@sixb/agent 0.2.3 → 0.2.5

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (149) hide show
  1. package/README.md +3 -3
  2. package/dist/agent-prompt.d.ts.map +1 -1
  3. package/dist/artifacts/layout-check/layout-check.d.ts +21 -24
  4. package/dist/artifacts/layout-check/layout-check.d.ts.map +1 -1
  5. package/dist/artifacts/publish-tool.d.ts +26 -44
  6. package/dist/artifacts/publish-tool.d.ts.map +1 -1
  7. package/dist/ask-tool.d.ts +62 -0
  8. package/dist/ask-tool.d.ts.map +1 -0
  9. package/dist/attachments.d.ts +1 -1
  10. package/dist/attachments.d.ts.map +1 -1
  11. package/dist/background-commands.d.ts +3 -3
  12. package/dist/background-commands.d.ts.map +1 -1
  13. package/dist/background-notices.d.ts.map +1 -1
  14. package/dist/child-files.d.ts +1 -1
  15. package/dist/child-questions.d.ts +1 -1
  16. package/dist/index.d.ts +1 -2
  17. package/dist/index.d.ts.map +1 -1
  18. package/dist/index.js +10457 -7971
  19. package/dist/index.js.map +46 -38
  20. package/dist/run-agent-turn.d.ts +3 -14
  21. package/dist/run-agent-turn.d.ts.map +1 -1
  22. package/dist/run-conversation-turn.d.ts.map +1 -1
  23. package/dist/run-environment.d.ts +4 -1
  24. package/dist/run-environment.d.ts.map +1 -1
  25. package/dist/run-subagent.d.ts +1 -1
  26. package/dist/run-subagent.d.ts.map +1 -1
  27. package/dist/sandbox-api-context.d.ts.map +1 -1
  28. package/dist/search-script.d.ts +7 -0
  29. package/dist/search-script.d.ts.map +1 -0
  30. package/dist/skills/built-in.d.ts.map +1 -1
  31. package/dist/subagent-tools.d.ts +20 -1
  32. package/dist/subagent-tools.d.ts.map +1 -1
  33. package/dist/tools/background.d.ts +50 -18
  34. package/dist/tools/background.d.ts.map +1 -1
  35. package/dist/tools/bash.d.ts +1 -0
  36. package/dist/tools/bash.d.ts.map +1 -1
  37. package/dist/tools/edit-text.d.ts +22 -0
  38. package/dist/tools/edit-text.d.ts.map +1 -0
  39. package/dist/tools/edit.d.ts +54 -0
  40. package/dist/tools/edit.d.ts.map +1 -0
  41. package/dist/tools/file-writes.d.ts +52 -0
  42. package/dist/tools/file-writes.d.ts.map +1 -0
  43. package/dist/tools/glob.d.ts +41 -0
  44. package/dist/tools/glob.d.ts.map +1 -0
  45. package/dist/tools/grep.d.ts +64 -0
  46. package/dist/tools/grep.d.ts.map +1 -0
  47. package/dist/tools/model-spec.d.ts.map +1 -1
  48. package/dist/tools/read.d.ts +16 -3
  49. package/dist/tools/read.d.ts.map +1 -1
  50. package/dist/tools/sandbox-search.d.ts +26 -0
  51. package/dist/tools/sandbox-search.d.ts.map +1 -0
  52. package/dist/tools/search/command.d.ts +15 -0
  53. package/dist/tools/search/command.d.ts.map +1 -0
  54. package/dist/tools/search/gitignore.d.ts +28 -0
  55. package/dist/tools/search/gitignore.d.ts.map +1 -0
  56. package/dist/tools/search/glob.d.ts +15 -0
  57. package/dist/tools/search/glob.d.ts.map +1 -0
  58. package/dist/tools/search/main.d.ts +2 -0
  59. package/dist/tools/search/main.d.ts.map +1 -0
  60. package/dist/tools/search/match.d.ts +30 -0
  61. package/dist/tools/search/match.d.ts.map +1 -0
  62. package/dist/tools/search/pattern.d.ts +19 -0
  63. package/dist/tools/search/pattern.d.ts.map +1 -0
  64. package/dist/tools/search/read.d.ts +62 -0
  65. package/dist/tools/search/read.d.ts.map +1 -0
  66. package/dist/tools/search/request.d.ts +80 -0
  67. package/dist/tools/search/request.d.ts.map +1 -0
  68. package/dist/tools/search/search.d.ts +4 -0
  69. package/dist/tools/search/search.d.ts.map +1 -0
  70. package/dist/tools/search/walk.d.ts +43 -0
  71. package/dist/tools/search/walk.d.ts.map +1 -0
  72. package/dist/tools/tool-outputs.d.ts.map +1 -1
  73. package/dist/tools/view-file.d.ts +20 -23
  74. package/dist/tools/view-file.d.ts.map +1 -1
  75. package/dist/tools/web-fetch/tool.d.ts +1 -0
  76. package/dist/tools/web-fetch/tool.d.ts.map +1 -1
  77. package/dist/tools/web-search.d.ts +1 -0
  78. package/dist/tools/web-search.d.ts.map +1 -1
  79. package/dist/tools/write.d.ts +36 -0
  80. package/dist/tools/write.d.ts.map +1 -0
  81. package/dist/turn-inbox.d.ts +75 -0
  82. package/dist/turn-inbox.d.ts.map +1 -0
  83. package/package.json +6 -4
  84. package/src/agent-prompt.ts +10 -9
  85. package/src/agent-runtime/preflight.ts +1 -0
  86. package/src/artifacts/layout-check/check.cjs +176 -95
  87. package/src/artifacts/layout-check/layout-check.ts +118 -76
  88. package/src/artifacts/publish-tool.ts +123 -192
  89. package/src/ask-tool.ts +223 -0
  90. package/src/attachments.ts +7 -6
  91. package/src/background-commands.ts +5 -5
  92. package/src/background-notices.ts +5 -12
  93. package/src/child-files.ts +1 -1
  94. package/src/child-questions.ts +1 -1
  95. package/src/context-compaction.ts +4 -1
  96. package/src/generated/search.mjs +985 -0
  97. package/src/index.ts +1 -2
  98. package/src/run-agent-turn.ts +99 -133
  99. package/src/run-child.ts +1 -1
  100. package/src/run-conversation-turn.ts +8 -5
  101. package/src/run-environment.ts +40 -21
  102. package/src/run-subagent.ts +4 -4
  103. package/src/sandbox-api-context.ts +3 -0
  104. package/src/search-script-assets.d.ts +5 -0
  105. package/src/search-script.ts +9 -0
  106. package/src/skills/built-in.ts +16 -0
  107. package/src/skills/sixb-artifacts/SKILL.md +13 -13
  108. package/src/skills/sixb-artifacts/canvas.md +5 -4
  109. package/src/skills/sixb-artifacts/document.md +92 -5
  110. package/src/skills/sixb-artifacts/examples/document/service-proposal/design.css +114 -0
  111. package/src/skills/sixb-artifacts/examples/document/service-proposal/document.html +159 -0
  112. package/src/skills/sixb-artifacts/examples/document/service-proposal/images/fieldwise.svg +1 -0
  113. package/src/skills/sixb-artifacts/examples/document/service-proposal/images/harbor-pine.svg +1 -0
  114. package/src/skills/sixb-artifacts/page.md +2 -2
  115. package/src/skills/sixb-artifacts/queries.md +2 -1
  116. package/src/skills/sixb-artifacts/slides.md +2 -2
  117. package/src/subagent-tools.ts +115 -81
  118. package/src/tools/background.ts +227 -67
  119. package/src/tools/bash.ts +2 -1
  120. package/src/tools/edit-text.ts +241 -0
  121. package/src/tools/edit.ts +166 -0
  122. package/src/tools/file-writes.ts +270 -0
  123. package/src/tools/glob.ts +124 -0
  124. package/src/tools/grep.ts +190 -0
  125. package/src/tools/model-spec.ts +13 -13
  126. package/src/tools/read.ts +75 -9
  127. package/src/tools/sandbox-search.ts +219 -0
  128. package/src/tools/search/command.ts +116 -0
  129. package/src/tools/search/gitignore.ts +101 -0
  130. package/src/tools/search/glob.ts +90 -0
  131. package/src/tools/search/main.ts +9 -0
  132. package/src/tools/search/match.ts +278 -0
  133. package/src/tools/search/pattern.ts +178 -0
  134. package/src/tools/search/read.ts +237 -0
  135. package/src/tools/search/request.ts +101 -0
  136. package/src/tools/search/search.ts +297 -0
  137. package/src/tools/search/walk.ts +145 -0
  138. package/src/tools/tool-outputs.ts +9 -4
  139. package/src/tools/view-file.ts +78 -80
  140. package/src/tools/web-fetch/tool.ts +2 -1
  141. package/src/tools/web-search.ts +1 -0
  142. package/src/tools/write.ts +95 -0
  143. package/src/turn-inbox.ts +283 -0
  144. package/dist/artifacts/board-previews.d.ts +0 -41
  145. package/dist/artifacts/board-previews.d.ts.map +0 -1
  146. package/dist/ask-parent-tool.d.ts +0 -36
  147. package/dist/ask-parent-tool.d.ts.map +0 -1
  148. package/src/artifacts/board-previews.ts +0 -218
  149. package/src/ask-parent-tool.ts +0 -100
@@ -0,0 +1,116 @@
1
+ import {
2
+ type GlobSearchRequest,
3
+ type GrepSearchRequest,
4
+ SearchError,
5
+ type SearchRequest,
6
+ } from "./request"
7
+ import { globFiles, grepFiles } from "./search"
8
+
9
+ const USAGE = "Usage: search.mjs <request JSON>"
10
+
11
+ const KEYS: Readonly<Record<SearchRequest["tool"], readonly string[]>> = {
12
+ glob: ["tool", "pattern", "path", "contextDirectory", "timeoutMs"],
13
+ grep: ["tool", "pattern", "path", "include", "literal", "contextDirectory", "timeoutMs"],
14
+ }
15
+
16
+ /** What the process prints, and the code it exits with. */
17
+ export interface SearchCommandResult {
18
+ readonly exitCode: number
19
+ readonly stdout: string
20
+ readonly stderr: string
21
+ }
22
+
23
+ /**
24
+ * `search.mjs '<request JSON>'`: the Agent's `glob` and `grep`, the same in every sandbox. The
25
+ * request is one JSON argument, so no shell quoting reaches a pattern; the result is one JSON
26
+ * line. A search that cannot run exits 2 with `{"error":{"code","message"}}` on stderr: a request
27
+ * it does not understand is `invalid_arguments`, and a search the request names wrongly has its
28
+ * own code ({@link SearchError}). Anything else exits 3, as `internal_error`.
29
+ */
30
+ export async function runSearchCommand(
31
+ args: readonly string[],
32
+ workingDirectory: string
33
+ ): Promise<SearchCommandResult> {
34
+ try {
35
+ const request = parseRequest(args)
36
+ const result =
37
+ request.tool === "glob"
38
+ ? globFiles(request, workingDirectory)
39
+ : await grepFiles(request, workingDirectory)
40
+
41
+ return { exitCode: 0, stdout: `${JSON.stringify(result)}\n`, stderr: "" }
42
+ } catch (error) {
43
+ if (error instanceof SearchError) return failure(2, error.code, error.message)
44
+ if (error instanceof RequestError) return failure(2, "invalid_arguments", error.message)
45
+
46
+ const message = error instanceof Error ? error.message : "The search failed unexpectedly."
47
+ return failure(3, "internal_error", message)
48
+ }
49
+ }
50
+
51
+ class RequestError extends Error {}
52
+
53
+ function failure(exitCode: number, code: string, message: string): SearchCommandResult {
54
+ return { exitCode, stdout: "", stderr: `${JSON.stringify({ error: { code, message } })}\n` }
55
+ }
56
+
57
+ function fail(message: string): never {
58
+ throw new RequestError(message)
59
+ }
60
+
61
+ function parseRequest(args: readonly string[]): SearchRequest {
62
+ if (args.length !== 1) fail(USAGE)
63
+
64
+ let value: unknown
65
+ try {
66
+ value = JSON.parse(args[0] as string)
67
+ } catch {
68
+ fail("The search request is not valid JSON.")
69
+ }
70
+ if (typeof value !== "object" || value === null || Array.isArray(value)) {
71
+ fail("The search request must be a JSON object.")
72
+ }
73
+
74
+ const input = value as Record<string, unknown>
75
+ const tool = input.tool
76
+ if (tool !== "glob" && tool !== "grep") fail("tool must be 'glob' or 'grep'.")
77
+ for (const key of Object.keys(input)) {
78
+ if (!KEYS[tool].includes(key)) fail(`The ${tool} request has an unknown property '${key}'.`)
79
+ }
80
+
81
+ const pattern = nonEmptyString(input.pattern, "pattern")
82
+ const timeoutMs = input.timeoutMs
83
+ if (typeof timeoutMs !== "number" || !Number.isSafeInteger(timeoutMs) || timeoutMs < 0) {
84
+ fail("timeoutMs must be a non-negative integer.")
85
+ }
86
+ const path = optionalString(input.path, "path")
87
+ const contextDirectory = optionalString(input.contextDirectory, "contextDirectory")
88
+ const scope = {
89
+ timeoutMs,
90
+ ...(path === undefined ? {} : { path }),
91
+ ...(contextDirectory === undefined ? {} : { contextDirectory }),
92
+ }
93
+
94
+ if (tool === "glob") return { tool, pattern, ...scope } satisfies GlobSearchRequest
95
+
96
+ if (input.literal !== undefined && typeof input.literal !== "boolean") {
97
+ fail("literal must be true or false.")
98
+ }
99
+ const include = optionalString(input.include, "include")
100
+ return {
101
+ tool,
102
+ pattern,
103
+ ...scope,
104
+ ...(include === undefined ? {} : { include }),
105
+ ...(input.literal === undefined ? {} : { literal: input.literal }),
106
+ } satisfies GrepSearchRequest
107
+ }
108
+
109
+ function nonEmptyString(value: unknown, field: string): string {
110
+ if (typeof value !== "string" || value === "") fail(`${field} must be a non-empty string.`)
111
+ return value
112
+ }
113
+
114
+ function optionalString(value: unknown, field: string): string | undefined {
115
+ return value === undefined ? undefined : nonEmptyString(value, field)
116
+ }
@@ -0,0 +1,101 @@
1
+ import { compilePattern, PatternError, pathPatternSource, segmentSource } from "./pattern"
2
+
3
+ /** One line of a `.gitignore`. */
4
+ export interface IgnoreRule {
5
+ /** `!pattern`: brings back what an earlier rule excluded. */
6
+ readonly negated: boolean
7
+ /** `pattern/`: matches folders only. */
8
+ readonly folderOnly: boolean
9
+ /** No `/` but a trailing one: matches a name at any depth below the file's folder. */
10
+ readonly anyDepth: boolean
11
+ readonly regex: RegExp
12
+ }
13
+
14
+ /** A `.gitignore` and the folder it applies to, an absolute path. */
15
+ export interface IgnoreFile {
16
+ readonly folder: string
17
+ readonly rules: readonly IgnoreRule[]
18
+ }
19
+
20
+ /**
21
+ * Parse a `.gitignore` as git reads it. A line that is not a valid pattern is skipped, as git
22
+ * skips it.
23
+ */
24
+ export function parseGitignore(text: string): IgnoreRule[] {
25
+ const rules: IgnoreRule[] = []
26
+
27
+ // Git drops a UTF-8 byte order mark, which would otherwise join the first pattern.
28
+ const body = text.startsWith("\uFEFF") ? text.slice(1) : text
29
+
30
+ for (const rawLine of body.split("\n")) {
31
+ const rule = parseLine(rawLine.endsWith("\r") ? rawLine.slice(0, -1) : rawLine)
32
+ if (rule) rules.push(rule)
33
+ }
34
+
35
+ return rules
36
+ }
37
+
38
+ function parseLine(rawLine: string): IgnoreRule | undefined {
39
+ // `\#` and `\!` stay escaped: the pattern compiler reads them as the characters themselves.
40
+ if (rawLine === "" || rawLine.startsWith("#")) return undefined
41
+
42
+ let line = withoutTrailingSpaces(rawLine)
43
+ const negated = line.startsWith("!")
44
+ if (negated) line = line.slice(1)
45
+
46
+ const folderOnly = line.endsWith("/")
47
+ line = line.replace(/\/+$/, "")
48
+
49
+ // A `/` at the start or in the middle anchors the pattern to the file's folder.
50
+ const anyDepth = !line.includes("/")
51
+ line = line.replace(/^\//, "")
52
+ if (!line) return undefined
53
+
54
+ try {
55
+ const regex = compilePattern(anyDepth ? segmentSource(line) : pathPatternSource(line))
56
+ return { negated, folderOnly, anyDepth, regex }
57
+ } catch (error) {
58
+ if (error instanceof PatternError) return undefined
59
+ throw error
60
+ }
61
+ }
62
+
63
+ /** Trailing spaces are dropped, unless a backslash escapes them. */
64
+ function withoutTrailingSpaces(line: string): string {
65
+ let end = line.length
66
+
67
+ while (end > 0 && line[end - 1] === " ") {
68
+ let backslashes = 0
69
+ while (line[end - 2 - backslashes] === "\\") backslashes += 1
70
+ if (backslashes % 2 === 1) break
71
+ end -= 1
72
+ }
73
+
74
+ return line.slice(0, end)
75
+ }
76
+
77
+ /**
78
+ * Whether the `.gitignore` files that apply exclude an entry, given as an absolute path. Files
79
+ * come from the outermost folder in, and a later rule wins over an earlier one, so a deeper file
80
+ * overrides its parents. What lies inside an excluded folder is never asked about: the walk does
81
+ * not enter it, so a rule cannot bring it back, as in git.
82
+ */
83
+ export function isIgnored(files: readonly IgnoreFile[], path: string, isFolder: boolean): boolean {
84
+ const name = path.slice(path.lastIndexOf("/") + 1)
85
+ let ignored = false
86
+
87
+ for (const file of files) {
88
+ const prefix = file.folder.endsWith("/") ? file.folder : `${file.folder}/`
89
+ if (!path.startsWith(prefix)) continue
90
+ const relative = path.slice(prefix.length)
91
+
92
+ for (const rule of file.rules) {
93
+ if (rule.folderOnly && !isFolder) continue
94
+ // Only a rule that would change the verdict so far needs testing.
95
+ if (rule.negated !== ignored) continue
96
+ if (rule.regex.test(rule.anyDepth ? name : relative)) ignored = !rule.negated
97
+ }
98
+ }
99
+
100
+ return ignored
101
+ }
@@ -0,0 +1,90 @@
1
+ import {
2
+ compilePattern,
3
+ expandBraces,
4
+ PatternError,
5
+ pathPatternSource,
6
+ segmentSource,
7
+ } from "./pattern"
8
+
9
+ /**
10
+ * A `glob` pattern, matched against paths relative to where the search starts. A pattern without
11
+ * `/` matches a name at any depth (`*.md` finds every Markdown file), as in `.gitignore`; one with
12
+ * `/` matches from the start, with `**` for any number of folders.
13
+ */
14
+ export interface GlobMatcher {
15
+ matches(relativePath: string): boolean
16
+ /** Whether a folder, relative to where the search starts, may hold a match. */
17
+ mayContain(relativeFolder: string): boolean
18
+ /** A segment starts with `.`: the pattern looks for hidden files. */
19
+ readonly namesHidden: boolean
20
+ readonly namesNodeModules: boolean
21
+ }
22
+
23
+ interface Alternative {
24
+ /** Matched against the name alone, at any depth. */
25
+ readonly anyDepth: boolean
26
+ readonly regex: RegExp
27
+ readonly segments: readonly string[]
28
+ }
29
+
30
+ export function compileGlob(pattern: string): GlobMatcher {
31
+ const alternatives = expandBraces(pattern).map(compileAlternative)
32
+ const segments = alternatives.flatMap((alternative) => alternative.segments)
33
+ const folderSegments = new Map<string, RegExp>()
34
+
35
+ // One folder segment against one pattern segment, compiled once for the whole walk.
36
+ const segmentMatches = (patternSegment: string, name: string): boolean => {
37
+ let regex = folderSegments.get(patternSegment)
38
+ if (!regex) {
39
+ regex = compilePattern(segmentSource(patternSegment))
40
+ folderSegments.set(patternSegment, regex)
41
+ }
42
+ return regex.test(name)
43
+ }
44
+
45
+ return {
46
+ matches(relativePath) {
47
+ const name = relativePath.slice(relativePath.lastIndexOf("/") + 1)
48
+ return alternatives.some((alternative) =>
49
+ alternative.regex.test(alternative.anyDepth ? name : relativePath)
50
+ )
51
+ },
52
+
53
+ mayContain(relativeFolder) {
54
+ const names = relativeFolder.split("/")
55
+ return alternatives.some((alternative) => {
56
+ if (alternative.anyDepth) return true
57
+
58
+ for (const [index, name] of names.entries()) {
59
+ const segment = alternative.segments[index]
60
+ if (segment === undefined) return false
61
+ if (segment === "**") return true
62
+ if (!segmentMatches(segment, name)) return false
63
+ }
64
+ // The folder matched as far as it goes: the pattern must still name what is inside.
65
+ return alternative.segments.length > names.length
66
+ })
67
+ },
68
+
69
+ namesHidden: segments.some((segment) => segment.startsWith(".")),
70
+ namesNodeModules: segments.some((segment) => segment.includes("node_modules")),
71
+ }
72
+ }
73
+
74
+ function compileAlternative(pattern: string): Alternative {
75
+ let relative = pattern
76
+ while (relative.startsWith("./")) relative = relative.slice(2)
77
+
78
+ // A leading `/` anchors the pattern where the search starts, as in `.gitignore`.
79
+ const anchored = relative.startsWith("/")
80
+ if (anchored) relative = relative.replace(/^\/+/, "")
81
+ if (!relative) throw new PatternError("it names no file")
82
+
83
+ const anyDepth = !anchored && !relative.includes("/")
84
+ const segments = relative.split("/")
85
+ return {
86
+ anyDepth,
87
+ regex: compilePattern(anyDepth ? segmentSource(relative) : pathPatternSource(relative)),
88
+ segments,
89
+ }
90
+ }
@@ -0,0 +1,9 @@
1
+ // The engine of the Agent's `glob` and `grep`, as a run installs it at `lib/search.mjs` and runs
2
+ // it on Bun or Node: `bun run generate:search-script` bundles it into src/generated/search.mjs.
3
+ import { runSearchCommand } from "./command"
4
+
5
+ const result = await runSearchCommand(process.argv.slice(2), process.cwd())
6
+
7
+ process.stdout.write(result.stdout)
8
+ process.stderr.write(result.stderr)
9
+ process.exitCode = result.exitCode
@@ -0,0 +1,278 @@
1
+ import { SEARCH_LIMITS, SearchError } from "./request"
2
+
3
+ /** Where a line first matched `grep`'s pattern, in UTF-16 units. */
4
+ export interface LineMatch {
5
+ readonly index: number
6
+ readonly length: number
7
+ }
8
+
9
+ export interface TextMatcher {
10
+ /**
11
+ * Where a line first matches, over its first {@link SEARCH_LIMITS.matchCharacters} characters.
12
+ */
13
+ line(line: string): LineMatch | undefined
14
+ /**
15
+ * Whether any line of a file may match: false only when none does, so a file that cannot match
16
+ * is never split into lines. It may say true of a file with no matching line.
17
+ */
18
+ mayMatch(text: string): boolean
19
+ }
20
+
21
+ /**
22
+ * `grep`'s pattern, matched line by line. Case is smart: ignored when the pattern has no uppercase
23
+ * letter. Literal text is a plain substring search, which no line can slow down.
24
+ */
25
+ export function textMatcher(pattern: string, literal: boolean): TextMatcher {
26
+ return literal ? literalMatcher(pattern) : regexMatcher(pattern)
27
+ }
28
+
29
+ function literalMatcher(pattern: string): TextMatcher {
30
+ const ignoreCase = !/\p{Lu}/u.test(pattern)
31
+ const needle = ignoreCase ? pattern.toLowerCase() : pattern
32
+
33
+ return {
34
+ line(line) {
35
+ const matched = linePrefix(line, SEARCH_LIMITS.matchCharacters)
36
+ const index = (ignoreCase ? matched.toLowerCase() : matched).indexOf(needle)
37
+
38
+ // Lowercasing lengthens a few characters (İ): the index is then near the match, not on it.
39
+ return index < 0 ? undefined : { index: Math.min(index, line.length), length: needle.length }
40
+ },
41
+
42
+ // A line's text is part of the file's. Lowercasing is the same character by character, but
43
+ // for Σ, which reads what follows it: only where a line is cut at 4,000 characters does that
44
+ // differ from the file's text, and a file with such a line is searched line by line.
45
+ mayMatch(text) {
46
+ if (!ignoreCase) return text.includes(needle)
47
+ if (hasLineOver(text, SEARCH_LIMITS.matchCharacters)) return true
48
+
49
+ return text.toLowerCase().includes(needle)
50
+ },
51
+ }
52
+ }
53
+
54
+ function regexMatcher(pattern: string): TextMatcher {
55
+ const regex = lineRegex(pattern)
56
+ const whole = wholeTextRegex(pattern, regex)
57
+
58
+ return {
59
+ line(line) {
60
+ const found = regex.exec(linePrefix(line, SEARCH_LIMITS.matchCharacters))
61
+ return found ? { index: found.index, length: found[0].length } : undefined
62
+ },
63
+
64
+ mayMatch(text) {
65
+ if (!whole || !linesAreWhole(text)) return true
66
+ return whole.test(text)
67
+ },
68
+ }
69
+ }
70
+
71
+ /**
72
+ * Unicode mode first, so `\p{…}` works; without it when that refuses what plain mode accepts.
73
+ * Case is ignored when the pattern writes no uppercase letter and no property escape: `\p{Lu}`
74
+ * ignoring case would match every cased letter.
75
+ */
76
+ function lineRegex(pattern: string): RegExp {
77
+ const written = pattern.replace(/\\(?:[pP]\{[^}]*\}|.)/g, "")
78
+ const flags = /\p{Lu}/u.test(written) || /\\[pP]\{/.test(pattern) ? "" : "i"
79
+
80
+ try {
81
+ return new RegExp(pattern, `${flags}u`)
82
+ } catch {}
83
+
84
+ try {
85
+ return new RegExp(pattern, flags)
86
+ } catch (error) {
87
+ const message = error instanceof Error ? error.message : String(error)
88
+ throw new SearchError(
89
+ "invalid_pattern",
90
+ message.replace(/^Invalid regular expression: (?:\/.*\/[a-z]*: )?/, "")
91
+ )
92
+ }
93
+ }
94
+
95
+ /**
96
+ * The line regex with `m`, to test a whole file at once; nothing for a pattern written with a
97
+ * lookaround or a modifier group, or that can match a line break. The test stands in for the lines
98
+ * only on a file whose lines are whole ({@link linesAreWhole}).
99
+ *
100
+ * Why a line's match is then a match in the whole text: the line is a stretch of it, and the
101
+ * same characters match the same way. The pattern consumes no line break, so a match never
102
+ * leaves the line: what can still differ is what its assertions read.
103
+ *
104
+ * - `^` and `$`: with `m`, they hold where a line starts or ends, `\n` and `\r\n` alike. With no
105
+ * lone `\r` and no U+2028 or U+2029 in the file, those are exactly a line's edges: in the line
106
+ * and in the whole text, they hold at the same places.
107
+ * - `\b` and `\B`: past a line's edge, the whole text holds `\n` or `\r`, which is no word
108
+ * character, as nothing is: they read the same at a line's edges.
109
+ * - Backreferences repeat what their group matched in the line.
110
+ * - A lookaround could say no where an assertion says yes, and is rare in a search: a pattern
111
+ * with one is searched line by line, so nothing above rests on how one reads. A modifier group
112
+ * changes the flags themselves: `(?-m:^)` turns `m` back off.
113
+ *
114
+ * A match in the whole text that no line holds, across lines, is never reported: it only costs a
115
+ * file split into lines.
116
+ *
117
+ * Why the test stays fast: a line bounds what each try can read, and with it how far a regex
118
+ * backtracks. A pattern that can match a line break backtracks over the whole file instead —
119
+ * `[^x]*y` takes seconds on 100 KB — which is the other reason it is searched line by line.
120
+ */
121
+ function wholeTextRegex(pattern: string, line: RegExp): RegExp | undefined {
122
+ if (!readsOneLine(pattern)) return undefined
123
+ return new RegExp(line.source, `${line.flags}m`)
124
+ }
125
+
126
+ /**
127
+ * Whether the line breaks `m` sees are the lines `grep` splits, each matched whole: no line is
128
+ * cut at {@link SEARCH_LIMITS.matchCharacters}, and no lone `\r`, U+2028 or U+2029, which end a
129
+ * line for `^` and `$` but stay inside one for {@link lines}.
130
+ */
131
+ function linesAreWhole(text: string): boolean {
132
+ return !hasLineOver(text, SEARCH_LIMITS.matchCharacters) && !hasLoneBreak(text)
133
+ }
134
+
135
+ /** A `\r` not followed by `\n`, U+2028 or U+2029. Searched for as text, faster than a regex. */
136
+ function hasLoneBreak(text: string): boolean {
137
+ if (text.includes("\u2028") || text.includes("\u2029")) return true
138
+
139
+ for (let index = text.indexOf("\r"); index >= 0; index = text.indexOf("\r", index + 1)) {
140
+ if (text.charCodeAt(index + 1) !== LINE_FEED) return true
141
+ }
142
+
143
+ return false
144
+ }
145
+
146
+ const LINE_FEED = 0x0a
147
+
148
+ /** Escapes that never match a line break, outside a class and inside one. */
149
+ const ESCAPES = new Set(["b", "B", "d", "w", "S", "k"])
150
+ const CLASS_ESCAPES = new Set(["d", "w", "S"])
151
+
152
+ /**
153
+ * Whether a pattern, as written, has no lookaround, no modifier group, and cannot match `\n`,
154
+ * `\r`, U+2028 or U+2029. What it cannot rule out says no:
155
+ *
156
+ * - `(?=`, `(?!`, `(?<=`, `(?<!`, and `(?i:`, `(?m:`, `(?s:`, `(?-…:` outside a class.
157
+ * - `.` never matches a line break, without the `s` flag, nor do `\d`, `\w` and `\S`.
158
+ * - A negated class does, unless it names them: `[^…]` says no.
159
+ * - `\s`, `\D`, `\W`, `\n`, `\r`, `\t`, `\v`, `\f`, `\0`, `\x…`, `\u…`, `\c…`, `\p{…}` and digit
160
+ * escapes (an octal `\12` is `\n` without the `u` flag) all may, alone or as a range's end.
161
+ * - So may a control character or a line separator written as itself, escaped or not, and a
162
+ * range that spans one.
163
+ * - `\b` and `\B` are assertions, and `\k<name>` repeats a group's text; in a class, `\b` is a
164
+ * backspace, a control character.
165
+ */
166
+ function readsOneLine(pattern: string): boolean {
167
+ let inClass = false
168
+ /** The class's last character, a range's start. */
169
+ let previous: number | undefined
170
+
171
+ for (let index = 0; index < pattern.length; index += 1) {
172
+ let code = pattern.charCodeAt(index)
173
+ if (isLineBreakOrControl(code)) return false
174
+
175
+ if (code === BACKSLASH) {
176
+ index += 1
177
+ const escaped = pattern[index] ?? ""
178
+ if (/[A-Za-z0-9]/.test(escaped)) {
179
+ if (!(inClass ? CLASS_ESCAPES : ESCAPES).has(escaped)) return false
180
+ previous = undefined
181
+ continue
182
+ }
183
+
184
+ // An escaped symbol is the symbol itself.
185
+ code = pattern.charCodeAt(index)
186
+ if (isLineBreakOrControl(code)) return false
187
+ } else if (!inClass && code === OPEN_PAREN) {
188
+ if (LOOKAROUND_OR_MODIFIER.test(pattern.slice(index, index + 4))) return false
189
+ } else if (!inClass && code === OPEN_BRACKET) {
190
+ if (pattern[index + 1] === "^") return false
191
+ inClass = true
192
+ previous = undefined
193
+ continue
194
+ } else if (inClass && code === CLOSE_BRACKET) {
195
+ inClass = false
196
+ continue
197
+ } else if (inClass && code === DASH && previous !== undefined && index + 1 < pattern.length) {
198
+ const end =
199
+ pattern[index + 1] === "\\" ? pattern.charCodeAt(index + 2) : pattern.charCodeAt(index + 1)
200
+ if (previous <= 0x2029 && end >= 0x2028) return false
201
+ }
202
+
203
+ previous = inClass ? code : undefined
204
+ }
205
+
206
+ return true
207
+ }
208
+
209
+ /** At a `(`: a lookaround, or a group that changes the `i`, `m` or `s` flag. */
210
+ const LOOKAROUND_OR_MODIFIER = /^\(\?(?:[=!]|<[=!]|[ims-])/
211
+
212
+ const OPEN_PAREN = 0x28
213
+ const BACKSLASH = 0x5c
214
+ const OPEN_BRACKET = 0x5b
215
+ const CLOSE_BRACKET = 0x5d
216
+ const DASH = 0x2d
217
+
218
+ function isLineBreakOrControl(code: number): boolean {
219
+ return code < 0x20 || code === 0x2028 || code === 0x2029
220
+ }
221
+
222
+ /** Whether a line of the text, its `\r` included, is longer than `length` UTF-16 units. */
223
+ function hasLineOver(text: string, length: number): boolean {
224
+ if (text.length <= length) return false
225
+
226
+ for (let start = 0; start <= text.length; ) {
227
+ const newline = text.indexOf("\n", start)
228
+ const end = newline < 0 ? text.length : newline
229
+ if (end - start > length) return true
230
+ start = end + 1
231
+ }
232
+
233
+ return false
234
+ }
235
+
236
+ /** A file's lines, without their `\n` or `\r\n`. */
237
+ export function lines(text: string): string[] {
238
+ const split = text.split("\n")
239
+ if (split.at(-1) === "") split.pop()
240
+ return split.map((line) => (line.endsWith("\r") ? line.slice(0, -1) : line))
241
+ }
242
+
243
+ /**
244
+ * A line as `grep` shows it: whole when it fits, else cut to {@link SEARCH_LIMITS.lineCharacters}
245
+ * characters around its first match, so a match far into a minified line is still seen. A cut
246
+ * never splits a character written as two units.
247
+ */
248
+ export function cutLine(line: string, match: LineMatch): string {
249
+ const budget = SEARCH_LIMITS.lineCharacters
250
+ if (line.length <= budget) return line
251
+
252
+ // The window opens a little before the match, or ends with the line, the leading `…` in the
253
+ // budget; a match within the line's first characters keeps its start.
254
+ const room = budget - 1
255
+ const start =
256
+ match.index + match.length <= budget
257
+ ? 0
258
+ : Math.max(0, Math.min(match.index - SEARCH_LIMITS.matchLead, line.length - room))
259
+ if (start === 0) return `${linePrefix(line, budget)}…`
260
+
261
+ const from = isLowSurrogate(line.charCodeAt(start)) ? start + 1 : start
262
+ // One unit past the room, so a character the room would split is seen and left out.
263
+ const window = linePrefix(line.slice(from, from + room + 1), room)
264
+
265
+ return `…${window}${from + window.length < line.length ? "…" : ""}`
266
+ }
267
+
268
+ function isLowSurrogate(code: number): boolean {
269
+ return code >= 0xdc00 && code <= 0xdfff
270
+ }
271
+
272
+ /** A line's first `length` UTF-16 units, never half of a character written as two. */
273
+ function linePrefix(line: string, length: number): string {
274
+ if (line.length <= length) return line
275
+
276
+ const code = line.charCodeAt(length - 1)
277
+ return line.slice(0, code >= 0xd800 && code <= 0xdbff ? length - 1 : length)
278
+ }