github.com/flyinox/gosm@v0.0.0-20171117061539-16768cb62077/src/go/printer/printer.go (about) 1 // Copyright 2009 The Go Authors. All rights reserved. 2 // Use of this source code is governed by a BSD-style 3 // license that can be found in the LICENSE file. 4 5 // Package printer implements printing of AST nodes. 6 package printer 7 8 import ( 9 "fmt" 10 "go/ast" 11 "go/token" 12 "io" 13 "os" 14 "strconv" 15 "strings" 16 "text/tabwriter" 17 "unicode" 18 ) 19 20 const ( 21 maxNewlines = 2 // max. number of newlines between source text 22 debug = false // enable for debugging 23 infinity = 1 << 30 24 ) 25 26 type whiteSpace byte 27 28 const ( 29 ignore = whiteSpace(0) 30 blank = whiteSpace(' ') 31 vtab = whiteSpace('\v') 32 newline = whiteSpace('\n') 33 formfeed = whiteSpace('\f') 34 indent = whiteSpace('>') 35 unindent = whiteSpace('<') 36 ) 37 38 // A pmode value represents the current printer mode. 39 type pmode int 40 41 const ( 42 noExtraBlank pmode = 1 << iota // disables extra blank after /*-style comment 43 noExtraLinebreak // disables extra line break after /*-style comment 44 ) 45 46 type commentInfo struct { 47 cindex int // current comment index 48 comment *ast.CommentGroup // = printer.comments[cindex]; or nil 49 commentOffset int // = printer.posFor(printer.comments[cindex].List[0].Pos()).Offset; or infinity 50 commentNewline bool // true if the comment group contains newlines 51 } 52 53 type printer struct { 54 // Configuration (does not change after initialization) 55 Config 56 fset *token.FileSet 57 58 // Current state 59 output []byte // raw printer result 60 indent int // current indentation 61 level int // level == 0: outside composite literal; level > 0: inside composite literal 62 mode pmode // current printer mode 63 impliedSemi bool // if set, a linebreak implies a semicolon 64 lastTok token.Token // last token printed (token.ILLEGAL if it's whitespace) 65 prevOpen token.Token // previous non-brace "open" token (, [, or token.ILLEGAL 66 wsbuf []whiteSpace // delayed white space 67 68 // Positions 69 // The out position differs from the pos position when the result 70 // formatting differs from the source formatting (in the amount of 71 // white space). If there's a difference and SourcePos is set in 72 // ConfigMode, //line directives are used in the output to restore 73 // original source positions for a reader. 74 pos token.Position // current position in AST (source) space 75 out token.Position // current position in output space 76 last token.Position // value of pos after calling writeString 77 linePtr *int // if set, record out.Line for the next token in *linePtr 78 79 // The list of all source comments, in order of appearance. 80 comments []*ast.CommentGroup // may be nil 81 useNodeComments bool // if not set, ignore lead and line comments of nodes 82 83 // Information about p.comments[p.cindex]; set up by nextComment. 84 commentInfo 85 86 // Cache of already computed node sizes. 87 nodeSizes map[ast.Node]int 88 89 // Cache of most recently computed line position. 90 cachedPos token.Pos 91 cachedLine int // line corresponding to cachedPos 92 } 93 94 func (p *printer) init(cfg *Config, fset *token.FileSet, nodeSizes map[ast.Node]int) { 95 p.Config = *cfg 96 p.fset = fset 97 p.pos = token.Position{Line: 1, Column: 1} 98 p.out = token.Position{Line: 1, Column: 1} 99 p.wsbuf = make([]whiteSpace, 0, 16) // whitespace sequences are short 100 p.nodeSizes = nodeSizes 101 p.cachedPos = -1 102 } 103 104 func (p *printer) internalError(msg ...interface{}) { 105 if debug { 106 fmt.Print(p.pos.String() + ": ") 107 fmt.Println(msg...) 108 panic("go/printer") 109 } 110 } 111 112 // commentsHaveNewline reports whether a list of comments belonging to 113 // an *ast.CommentGroup contains newlines. Because the position information 114 // may only be partially correct, we also have to read the comment text. 115 func (p *printer) commentsHaveNewline(list []*ast.Comment) bool { 116 // len(list) > 0 117 line := p.lineFor(list[0].Pos()) 118 for i, c := range list { 119 if i > 0 && p.lineFor(list[i].Pos()) != line { 120 // not all comments on the same line 121 return true 122 } 123 if t := c.Text; len(t) >= 2 && (t[1] == '/' || strings.Contains(t, "\n")) { 124 return true 125 } 126 } 127 _ = line 128 return false 129 } 130 131 func (p *printer) nextComment() { 132 for p.cindex < len(p.comments) { 133 c := p.comments[p.cindex] 134 p.cindex++ 135 if list := c.List; len(list) > 0 { 136 p.comment = c 137 p.commentOffset = p.posFor(list[0].Pos()).Offset 138 p.commentNewline = p.commentsHaveNewline(list) 139 return 140 } 141 // we should not reach here (correct ASTs don't have empty 142 // ast.CommentGroup nodes), but be conservative and try again 143 } 144 // no more comments 145 p.commentOffset = infinity 146 } 147 148 // commentBefore reports whether the current comment group occurs 149 // before the next position in the source code and printing it does 150 // not introduce implicit semicolons. 151 // 152 func (p *printer) commentBefore(next token.Position) bool { 153 return p.commentOffset < next.Offset && (!p.impliedSemi || !p.commentNewline) 154 } 155 156 // commentSizeBefore returns the estimated size of the 157 // comments on the same line before the next position. 158 // 159 func (p *printer) commentSizeBefore(next token.Position) int { 160 // save/restore current p.commentInfo (p.nextComment() modifies it) 161 defer func(info commentInfo) { 162 p.commentInfo = info 163 }(p.commentInfo) 164 165 size := 0 166 for p.commentBefore(next) { 167 for _, c := range p.comment.List { 168 size += len(c.Text) 169 } 170 p.nextComment() 171 } 172 return size 173 } 174 175 // recordLine records the output line number for the next non-whitespace 176 // token in *linePtr. It is used to compute an accurate line number for a 177 // formatted construct, independent of pending (not yet emitted) whitespace 178 // or comments. 179 // 180 func (p *printer) recordLine(linePtr *int) { 181 p.linePtr = linePtr 182 } 183 184 // linesFrom returns the number of output lines between the current 185 // output line and the line argument, ignoring any pending (not yet 186 // emitted) whitespace or comments. It is used to compute an accurate 187 // size (in number of lines) for a formatted construct. 188 // 189 func (p *printer) linesFrom(line int) int { 190 return p.out.Line - line 191 } 192 193 func (p *printer) posFor(pos token.Pos) token.Position { 194 // not used frequently enough to cache entire token.Position 195 return p.fset.Position(pos) 196 } 197 198 func (p *printer) lineFor(pos token.Pos) int { 199 if pos != p.cachedPos { 200 p.cachedPos = pos 201 p.cachedLine = p.fset.Position(pos).Line 202 } 203 return p.cachedLine 204 } 205 206 // writeLineDirective writes a //line directive if necessary. 207 func (p *printer) writeLineDirective(pos token.Position) { 208 if pos.IsValid() && (p.out.Line != pos.Line || p.out.Filename != pos.Filename) { 209 p.output = append(p.output, tabwriter.Escape) // protect '\n' in //line from tabwriter interpretation 210 p.output = append(p.output, fmt.Sprintf("//line %s:%d\n", pos.Filename, pos.Line)...) 211 p.output = append(p.output, tabwriter.Escape) 212 // p.out must match the //line directive 213 p.out.Filename = pos.Filename 214 p.out.Line = pos.Line 215 } 216 } 217 218 // writeIndent writes indentation. 219 func (p *printer) writeIndent() { 220 // use "hard" htabs - indentation columns 221 // must not be discarded by the tabwriter 222 n := p.Config.Indent + p.indent // include base indentation 223 for i := 0; i < n; i++ { 224 p.output = append(p.output, '\t') 225 } 226 227 // update positions 228 p.pos.Offset += n 229 p.pos.Column += n 230 p.out.Column += n 231 } 232 233 // writeByte writes ch n times to p.output and updates p.pos. 234 // Only used to write formatting (white space) characters. 235 func (p *printer) writeByte(ch byte, n int) { 236 if p.out.Column == 1 { 237 // no need to write line directives before white space 238 p.writeIndent() 239 } 240 241 for i := 0; i < n; i++ { 242 p.output = append(p.output, ch) 243 } 244 245 // update positions 246 p.pos.Offset += n 247 if ch == '\n' || ch == '\f' { 248 p.pos.Line += n 249 p.out.Line += n 250 p.pos.Column = 1 251 p.out.Column = 1 252 return 253 } 254 p.pos.Column += n 255 p.out.Column += n 256 } 257 258 // writeString writes the string s to p.output and updates p.pos, p.out, 259 // and p.last. If isLit is set, s is escaped w/ tabwriter.Escape characters 260 // to protect s from being interpreted by the tabwriter. 261 // 262 // Note: writeString is only used to write Go tokens, literals, and 263 // comments, all of which must be written literally. Thus, it is correct 264 // to always set isLit = true. However, setting it explicitly only when 265 // needed (i.e., when we don't know that s contains no tabs or line breaks) 266 // avoids processing extra escape characters and reduces run time of the 267 // printer benchmark by up to 10%. 268 // 269 func (p *printer) writeString(pos token.Position, s string, isLit bool) { 270 if p.out.Column == 1 { 271 if p.Config.Mode&SourcePos != 0 { 272 p.writeLineDirective(pos) 273 } 274 p.writeIndent() 275 } 276 277 if pos.IsValid() { 278 // update p.pos (if pos is invalid, continue with existing p.pos) 279 // Note: Must do this after handling line beginnings because 280 // writeIndent updates p.pos if there's indentation, but p.pos 281 // is the position of s. 282 p.pos = pos 283 } 284 285 if isLit { 286 // Protect s such that is passes through the tabwriter 287 // unchanged. Note that valid Go programs cannot contain 288 // tabwriter.Escape bytes since they do not appear in legal 289 // UTF-8 sequences. 290 p.output = append(p.output, tabwriter.Escape) 291 } 292 293 if debug { 294 p.output = append(p.output, fmt.Sprintf("/*%s*/", pos)...) // do not update p.pos! 295 } 296 p.output = append(p.output, s...) 297 298 // update positions 299 nlines := 0 300 var li int // index of last newline; valid if nlines > 0 301 for i := 0; i < len(s); i++ { 302 // Go tokens cannot contain '\f' - no need to look for it 303 if s[i] == '\n' { 304 nlines++ 305 li = i 306 } 307 } 308 p.pos.Offset += len(s) 309 if nlines > 0 { 310 p.pos.Line += nlines 311 p.out.Line += nlines 312 c := len(s) - li 313 p.pos.Column = c 314 p.out.Column = c 315 } else { 316 p.pos.Column += len(s) 317 p.out.Column += len(s) 318 } 319 320 if isLit { 321 p.output = append(p.output, tabwriter.Escape) 322 } 323 324 p.last = p.pos 325 } 326 327 // writeCommentPrefix writes the whitespace before a comment. 328 // If there is any pending whitespace, it consumes as much of 329 // it as is likely to help position the comment nicely. 330 // pos is the comment position, next the position of the item 331 // after all pending comments, prev is the previous comment in 332 // a group of comments (or nil), and tok is the next token. 333 // 334 func (p *printer) writeCommentPrefix(pos, next token.Position, prev *ast.Comment, tok token.Token) { 335 if len(p.output) == 0 { 336 // the comment is the first item to be printed - don't write any whitespace 337 return 338 } 339 340 if pos.IsValid() && pos.Filename != p.last.Filename { 341 // comment in a different file - separate with newlines 342 p.writeByte('\f', maxNewlines) 343 return 344 } 345 346 if pos.Line == p.last.Line && (prev == nil || prev.Text[1] != '/') { 347 // comment on the same line as last item: 348 // separate with at least one separator 349 hasSep := false 350 if prev == nil { 351 // first comment of a comment group 352 j := 0 353 for i, ch := range p.wsbuf { 354 switch ch { 355 case blank: 356 // ignore any blanks before a comment 357 p.wsbuf[i] = ignore 358 continue 359 case vtab: 360 // respect existing tabs - important 361 // for proper formatting of commented structs 362 hasSep = true 363 continue 364 case indent: 365 // apply pending indentation 366 continue 367 } 368 j = i 369 break 370 } 371 p.writeWhitespace(j) 372 } 373 // make sure there is at least one separator 374 if !hasSep { 375 sep := byte('\t') 376 if pos.Line == next.Line { 377 // next item is on the same line as the comment 378 // (which must be a /*-style comment): separate 379 // with a blank instead of a tab 380 sep = ' ' 381 } 382 p.writeByte(sep, 1) 383 } 384 385 } else { 386 // comment on a different line: 387 // separate with at least one line break 388 droppedLinebreak := false 389 j := 0 390 for i, ch := range p.wsbuf { 391 switch ch { 392 case blank, vtab: 393 // ignore any horizontal whitespace before line breaks 394 p.wsbuf[i] = ignore 395 continue 396 case indent: 397 // apply pending indentation 398 continue 399 case unindent: 400 // if this is not the last unindent, apply it 401 // as it is (likely) belonging to the last 402 // construct (e.g., a multi-line expression list) 403 // and is not part of closing a block 404 if i+1 < len(p.wsbuf) && p.wsbuf[i+1] == unindent { 405 continue 406 } 407 // if the next token is not a closing }, apply the unindent 408 // if it appears that the comment is aligned with the 409 // token; otherwise assume the unindent is part of a 410 // closing block and stop (this scenario appears with 411 // comments before a case label where the comments 412 // apply to the next case instead of the current one) 413 if tok != token.RBRACE && pos.Column == next.Column { 414 continue 415 } 416 case newline, formfeed: 417 p.wsbuf[i] = ignore 418 droppedLinebreak = prev == nil // record only if first comment of a group 419 } 420 j = i 421 break 422 } 423 p.writeWhitespace(j) 424 425 // determine number of linebreaks before the comment 426 n := 0 427 if pos.IsValid() && p.last.IsValid() { 428 n = pos.Line - p.last.Line 429 if n < 0 { // should never happen 430 n = 0 431 } 432 } 433 434 // at the package scope level only (p.indent == 0), 435 // add an extra newline if we dropped one before: 436 // this preserves a blank line before documentation 437 // comments at the package scope level (issue 2570) 438 if p.indent == 0 && droppedLinebreak { 439 n++ 440 } 441 442 // make sure there is at least one line break 443 // if the previous comment was a line comment 444 if n == 0 && prev != nil && prev.Text[1] == '/' { 445 n = 1 446 } 447 448 if n > 0 { 449 // use formfeeds to break columns before a comment; 450 // this is analogous to using formfeeds to separate 451 // individual lines of /*-style comments 452 p.writeByte('\f', nlimit(n)) 453 } 454 } 455 } 456 457 // Returns true if s contains only white space 458 // (only tabs and blanks can appear in the printer's context). 459 // 460 func isBlank(s string) bool { 461 for i := 0; i < len(s); i++ { 462 if s[i] > ' ' { 463 return false 464 } 465 } 466 return true 467 } 468 469 // commonPrefix returns the common prefix of a and b. 470 func commonPrefix(a, b string) string { 471 i := 0 472 for i < len(a) && i < len(b) && a[i] == b[i] && (a[i] <= ' ' || a[i] == '*') { 473 i++ 474 } 475 return a[0:i] 476 } 477 478 // trimRight returns s with trailing whitespace removed. 479 func trimRight(s string) string { 480 return strings.TrimRightFunc(s, unicode.IsSpace) 481 } 482 483 // stripCommonPrefix removes a common prefix from /*-style comment lines (unless no 484 // comment line is indented, all but the first line have some form of space prefix). 485 // The prefix is computed using heuristics such that is likely that the comment 486 // contents are nicely laid out after re-printing each line using the printer's 487 // current indentation. 488 // 489 func stripCommonPrefix(lines []string) { 490 if len(lines) <= 1 { 491 return // at most one line - nothing to do 492 } 493 // len(lines) > 1 494 495 // The heuristic in this function tries to handle a few 496 // common patterns of /*-style comments: Comments where 497 // the opening /* and closing */ are aligned and the 498 // rest of the comment text is aligned and indented with 499 // blanks or tabs, cases with a vertical "line of stars" 500 // on the left, and cases where the closing */ is on the 501 // same line as the last comment text. 502 503 // Compute maximum common white prefix of all but the first, 504 // last, and blank lines, and replace blank lines with empty 505 // lines (the first line starts with /* and has no prefix). 506 // In cases where only the first and last lines are not blank, 507 // such as two-line comments, or comments where all inner lines 508 // are blank, consider the last line for the prefix computation 509 // since otherwise the prefix would be empty. 510 // 511 // Note that the first and last line are never empty (they 512 // contain the opening /* and closing */ respectively) and 513 // thus they can be ignored by the blank line check. 514 prefix := "" 515 prefixSet := false 516 if len(lines) > 2 { 517 for i, line := range lines[1 : len(lines)-1] { 518 if isBlank(line) { 519 lines[1+i] = "" // range starts with lines[1] 520 } else { 521 if !prefixSet { 522 prefix = line 523 prefixSet = true 524 } 525 prefix = commonPrefix(prefix, line) 526 } 527 528 } 529 } 530 // If we don't have a prefix yet, consider the last line. 531 if !prefixSet { 532 line := lines[len(lines)-1] 533 prefix = commonPrefix(line, line) 534 } 535 536 /* 537 * Check for vertical "line of stars" and correct prefix accordingly. 538 */ 539 lineOfStars := false 540 if i := strings.Index(prefix, "*"); i >= 0 { 541 // Line of stars present. 542 if i > 0 && prefix[i-1] == ' ' { 543 i-- // remove trailing blank from prefix so stars remain aligned 544 } 545 prefix = prefix[0:i] 546 lineOfStars = true 547 } else { 548 // No line of stars present. 549 // Determine the white space on the first line after the /* 550 // and before the beginning of the comment text, assume two 551 // blanks instead of the /* unless the first character after 552 // the /* is a tab. If the first comment line is empty but 553 // for the opening /*, assume up to 3 blanks or a tab. This 554 // whitespace may be found as suffix in the common prefix. 555 first := lines[0] 556 if isBlank(first[2:]) { 557 // no comment text on the first line: 558 // reduce prefix by up to 3 blanks or a tab 559 // if present - this keeps comment text indented 560 // relative to the /* and */'s if it was indented 561 // in the first place 562 i := len(prefix) 563 for n := 0; n < 3 && i > 0 && prefix[i-1] == ' '; n++ { 564 i-- 565 } 566 if i == len(prefix) && i > 0 && prefix[i-1] == '\t' { 567 i-- 568 } 569 prefix = prefix[0:i] 570 } else { 571 // comment text on the first line 572 suffix := make([]byte, len(first)) 573 n := 2 // start after opening /* 574 for n < len(first) && first[n] <= ' ' { 575 suffix[n] = first[n] 576 n++ 577 } 578 if n > 2 && suffix[2] == '\t' { 579 // assume the '\t' compensates for the /* 580 suffix = suffix[2:n] 581 } else { 582 // otherwise assume two blanks 583 suffix[0], suffix[1] = ' ', ' ' 584 suffix = suffix[0:n] 585 } 586 // Shorten the computed common prefix by the length of 587 // suffix, if it is found as suffix of the prefix. 588 prefix = strings.TrimSuffix(prefix, string(suffix)) 589 } 590 } 591 592 // Handle last line: If it only contains a closing */, align it 593 // with the opening /*, otherwise align the text with the other 594 // lines. 595 last := lines[len(lines)-1] 596 closing := "*/" 597 i := strings.Index(last, closing) // i >= 0 (closing is always present) 598 if isBlank(last[0:i]) { 599 // last line only contains closing */ 600 if lineOfStars { 601 closing = " */" // add blank to align final star 602 } 603 lines[len(lines)-1] = prefix + closing 604 } else { 605 // last line contains more comment text - assume 606 // it is aligned like the other lines and include 607 // in prefix computation 608 prefix = commonPrefix(prefix, last) 609 } 610 611 // Remove the common prefix from all but the first and empty lines. 612 for i, line := range lines { 613 if i > 0 && line != "" { 614 lines[i] = line[len(prefix):] 615 } 616 } 617 } 618 619 func (p *printer) writeComment(comment *ast.Comment) { 620 text := comment.Text 621 pos := p.posFor(comment.Pos()) 622 623 const linePrefix = "//line " 624 if strings.HasPrefix(text, linePrefix) && (!pos.IsValid() || pos.Column == 1) { 625 // possibly a line directive 626 ldir := strings.TrimSpace(text[len(linePrefix):]) 627 if i := strings.LastIndex(ldir, ":"); i >= 0 { 628 if line, err := strconv.Atoi(ldir[i+1:]); err == nil && line > 0 { 629 // The line directive we are about to print changed 630 // the Filename and Line number used for subsequent 631 // tokens. We have to update our AST-space position 632 // accordingly and suspend indentation temporarily. 633 indent := p.indent 634 p.indent = 0 635 defer func() { 636 p.pos.Filename = ldir[:i] 637 p.pos.Line = line 638 p.pos.Column = 1 639 p.indent = indent 640 }() 641 } 642 } 643 } 644 645 // shortcut common case of //-style comments 646 if text[1] == '/' { 647 p.writeString(pos, trimRight(text), true) 648 return 649 } 650 651 // for /*-style comments, print line by line and let the 652 // write function take care of the proper indentation 653 lines := strings.Split(text, "\n") 654 655 // The comment started in the first column but is going 656 // to be indented. For an idempotent result, add indentation 657 // to all lines such that they look like they were indented 658 // before - this will make sure the common prefix computation 659 // is the same independent of how many times formatting is 660 // applied (was issue 1835). 661 if pos.IsValid() && pos.Column == 1 && p.indent > 0 { 662 for i, line := range lines[1:] { 663 lines[1+i] = " " + line 664 } 665 } 666 667 stripCommonPrefix(lines) 668 669 // write comment lines, separated by formfeed, 670 // without a line break after the last line 671 for i, line := range lines { 672 if i > 0 { 673 p.writeByte('\f', 1) 674 pos = p.pos 675 } 676 if len(line) > 0 { 677 p.writeString(pos, trimRight(line), true) 678 } 679 } 680 } 681 682 // writeCommentSuffix writes a line break after a comment if indicated 683 // and processes any leftover indentation information. If a line break 684 // is needed, the kind of break (newline vs formfeed) depends on the 685 // pending whitespace. The writeCommentSuffix result indicates if a 686 // newline was written or if a formfeed was dropped from the whitespace 687 // buffer. 688 // 689 func (p *printer) writeCommentSuffix(needsLinebreak bool) (wroteNewline, droppedFF bool) { 690 for i, ch := range p.wsbuf { 691 switch ch { 692 case blank, vtab: 693 // ignore trailing whitespace 694 p.wsbuf[i] = ignore 695 case indent, unindent: 696 // don't lose indentation information 697 case newline, formfeed: 698 // if we need a line break, keep exactly one 699 // but remember if we dropped any formfeeds 700 if needsLinebreak { 701 needsLinebreak = false 702 wroteNewline = true 703 } else { 704 if ch == formfeed { 705 droppedFF = true 706 } 707 p.wsbuf[i] = ignore 708 } 709 } 710 } 711 p.writeWhitespace(len(p.wsbuf)) 712 713 // make sure we have a line break 714 if needsLinebreak { 715 p.writeByte('\n', 1) 716 wroteNewline = true 717 } 718 719 return 720 } 721 722 // containsLinebreak reports whether the whitespace buffer contains any line breaks. 723 func (p *printer) containsLinebreak() bool { 724 for _, ch := range p.wsbuf { 725 if ch == newline || ch == formfeed { 726 return true 727 } 728 } 729 return false 730 } 731 732 // intersperseComments consumes all comments that appear before the next token 733 // tok and prints it together with the buffered whitespace (i.e., the whitespace 734 // that needs to be written before the next token). A heuristic is used to mix 735 // the comments and whitespace. The intersperseComments result indicates if a 736 // newline was written or if a formfeed was dropped from the whitespace buffer. 737 // 738 func (p *printer) intersperseComments(next token.Position, tok token.Token) (wroteNewline, droppedFF bool) { 739 var last *ast.Comment 740 for p.commentBefore(next) { 741 for _, c := range p.comment.List { 742 p.writeCommentPrefix(p.posFor(c.Pos()), next, last, tok) 743 p.writeComment(c) 744 last = c 745 } 746 p.nextComment() 747 } 748 749 if last != nil { 750 // If the last comment is a /*-style comment and the next item 751 // follows on the same line but is not a comma, and not a "closing" 752 // token immediately following its corresponding "opening" token, 753 // add an extra separator unless explicitly disabled. Use a blank 754 // as separator unless we have pending linebreaks, they are not 755 // disabled, and we are outside a composite literal, in which case 756 // we want a linebreak (issue 15137). 757 // TODO(gri) This has become overly complicated. We should be able 758 // to track whether we're inside an expression or statement and 759 // use that information to decide more directly. 760 needsLinebreak := false 761 if p.mode&noExtraBlank == 0 && 762 last.Text[1] == '*' && p.lineFor(last.Pos()) == next.Line && 763 tok != token.COMMA && 764 (tok != token.RPAREN || p.prevOpen == token.LPAREN) && 765 (tok != token.RBRACK || p.prevOpen == token.LBRACK) { 766 if p.containsLinebreak() && p.mode&noExtraLinebreak == 0 && p.level == 0 { 767 needsLinebreak = true 768 } else { 769 p.writeByte(' ', 1) 770 } 771 } 772 // Ensure that there is a line break after a //-style comment, 773 // before EOF, and before a closing '}' unless explicitly disabled. 774 if last.Text[1] == '/' || 775 tok == token.EOF || 776 tok == token.RBRACE && p.mode&noExtraLinebreak == 0 { 777 needsLinebreak = true 778 } 779 return p.writeCommentSuffix(needsLinebreak) 780 } 781 782 // no comment was written - we should never reach here since 783 // intersperseComments should not be called in that case 784 p.internalError("intersperseComments called without pending comments") 785 return 786 } 787 788 // whiteWhitespace writes the first n whitespace entries. 789 func (p *printer) writeWhitespace(n int) { 790 // write entries 791 for i := 0; i < n; i++ { 792 switch ch := p.wsbuf[i]; ch { 793 case ignore: 794 // ignore! 795 case indent: 796 p.indent++ 797 case unindent: 798 p.indent-- 799 if p.indent < 0 { 800 p.internalError("negative indentation:", p.indent) 801 p.indent = 0 802 } 803 case newline, formfeed: 804 // A line break immediately followed by a "correcting" 805 // unindent is swapped with the unindent - this permits 806 // proper label positioning. If a comment is between 807 // the line break and the label, the unindent is not 808 // part of the comment whitespace prefix and the comment 809 // will be positioned correctly indented. 810 if i+1 < n && p.wsbuf[i+1] == unindent { 811 // Use a formfeed to terminate the current section. 812 // Otherwise, a long label name on the next line leading 813 // to a wide column may increase the indentation column 814 // of lines before the label; effectively leading to wrong 815 // indentation. 816 p.wsbuf[i], p.wsbuf[i+1] = unindent, formfeed 817 i-- // do it again 818 continue 819 } 820 fallthrough 821 default: 822 p.writeByte(byte(ch), 1) 823 } 824 } 825 826 // shift remaining entries down 827 l := copy(p.wsbuf, p.wsbuf[n:]) 828 p.wsbuf = p.wsbuf[:l] 829 } 830 831 // ---------------------------------------------------------------------------- 832 // Printing interface 833 834 // nlines limits n to maxNewlines. 835 func nlimit(n int) int { 836 if n > maxNewlines { 837 n = maxNewlines 838 } 839 return n 840 } 841 842 func mayCombine(prev token.Token, next byte) (b bool) { 843 switch prev { 844 case token.INT: 845 b = next == '.' // 1. 846 case token.ADD: 847 b = next == '+' // ++ 848 case token.SUB: 849 b = next == '-' // -- 850 case token.QUO: 851 b = next == '*' // /* 852 case token.LSS: 853 b = next == '-' || next == '<' // <- or << 854 case token.AND: 855 b = next == '&' || next == '^' // && or &^ 856 } 857 return 858 } 859 860 // print prints a list of "items" (roughly corresponding to syntactic 861 // tokens, but also including whitespace and formatting information). 862 // It is the only print function that should be called directly from 863 // any of the AST printing functions in nodes.go. 864 // 865 // Whitespace is accumulated until a non-whitespace token appears. Any 866 // comments that need to appear before that token are printed first, 867 // taking into account the amount and structure of any pending white- 868 // space for best comment placement. Then, any leftover whitespace is 869 // printed, followed by the actual token. 870 // 871 func (p *printer) print(args ...interface{}) { 872 for _, arg := range args { 873 // information about the current arg 874 var data string 875 var isLit bool 876 var impliedSemi bool // value for p.impliedSemi after this arg 877 878 // record previous opening token, if any 879 switch p.lastTok { 880 case token.ILLEGAL: 881 // ignore (white space) 882 case token.LPAREN, token.LBRACK: 883 p.prevOpen = p.lastTok 884 default: 885 // other tokens followed any opening token 886 p.prevOpen = token.ILLEGAL 887 } 888 889 switch x := arg.(type) { 890 case pmode: 891 // toggle printer mode 892 p.mode ^= x 893 continue 894 895 case whiteSpace: 896 if x == ignore { 897 // don't add ignore's to the buffer; they 898 // may screw up "correcting" unindents (see 899 // LabeledStmt) 900 continue 901 } 902 i := len(p.wsbuf) 903 if i == cap(p.wsbuf) { 904 // Whitespace sequences are very short so this should 905 // never happen. Handle gracefully (but possibly with 906 // bad comment placement) if it does happen. 907 p.writeWhitespace(i) 908 i = 0 909 } 910 p.wsbuf = p.wsbuf[0 : i+1] 911 p.wsbuf[i] = x 912 if x == newline || x == formfeed { 913 // newlines affect the current state (p.impliedSemi) 914 // and not the state after printing arg (impliedSemi) 915 // because comments can be interspersed before the arg 916 // in this case 917 p.impliedSemi = false 918 } 919 p.lastTok = token.ILLEGAL 920 continue 921 922 case *ast.Ident: 923 data = x.Name 924 impliedSemi = true 925 p.lastTok = token.IDENT 926 927 case *ast.BasicLit: 928 data = x.Value 929 isLit = true 930 impliedSemi = true 931 p.lastTok = x.Kind 932 933 case token.Token: 934 s := x.String() 935 if mayCombine(p.lastTok, s[0]) { 936 // the previous and the current token must be 937 // separated by a blank otherwise they combine 938 // into a different incorrect token sequence 939 // (except for token.INT followed by a '.' this 940 // should never happen because it is taken care 941 // of via binary expression formatting) 942 if len(p.wsbuf) != 0 { 943 p.internalError("whitespace buffer not empty") 944 } 945 p.wsbuf = p.wsbuf[0:1] 946 p.wsbuf[0] = ' ' 947 } 948 data = s 949 // some keywords followed by a newline imply a semicolon 950 switch x { 951 case token.BREAK, token.CONTINUE, token.FALLTHROUGH, token.RETURN, 952 token.INC, token.DEC, token.RPAREN, token.RBRACK, token.RBRACE: 953 impliedSemi = true 954 } 955 p.lastTok = x 956 957 case token.Pos: 958 if x.IsValid() { 959 p.pos = p.posFor(x) // accurate position of next item 960 } 961 continue 962 963 case string: 964 // incorrect AST - print error message 965 data = x 966 isLit = true 967 impliedSemi = true 968 p.lastTok = token.STRING 969 970 default: 971 fmt.Fprintf(os.Stderr, "print: unsupported argument %v (%T)\n", arg, arg) 972 panic("go/printer type") 973 } 974 // data != "" 975 976 next := p.pos // estimated/accurate position of next item 977 wroteNewline, droppedFF := p.flush(next, p.lastTok) 978 979 // intersperse extra newlines if present in the source and 980 // if they don't cause extra semicolons (don't do this in 981 // flush as it will cause extra newlines at the end of a file) 982 if !p.impliedSemi { 983 n := nlimit(next.Line - p.pos.Line) 984 // don't exceed maxNewlines if we already wrote one 985 if wroteNewline && n == maxNewlines { 986 n = maxNewlines - 1 987 } 988 if n > 0 { 989 ch := byte('\n') 990 if droppedFF { 991 ch = '\f' // use formfeed since we dropped one before 992 } 993 p.writeByte(ch, n) 994 impliedSemi = false 995 } 996 } 997 998 // the next token starts now - record its line number if requested 999 if p.linePtr != nil { 1000 *p.linePtr = p.out.Line 1001 p.linePtr = nil 1002 } 1003 1004 p.writeString(next, data, isLit) 1005 p.impliedSemi = impliedSemi 1006 } 1007 } 1008 1009 // flush prints any pending comments and whitespace occurring textually 1010 // before the position of the next token tok. The flush result indicates 1011 // if a newline was written or if a formfeed was dropped from the whitespace 1012 // buffer. 1013 // 1014 func (p *printer) flush(next token.Position, tok token.Token) (wroteNewline, droppedFF bool) { 1015 if p.commentBefore(next) { 1016 // if there are comments before the next item, intersperse them 1017 wroteNewline, droppedFF = p.intersperseComments(next, tok) 1018 } else { 1019 // otherwise, write any leftover whitespace 1020 p.writeWhitespace(len(p.wsbuf)) 1021 } 1022 return 1023 } 1024 1025 // getNode returns the ast.CommentGroup associated with n, if any. 1026 func getDoc(n ast.Node) *ast.CommentGroup { 1027 switch n := n.(type) { 1028 case *ast.Field: 1029 return n.Doc 1030 case *ast.ImportSpec: 1031 return n.Doc 1032 case *ast.ValueSpec: 1033 return n.Doc 1034 case *ast.TypeSpec: 1035 return n.Doc 1036 case *ast.GenDecl: 1037 return n.Doc 1038 case *ast.FuncDecl: 1039 return n.Doc 1040 case *ast.File: 1041 return n.Doc 1042 } 1043 return nil 1044 } 1045 1046 func getLastComment(n ast.Node) *ast.CommentGroup { 1047 switch n := n.(type) { 1048 case *ast.Field: 1049 return n.Comment 1050 case *ast.ImportSpec: 1051 return n.Comment 1052 case *ast.ValueSpec: 1053 return n.Comment 1054 case *ast.TypeSpec: 1055 return n.Comment 1056 case *ast.GenDecl: 1057 if len(n.Specs) > 0 { 1058 return getLastComment(n.Specs[len(n.Specs)-1]) 1059 } 1060 case *ast.File: 1061 if len(n.Comments) > 0 { 1062 return n.Comments[len(n.Comments)-1] 1063 } 1064 } 1065 return nil 1066 } 1067 1068 func (p *printer) printNode(node interface{}) error { 1069 // unpack *CommentedNode, if any 1070 var comments []*ast.CommentGroup 1071 if cnode, ok := node.(*CommentedNode); ok { 1072 node = cnode.Node 1073 comments = cnode.Comments 1074 } 1075 1076 if comments != nil { 1077 // commented node - restrict comment list to relevant range 1078 n, ok := node.(ast.Node) 1079 if !ok { 1080 goto unsupported 1081 } 1082 beg := n.Pos() 1083 end := n.End() 1084 // if the node has associated documentation, 1085 // include that commentgroup in the range 1086 // (the comment list is sorted in the order 1087 // of the comment appearance in the source code) 1088 if doc := getDoc(n); doc != nil { 1089 beg = doc.Pos() 1090 } 1091 if com := getLastComment(n); com != nil { 1092 if e := com.End(); e > end { 1093 end = e 1094 } 1095 } 1096 // token.Pos values are global offsets, we can 1097 // compare them directly 1098 i := 0 1099 for i < len(comments) && comments[i].End() < beg { 1100 i++ 1101 } 1102 j := i 1103 for j < len(comments) && comments[j].Pos() < end { 1104 j++ 1105 } 1106 if i < j { 1107 p.comments = comments[i:j] 1108 } 1109 } else if n, ok := node.(*ast.File); ok { 1110 // use ast.File comments, if any 1111 p.comments = n.Comments 1112 } 1113 1114 // if there are no comments, use node comments 1115 p.useNodeComments = p.comments == nil 1116 1117 // get comments ready for use 1118 p.nextComment() 1119 1120 // format node 1121 switch n := node.(type) { 1122 case ast.Expr: 1123 p.expr(n) 1124 case ast.Stmt: 1125 // A labeled statement will un-indent to position the label. 1126 // Set p.indent to 1 so we don't get indent "underflow". 1127 if _, ok := n.(*ast.LabeledStmt); ok { 1128 p.indent = 1 1129 } 1130 p.stmt(n, false) 1131 case ast.Decl: 1132 p.decl(n) 1133 case ast.Spec: 1134 p.spec(n, 1, false) 1135 case []ast.Stmt: 1136 // A labeled statement will un-indent to position the label. 1137 // Set p.indent to 1 so we don't get indent "underflow". 1138 for _, s := range n { 1139 if _, ok := s.(*ast.LabeledStmt); ok { 1140 p.indent = 1 1141 } 1142 } 1143 p.stmtList(n, 0, false) 1144 case []ast.Decl: 1145 p.declList(n) 1146 case *ast.File: 1147 p.file(n) 1148 default: 1149 goto unsupported 1150 } 1151 1152 return nil 1153 1154 unsupported: 1155 return fmt.Errorf("go/printer: unsupported node type %T", node) 1156 } 1157 1158 // ---------------------------------------------------------------------------- 1159 // Trimmer 1160 1161 // A trimmer is an io.Writer filter for stripping tabwriter.Escape 1162 // characters, trailing blanks and tabs, and for converting formfeed 1163 // and vtab characters into newlines and htabs (in case no tabwriter 1164 // is used). Text bracketed by tabwriter.Escape characters is passed 1165 // through unchanged. 1166 // 1167 type trimmer struct { 1168 output io.Writer 1169 state int 1170 space []byte 1171 } 1172 1173 // trimmer is implemented as a state machine. 1174 // It can be in one of the following states: 1175 const ( 1176 inSpace = iota // inside space 1177 inEscape // inside text bracketed by tabwriter.Escapes 1178 inText // inside text 1179 ) 1180 1181 func (p *trimmer) resetSpace() { 1182 p.state = inSpace 1183 p.space = p.space[0:0] 1184 } 1185 1186 // Design note: It is tempting to eliminate extra blanks occurring in 1187 // whitespace in this function as it could simplify some 1188 // of the blanks logic in the node printing functions. 1189 // However, this would mess up any formatting done by 1190 // the tabwriter. 1191 1192 var aNewline = []byte("\n") 1193 1194 func (p *trimmer) Write(data []byte) (n int, err error) { 1195 // invariants: 1196 // p.state == inSpace: 1197 // p.space is unwritten 1198 // p.state == inEscape, inText: 1199 // data[m:n] is unwritten 1200 m := 0 1201 var b byte 1202 for n, b = range data { 1203 if b == '\v' { 1204 b = '\t' // convert to htab 1205 } 1206 switch p.state { 1207 case inSpace: 1208 switch b { 1209 case '\t', ' ': 1210 p.space = append(p.space, b) 1211 case '\n', '\f': 1212 p.resetSpace() // discard trailing space 1213 _, err = p.output.Write(aNewline) 1214 case tabwriter.Escape: 1215 _, err = p.output.Write(p.space) 1216 p.state = inEscape 1217 m = n + 1 // +1: skip tabwriter.Escape 1218 default: 1219 _, err = p.output.Write(p.space) 1220 p.state = inText 1221 m = n 1222 } 1223 case inEscape: 1224 if b == tabwriter.Escape { 1225 _, err = p.output.Write(data[m:n]) 1226 p.resetSpace() 1227 } 1228 case inText: 1229 switch b { 1230 case '\t', ' ': 1231 _, err = p.output.Write(data[m:n]) 1232 p.resetSpace() 1233 p.space = append(p.space, b) 1234 case '\n', '\f': 1235 _, err = p.output.Write(data[m:n]) 1236 p.resetSpace() 1237 if err == nil { 1238 _, err = p.output.Write(aNewline) 1239 } 1240 case tabwriter.Escape: 1241 _, err = p.output.Write(data[m:n]) 1242 p.state = inEscape 1243 m = n + 1 // +1: skip tabwriter.Escape 1244 } 1245 default: 1246 panic("unreachable") 1247 } 1248 if err != nil { 1249 return 1250 } 1251 } 1252 n = len(data) 1253 1254 switch p.state { 1255 case inEscape, inText: 1256 _, err = p.output.Write(data[m:n]) 1257 p.resetSpace() 1258 } 1259 1260 return 1261 } 1262 1263 // ---------------------------------------------------------------------------- 1264 // Public interface 1265 1266 // A Mode value is a set of flags (or 0). They control printing. 1267 type Mode uint 1268 1269 const ( 1270 RawFormat Mode = 1 << iota // do not use a tabwriter; if set, UseSpaces is ignored 1271 TabIndent // use tabs for indentation independent of UseSpaces 1272 UseSpaces // use spaces instead of tabs for alignment 1273 SourcePos // emit //line directives to preserve original source positions 1274 ) 1275 1276 // A Config node controls the output of Fprint. 1277 type Config struct { 1278 Mode Mode // default: 0 1279 Tabwidth int // default: 8 1280 Indent int // default: 0 (all code is indented at least by this much) 1281 } 1282 1283 // fprint implements Fprint and takes a nodesSizes map for setting up the printer state. 1284 func (cfg *Config) fprint(output io.Writer, fset *token.FileSet, node interface{}, nodeSizes map[ast.Node]int) (err error) { 1285 // print node 1286 var p printer 1287 p.init(cfg, fset, nodeSizes) 1288 if err = p.printNode(node); err != nil { 1289 return 1290 } 1291 // print outstanding comments 1292 p.impliedSemi = false // EOF acts like a newline 1293 p.flush(token.Position{Offset: infinity, Line: infinity}, token.EOF) 1294 1295 // redirect output through a trimmer to eliminate trailing whitespace 1296 // (Input to a tabwriter must be untrimmed since trailing tabs provide 1297 // formatting information. The tabwriter could provide trimming 1298 // functionality but no tabwriter is used when RawFormat is set.) 1299 output = &trimmer{output: output} 1300 1301 // redirect output through a tabwriter if necessary 1302 if cfg.Mode&RawFormat == 0 { 1303 minwidth := cfg.Tabwidth 1304 1305 padchar := byte('\t') 1306 if cfg.Mode&UseSpaces != 0 { 1307 padchar = ' ' 1308 } 1309 1310 twmode := tabwriter.DiscardEmptyColumns 1311 if cfg.Mode&TabIndent != 0 { 1312 minwidth = 0 1313 twmode |= tabwriter.TabIndent 1314 } 1315 1316 output = tabwriter.NewWriter(output, minwidth, cfg.Tabwidth, 1, padchar, twmode) 1317 } 1318 1319 // write printer result via tabwriter/trimmer to output 1320 if _, err = output.Write(p.output); err != nil { 1321 return 1322 } 1323 1324 // flush tabwriter, if any 1325 if tw, _ := output.(*tabwriter.Writer); tw != nil { 1326 err = tw.Flush() 1327 } 1328 1329 return 1330 } 1331 1332 // A CommentedNode bundles an AST node and corresponding comments. 1333 // It may be provided as argument to any of the Fprint functions. 1334 // 1335 type CommentedNode struct { 1336 Node interface{} // *ast.File, or ast.Expr, ast.Decl, ast.Spec, or ast.Stmt 1337 Comments []*ast.CommentGroup 1338 } 1339 1340 // Fprint "pretty-prints" an AST node to output for a given configuration cfg. 1341 // Position information is interpreted relative to the file set fset. 1342 // The node type must be *ast.File, *CommentedNode, []ast.Decl, []ast.Stmt, 1343 // or assignment-compatible to ast.Expr, ast.Decl, ast.Spec, or ast.Stmt. 1344 // 1345 func (cfg *Config) Fprint(output io.Writer, fset *token.FileSet, node interface{}) error { 1346 return cfg.fprint(output, fset, node, make(map[ast.Node]int)) 1347 } 1348 1349 // Fprint "pretty-prints" an AST node to output. 1350 // It calls Config.Fprint with default settings. 1351 // Note that gofmt uses tabs for indentation but spaces for alignment; 1352 // use format.Node (package go/format) for output that matches gofmt. 1353 // 1354 func Fprint(output io.Writer, fset *token.FileSet, node interface{}) error { 1355 return (&Config{Tabwidth: 8}).Fprint(output, fset, node) 1356 }