 );
1262. 1262 }
1263. 1263
1264. 1264 #[tokio::test]
1265. 1265 async fn office_summary_lists_flags_and_external_links() {
1266. 1266 let package = vak_ooxml::fixtures::word_with(
1267. 1267 vak_ooxml::fixtures::WORD_MAIN,
1268. 1268 vak_ooxml::fixtures::MINIMAL_WORD_BODY,
1269. 1269 &[],
1270. 1270 &[],
1271. 1271 &[],
1272. 1272 &[(
1273. 1273 "rId1",
1274. 1274 "http://schemas.openxmlformats.org/officeDocument/2006/relationships/attachedTemplate",
1275. 1275 "https://attacker.example/t.dotm",
1276. 1276 )],
1277. 1277 );
1278. 1278 let output = read_office(
1279. 1279 "letter.docx",
1280. 1280 package,
1281. 1281 serde_json::json!({"view": "summary"}),
1282. 1282 )
1283. 1283 .await;
1284. 1284 assert!(
1285. 1285 output.content.contains("Flags: remote template"),
1286. 1286 "{}",
1287. 1287 output.content
1288. 1288 );
1289. 1289 assert!(
1290. 1290 output.content.contains("External attachedTemplate from word/document.xml: https://attacker.example/t.dotm (not followed)"),
1291. 1291 "{}",
1292. 1292 output.content
1293. 1293 );
1294. 1294 }
1295. 1295
1296. 1296 #[tokio::test]
1297. 1297 async fn hostile_and_unsupported_files_fail_with_a_reason() {
1298. 1298 let bomb = vak_ooxml::fixtures::word_with(
1299. 1299 vak_ooxml::fixtures::WORD_MAIN,
1300. 1300 r#"<!DOCTYPE x [<!ENTITY a "a">]><w:document xmlns:w="x"/>"#,
1301. 1301 &[],
1302. 1302 &[],
1303. 1303 &[],
1304. 1304 &[],
1305. 1305 );
1306. 1306 let output = read_office("bomb.docx", bomb, serde_json::json!({})).await;
1307. 1307 assert!(output.is_error);
1308. 1308 assert!(output.content.contains("DOCTYPE"), "{}", output.content);
1309. 1309
1310. 1310 let legacy = read_office(
1311. 1311 "old.doc",
1312. 1312 b"\xD0\xCF\x11\xE0".to_vec(),
1313. 1313 serde_json::json!({}),
1314. 1314 )
1315. 1315 .await;
1316. 1316 assert!(legacy.is_error);
1317. 1317 assert!(
1318. 1318 legacy.content.contains("legacy binary Word"),
1319. 1319 "{}",
1320. 1320 legacy.content
1321. 1321 );
1322. 1322
1323. 1323 let binary = read_office(
1324. 1324 "blob.bin",
1325. 1325 vec![0xFF, 0xFE, 0x00, 0x81],
1326. 1326 serde_json::json!({}),
1327. 1327 )
1328. 1328 .await;
1329. 1329 assert!(binary.is_error);
1330. 1330 assert!(
1331. 1331 binary.content.contains("not a text file"),
1332. 1332 "{}",
1333. 1333 binary.content
1334. 1334 );
1335. 1335 }
1336. 1336
1337. 1337 #[tokio::test]
1338. 1338 async fn office_output_is_digest_bound_and_pages_a_huge_paragraph() {
1339. 1339 let giant: String = (0..400_000u64)
1340. 1340 .map(|index| format!("w{} ", index.wrapping_mul(2_654_435_761) % 1_000_003))
1341. 1341 .collect();
1342. 1342 let document = format!(
1343. 1343 r#"<w:document xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main"><w:body><w:p><w:r><w:t>{giant}</w:t></w:r></w:p><w:p><w:r><w:t>tail</w:t></w:r></w:p></w:body></w:document>"#
1344. 1344 );
1345. 1345 let package = vak_ooxml::fixtures::word_with(
1346. 1346 vak_ooxml::fixtures::WORD_MAIN,
1347. 1347 &document,
1348. 1348 &[],
1349. 1349 &[],
1350. 1350 &[],
1351. 1351 &[],
1352. 1352 );
1353. 1353 let output = read_office("big.docx", package, serde_json::json!({"limit": 500})).await;
1354. 1354 assert!(!output.is_error, "{}", output.content);
1355. 1355 assert!(
1356. 1356 output.content.contains("· sha256 "),
1357. 1357 "anchors are bound to a digest"
1358. 1358 );
1359. 1359 assert!(
1360. 1360 output.content.contains("[p@1 ⟨part 1 of "),
1361. 1361 "a huge unit is split, not cut"
1362. 1362 );
1363. 1363 assert!(
1364. 1364 output.content.len() <= super::PAGE_BYTES + 4096,
1365. 1365 "a page stays inside the budget: {} bytes",
1366. 1366 output.content.len()
1367. 1367 );
1368. 1368 assert!(output.content.contains("continue with offset="));
1369. 1369 assert!(
1370. 1370 !output.content.contains("[p@2] tail"),
1371. 1371 "the rest waits for the next page"
1372. 1372 );
1373. 1373 }
1374. 1374}
1375. 1375