Migration
This is my new public git source viewer. Source coming soon. Please contact with concerns/bugs.
Contact me
internal/modules/structured.go
d658fcc4ea9f3bb562fd00dce8f863934560c56c
· 14.7 KB · 516 lines
raw
| 1 | package modules |
| 2 | |
| 3 | import ( |
| 4 | "bytes" |
| 5 | "encoding/json" |
| 6 | "html/template" |
| 7 | "regexp" |
| 8 | "sort" |
| 9 | "strconv" |
| 10 | "strings" |
| 11 | |
| 12 | "gopkg.in/yaml.v3" |
| 13 | ) |
| 14 | |
| 15 | // ParseNote reports why a JSON/YAML blob failed to parse, so the blob view |
| 16 | // can point at the offending line instead of leaving a reader to guess. |
| 17 | // Line is 0 when the underlying error can't be pinned to one. |
| 18 | type ParseNote struct { |
| 19 | Line int |
| 20 | Msg string |
| 21 | } |
| 22 | |
| 23 | // Anchor is a YAML anchor declared on a line, along with how many alias |
| 24 | // sites in the file point back at it. |
| 25 | type Anchor struct { |
| 26 | Name string |
| 27 | Uses int |
| 28 | } |
| 29 | |
| 30 | // StructRef is a resolved YAML alias: which anchor it names, the line that |
| 31 | // anchor is declared on, and the value it stands for rendered back to YAML. |
| 32 | // One per line -- a flow sequence like [*a, *b] resolves only its first. |
| 33 | type StructRef struct { |
| 34 | Name string |
| 35 | Line int |
| 36 | Preview string |
| 37 | } |
| 38 | |
| 39 | // maxStructuredSize bounds the work one file may cost the structured view: |
| 40 | // a full parse on the server, and a pass over every row in the browser at |
| 41 | // load. It matches maxHighlightSize on purpose -- past that point the file |
| 42 | // is already served as plain, unhighlighted text with no indent guides, and |
| 43 | // folding a file that has neither would be a strange half-feature. Half a |
| 44 | // megabyte is far above any hand-written config and below the size where a |
| 45 | // modest laptop notices the per-row work. |
| 46 | const maxStructuredSize = maxHighlightSize |
| 47 | |
| 48 | // structuredLang returns "json" or "yaml" for a recognised config |
| 49 | // extension, "" otherwise. The blob view only spends the extra parse and |
| 50 | // indent-guide pass on formats where nesting depth is the whole point. |
| 51 | func structuredLang(path string) string { |
| 52 | p := strings.ToLower(path) |
| 53 | switch { |
| 54 | case strings.HasSuffix(p, ".json"): |
| 55 | return "json" |
| 56 | case strings.HasSuffix(p, ".yaml"), strings.HasSuffix(p, ".yml"): |
| 57 | return "yaml" |
| 58 | default: |
| 59 | return "" |
| 60 | } |
| 61 | } |
| 62 | |
| 63 | func validateStructured(lang string, raw []byte) *ParseNote { |
| 64 | var err error |
| 65 | switch lang { |
| 66 | case "json": |
| 67 | err = json.Unmarshal(raw, new(any)) |
| 68 | case "yaml": |
| 69 | err = yaml.Unmarshal(raw, new(any)) |
| 70 | default: |
| 71 | return nil |
| 72 | } |
| 73 | if err == nil { |
| 74 | return nil |
| 75 | } |
| 76 | return &ParseNote{Line: parseErrorLine(lang, raw, err), Msg: err.Error()} |
| 77 | } |
| 78 | |
| 79 | var yamlLineRe = regexp.MustCompile(`line (\d+)`) |
| 80 | |
| 81 | func parseErrorLine(lang string, raw []byte, err error) int { |
| 82 | if lang == "json" { |
| 83 | switch e := err.(type) { |
| 84 | case *json.SyntaxError: |
| 85 | return bytes.Count(raw[:e.Offset], []byte("\n")) + 1 |
| 86 | case *json.UnmarshalTypeError: |
| 87 | return bytes.Count(raw[:e.Offset], []byte("\n")) + 1 |
| 88 | } |
| 89 | return 0 |
| 90 | } |
| 91 | // yaml.v3 errors are plain strings of the form "yaml: line N: ...", or |
| 92 | // for TypeError one such string per bad field -- take the first. |
| 93 | if m := yamlLineRe.FindStringSubmatch(err.Error()); m != nil { |
| 94 | n, _ := strconv.Atoi(m[1]) |
| 95 | return n |
| 96 | } |
| 97 | return 0 |
| 98 | } |
| 99 | |
| 100 | // indentUnit finds the smallest non-zero run of leading spaces across |
| 101 | // non-blank lines, so a 4-space file gets one guide per level and a |
| 102 | // 2-space file doesn't get a spurious extra one. |
| 103 | func indentUnit(raw []byte) int { |
| 104 | unit := 0 |
| 105 | for _, line := range bytes.Split(raw, []byte("\n")) { |
| 106 | n := 0 |
| 107 | for n < len(line) && line[n] == ' ' { |
| 108 | n++ |
| 109 | } |
| 110 | if n == 0 || n == len(line) { |
| 111 | continue // blank or all-whitespace |
| 112 | } |
| 113 | if unit == 0 || n < unit { |
| 114 | unit = n |
| 115 | } |
| 116 | } |
| 117 | if unit == 0 { |
| 118 | return 2 |
| 119 | } |
| 120 | return unit |
| 121 | } |
| 122 | |
| 123 | // annotateStructure fills in the per-line structure the blob view needs to |
| 124 | // fold, filter and cross-reference a config file: nesting depth, the extent |
| 125 | // of the block each line opens, the mapping key it declares, and (YAML only) |
| 126 | // the anchors and aliases tying parts of the file together. |
| 127 | // |
| 128 | // It works off raw rather than the highlighted HTML: indentation is the one |
| 129 | // signal that means the same thing in both languages, and it survives |
| 130 | // minified or oddly-formatted input by simply finding nothing to fold. |
| 131 | func annotateStructure(lines []Line, raw []byte, lang string) { |
| 132 | src := bytes.Split(raw, []byte("\n")) |
| 133 | unit := indentUnit(raw) |
| 134 | |
| 135 | ind := make([]int, len(src)) |
| 136 | for i, l := range src { |
| 137 | ind[i] = lineIndent(l) |
| 138 | } |
| 139 | ends := foldEnds(ind) |
| 140 | |
| 141 | keyRe := jsonKeyRe |
| 142 | if lang == "yaml" { |
| 143 | keyRe = yamlKeyRe |
| 144 | } |
| 145 | for i := range src { |
| 146 | if i >= len(lines) { |
| 147 | break |
| 148 | } |
| 149 | if ind[i] >= 0 { |
| 150 | lines[i].Depth = ind[i] / unit |
| 151 | if e := ends[i]; e > 0 && e <= len(lines) { |
| 152 | lines[i].FoldTo = e |
| 153 | } |
| 154 | } |
| 155 | if m := keyRe.FindSubmatch(src[i]); m != nil { |
| 156 | for _, g := range m[1:] { |
| 157 | if len(g) > 0 { |
| 158 | lines[i].Key = string(g) |
| 159 | break |
| 160 | } |
| 161 | } |
| 162 | } |
| 163 | } |
| 164 | if lang != "yaml" { |
| 165 | annotateJSONPaths(lines, raw) |
| 166 | return |
| 167 | } |
| 168 | anchors, refs, paths := yamlAnnotations(src) |
| 169 | for n, a := range anchors { |
| 170 | if n >= 1 && n <= len(lines) { |
| 171 | lines[n-1].Anchor = a |
| 172 | } |
| 173 | } |
| 174 | for n, r := range refs { |
| 175 | if n >= 1 && n <= len(lines) { |
| 176 | lines[n-1].Ref = r |
| 177 | } |
| 178 | } |
| 179 | for n, p := range paths { |
| 180 | if n >= 1 && n <= len(lines) { |
| 181 | lines[n-1].Path = p |
| 182 | } |
| 183 | } |
| 184 | } |
| 185 | |
| 186 | // lineIndent counts leading spaces, or returns -1 for a line that carries no |
| 187 | // structure (blank or whitespace-only) so folding steps over it rather than |
| 188 | // treating column 0 as a dedent. |
| 189 | func lineIndent(line []byte) int { |
| 190 | n := 0 |
| 191 | for n < len(line) && line[n] == ' ' { |
| 192 | n++ |
| 193 | } |
| 194 | if n == len(bytes.TrimRight(line, " \t\r")) { |
| 195 | return -1 |
| 196 | } |
| 197 | return n |
| 198 | } |
| 199 | |
| 200 | // foldEnds returns, per line, the 1-based number of the last line belonging |
| 201 | // to the block that line opens -- 0 when it opens nothing. A line opens a |
| 202 | // block when the next line carrying structure is indented further, and the |
| 203 | // block runs until indentation returns to that line's own level or less. |
| 204 | // Blank lines between siblings are absorbed; blank lines trailing a block |
| 205 | // are not. |
| 206 | func foldEnds(ind []int) []int { |
| 207 | ends := make([]int, len(ind)) |
| 208 | type frame struct { |
| 209 | ind, idx int |
| 210 | child bool |
| 211 | } |
| 212 | var stack []frame |
| 213 | prev := -1 // last line carrying structure |
| 214 | for i, in := range ind { |
| 215 | if in < 0 { |
| 216 | continue |
| 217 | } |
| 218 | for len(stack) > 0 && stack[len(stack)-1].ind >= in { |
| 219 | if t := stack[len(stack)-1]; t.child { |
| 220 | ends[t.idx] = prev + 1 |
| 221 | } |
| 222 | stack = stack[:len(stack)-1] |
| 223 | } |
| 224 | if len(stack) > 0 { |
| 225 | stack[len(stack)-1].child = true |
| 226 | } |
| 227 | stack = append(stack, frame{ind: in, idx: i}) |
| 228 | prev = i |
| 229 | } |
| 230 | for _, t := range stack { |
| 231 | if t.child { |
| 232 | ends[t.idx] = prev + 1 |
| 233 | } |
| 234 | } |
| 235 | return ends |
| 236 | } |
| 237 | |
| 238 | var ( |
| 239 | // The key a line declares, used to scope search to keys rather than |
| 240 | // whole lines. Both are deliberately shallow: a miss costs a narrower |
| 241 | // search, not a wrong render. |
| 242 | yamlKeyRe = regexp.MustCompile(`^\s*(?:-\s+)?(?:"([^"]*)"|'([^']*)'|([^\s#"'][^:#]*?))\s*:(?:\s|$)`) |
| 243 | jsonKeyRe = regexp.MustCompile(`^\s*"((?:[^"\\]|\\.)*)"\s*:`) |
| 244 | ) |
| 245 | |
| 246 | // yamlAnnotations resolves every alias in the file to the line its anchor is |
| 247 | // declared on, keyed by 1-based line number, and records each line's |
| 248 | // canonical value path. Anchors are per-document, so the whole stream is |
| 249 | // decoded document by document -- a k8s manifest that defines its anchors in |
| 250 | // the second document would otherwise come back empty. A malformed document |
| 251 | // ends the walk with whatever was found first; the parse-error banner |
| 252 | // already reports the failure itself. |
| 253 | func yamlAnnotations(src [][]byte) (map[int]*Anchor, map[int]*StructRef, map[int]string) { |
| 254 | anchors := map[int]*Anchor{} |
| 255 | refs := map[int]*StructRef{} |
| 256 | paths := map[int]string{} |
| 257 | dec := yaml.NewDecoder(bytes.NewReader(bytes.Join(src, []byte("\n")))) |
| 258 | for { |
| 259 | var doc yaml.Node |
| 260 | if err := dec.Decode(&doc); err != nil { |
| 261 | break |
| 262 | } |
| 263 | walkYAML(&doc, src, anchors, refs, paths, "") |
| 264 | } |
| 265 | return anchors, refs, paths |
| 266 | } |
| 267 | |
| 268 | func walkYAML(n *yaml.Node, src [][]byte, anchors map[int]*Anchor, refs map[int]*StructRef, paths map[int]string, path string) { |
| 269 | if n.Anchor != "" { |
| 270 | if line := anchorLine(src, n.Anchor, n.Line); line > 0 { |
| 271 | if _, ok := anchors[line]; !ok { |
| 272 | anchors[line] = &Anchor{Name: n.Anchor} |
| 273 | } |
| 274 | } |
| 275 | } |
| 276 | if n.Kind == yaml.AliasNode && n.Alias != nil { |
| 277 | target := anchorLine(src, n.Value, n.Alias.Line) |
| 278 | if target > 0 { |
| 279 | if _, ok := refs[n.Line]; !ok { |
| 280 | refs[n.Line] = &StructRef{Name: n.Value, Line: target, Preview: nodePreview(n.Alias)} |
| 281 | } |
| 282 | if a, ok := anchors[target]; ok { |
| 283 | a.Uses++ |
| 284 | } |
| 285 | } |
| 286 | // An alias displays only *name here; the aliased value's lines are |
| 287 | // wherever the anchor was declared. Recursing into n.Alias would |
| 288 | // emit paths pointing at those other lines, so record this line's |
| 289 | // path and stop. |
| 290 | recordPath(paths, n.Line, path) |
| 291 | return |
| 292 | } |
| 293 | |
| 294 | switch n.Kind { |
| 295 | case yaml.DocumentNode: |
| 296 | for _, c := range n.Content { |
| 297 | walkYAML(c, src, anchors, refs, paths, path) |
| 298 | } |
| 299 | case yaml.MappingNode: |
| 300 | for i := 0; i+1 < len(n.Content); i += 2 { |
| 301 | k, v := n.Content[i], n.Content[i+1] |
| 302 | child := joinPath(path, k.Value) |
| 303 | if k.Value == "<<" { |
| 304 | // A merge key is not a property of the mapping, and the |
| 305 | // rendered text never expands it, so a `<<` segment would |
| 306 | // name something the reader cannot see. Keep the parent. |
| 307 | child = path |
| 308 | } |
| 309 | // Walk the key too (anchors may be declared on one), but record |
| 310 | // the value's path last so it wins for the shared line. |
| 311 | walkYAML(k, src, anchors, refs, paths, path) |
| 312 | recordPath(paths, k.Line, child) |
| 313 | walkYAML(v, src, anchors, refs, paths, child) |
| 314 | } |
| 315 | case yaml.SequenceNode: |
| 316 | for i, c := range n.Content { |
| 317 | child := joinIndex(path, i) |
| 318 | recordPath(paths, c.Line, child) |
| 319 | walkYAML(c, src, anchors, refs, paths, child) |
| 320 | } |
| 321 | default: |
| 322 | recordPath(paths, n.Line, path) |
| 323 | } |
| 324 | } |
| 325 | |
| 326 | // anchorLine locates the line the "&name" token sits on. A node's own Line |
| 327 | // is where its *value* begins, which for a block mapping is the line after |
| 328 | // the key the anchor was written on -- so scan back from there for the |
| 329 | // token itself and point at that instead. |
| 330 | func anchorLine(src [][]byte, name string, at int) int { |
| 331 | tok := []byte("&" + name) |
| 332 | if at > len(src) { |
| 333 | at = len(src) |
| 334 | } |
| 335 | for i := at - 1; i >= 0 && i > at-64; i-- { |
| 336 | j := bytes.Index(src[i], tok) |
| 337 | if j < 0 { |
| 338 | continue |
| 339 | } |
| 340 | if k := j + len(tok); k < len(src[i]) && isAnchorChar(src[i][k]) { |
| 341 | continue // &namelonger |
| 342 | } |
| 343 | return i + 1 |
| 344 | } |
| 345 | return 0 |
| 346 | } |
| 347 | |
| 348 | func isAnchorChar(b byte) bool { |
| 349 | return b == '-' || b == '_' || b == '.' || |
| 350 | (b >= '0' && b <= '9') || (b >= 'a' && b <= 'z') || (b >= 'A' && b <= 'Z') |
| 351 | } |
| 352 | |
| 353 | // nodePreview renders what an alias actually stands for, short enough to sit |
| 354 | // in a tooltip. The anchor name is stripped from the copy so the preview |
| 355 | // reads as the value, not as its declaration. |
| 356 | func nodePreview(n *yaml.Node) string { |
| 357 | c := *n |
| 358 | c.Anchor = "" |
| 359 | var buf bytes.Buffer |
| 360 | enc := yaml.NewEncoder(&buf) |
| 361 | enc.SetIndent(2) // yaml.Marshal's default of 4 doesn't match most files |
| 362 | if err := enc.Encode(&c); err != nil { |
| 363 | return "" |
| 364 | } |
| 365 | enc.Close() |
| 366 | s := strings.TrimRight(buf.String(), "\n") |
| 367 | if ls := strings.Split(s, "\n"); len(ls) > 12 { |
| 368 | s = strings.Join(append(ls[:12], "…"), "\n") |
| 369 | } |
| 370 | if len(s) > 480 { |
| 371 | s = strings.ToValidUTF8(s[:480], "") + "…" |
| 372 | } |
| 373 | return s |
| 374 | } |
| 375 | |
| 376 | // withIndentGuides prefixes each line's leading spaces with one guide span |
| 377 | // per nesting level, so deeply nested JSON/YAML reads like a tree instead |
| 378 | // of a wall of spaces. Leading whitespace is always plain, unwrapped text |
| 379 | // at this point (see highlightLines' Text/TextWhitespace handling), so |
| 380 | // trimming it back off the rendered HTML here is safe. |
| 381 | func withIndentGuides(lines []template.HTML, unit int) []template.HTML { |
| 382 | out := make([]template.HTML, len(lines)) |
| 383 | for i, l := range lines { |
| 384 | s := string(l) |
| 385 | n := 0 |
| 386 | for n < len(s) && s[n] == ' ' { |
| 387 | n++ |
| 388 | } |
| 389 | if n < unit { |
| 390 | out[i] = l |
| 391 | continue |
| 392 | } |
| 393 | var b strings.Builder |
| 394 | levels, rem := n/unit, n%unit |
| 395 | for j := 0; j < levels; j++ { |
| 396 | b.WriteString(`<span class="ind" style="width:`) |
| 397 | b.WriteString(strconv.Itoa(unit)) |
| 398 | b.WriteString(`ch"></span>`) |
| 399 | } |
| 400 | if rem > 0 { |
| 401 | b.WriteString(strings.Repeat(" ", rem)) |
| 402 | } |
| 403 | b.WriteString(s[n:]) |
| 404 | out[i] = template.HTML(b.String()) |
| 405 | } |
| 406 | return out |
| 407 | } |
| 408 | |
| 409 | // --------------------------------------------------------------- paths |
| 410 | |
| 411 | // A path query names a value the way obj1.property3.items[4].val does. |
| 412 | // Both the server (here) and static/js/structured.js write the same |
| 413 | // canonical form so a query can be compared as a string: simple keys are |
| 414 | // dotted, indices attach directly, and anything else is bracket-quoted. |
| 415 | var ( |
| 416 | simplePathKey = regexp.MustCompile(`^[A-Za-z0-9_$-]+$`) |
| 417 | pathEscaper = strings.NewReplacer(`\`, `\\`, `"`, `\"`) |
| 418 | ) |
| 419 | |
| 420 | func joinPath(parent, key string) string { |
| 421 | if simplePathKey.MatchString(key) { |
| 422 | if parent == "" { |
| 423 | return key |
| 424 | } |
| 425 | return parent + "." + key |
| 426 | } |
| 427 | return parent + `["` + pathEscaper.Replace(key) + `"]` |
| 428 | } |
| 429 | |
| 430 | func joinIndex(parent string, i int) string { |
| 431 | return parent + "[" + strconv.Itoa(i) + "]" |
| 432 | } |
| 433 | |
| 434 | // recordPath keeps the last path recorded for a line. A line can carry more |
| 435 | // than one path (JSON/YAML flow style puts several values on one row), and |
| 436 | // the last one parsed is the deepest -- the one a reader naming a full path |
| 437 | // is most likely to mean. A container query still reaches its children by |
| 438 | // the boundary-aware prefix match in structured.js, so dropping the shallow |
| 439 | // path costs nothing there. |
| 440 | func recordPath(paths map[int]string, line int, path string) { |
| 441 | if path == "" || line <= 0 { |
| 442 | return |
| 443 | } |
| 444 | paths[line] = path |
| 445 | } |
| 446 | |
| 447 | // annotateJSONPaths records each line's canonical value path. It reuses |
| 448 | // encoding/json's own decoder rather than a second parser, so the offsets it |
| 449 | // sees are exactly the ones the format defines. |
| 450 | func annotateJSONPaths(lines []Line, raw []byte) { |
| 451 | paths := jsonPaths(raw) |
| 452 | for i := range lines { |
| 453 | if p, ok := paths[lines[i].N]; ok { |
| 454 | lines[i].Path = p |
| 455 | } |
| 456 | } |
| 457 | } |
| 458 | |
| 459 | func jsonPaths(raw []byte) map[int]string { |
| 460 | paths := map[int]string{} |
| 461 | lineOf := newLineIndex(raw) |
| 462 | dec := json.NewDecoder(bytes.NewReader(raw)) |
| 463 | |
| 464 | // walk consumes one JSON value at path. Objects record the path on the |
| 465 | // key's line and recurse into the value; arrays record it on each |
| 466 | // element's line. A malformed document simply stops -- the parse-error |
| 467 | // banner owns reporting that. |
| 468 | var walk func(path string) |
| 469 | walk = func(path string) { |
| 470 | tok, err := dec.Token() |
| 471 | if err != nil { |
| 472 | return |
| 473 | } |
| 474 | recordPath(paths, lineOf(dec.InputOffset()), path) |
| 475 | d, ok := tok.(json.Delim) |
| 476 | if !ok { |
| 477 | return |
| 478 | } |
| 479 | switch d { |
| 480 | case '{': |
| 481 | for dec.More() { |
| 482 | kt, err := dec.Token() |
| 483 | if err != nil { |
| 484 | return |
| 485 | } |
| 486 | key, _ := kt.(string) |
| 487 | child := joinPath(path, key) |
| 488 | recordPath(paths, lineOf(dec.InputOffset()), child) |
| 489 | walk(child) |
| 490 | } |
| 491 | dec.Token() // consume '}' |
| 492 | case '[': |
| 493 | for i := 0; dec.More(); i++ { |
| 494 | walk(joinIndex(path, i)) |
| 495 | } |
| 496 | dec.Token() // consume ']' |
| 497 | } |
| 498 | } |
| 499 | walk("") |
| 500 | return paths |
| 501 | } |
| 502 | |
| 503 | // newLineIndex maps a byte offset to a 1-based line number. Newline |
| 504 | // positions are found once and binary-searched per token, so indexing a |
| 505 | // whole file stays O(n log n) rather than counting from zero each time. |
| 506 | func newLineIndex(raw []byte) func(int64) int { |
| 507 | var newlines []int64 |
| 508 | for i, b := range raw { |
| 509 | if b == '\n' { |
| 510 | newlines = append(newlines, int64(i)) |
| 511 | } |
| 512 | } |
| 513 | return func(off int64) int { |
| 514 | return sort.Search(len(newlines), func(i int) bool { return newlines[i] >= off }) + 1 |
| 515 | } |
| 516 | } |