github.com/q45/go@v0.0.0-20151101211701-a4fb8c13db3f/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 mode pmode // current printer mode 62 impliedSemi bool // if set, a linebreak implies a semicolon 63 lastTok token.Token // last token printed (token.ILLEGAL if it's whitespace) 64 prevOpen token.Token // previous non-brace "open" token (, [, or token.ILLEGAL 65 wsbuf []whiteSpace // delayed white space 66 67 // Positions 68 // The out position differs from the pos position when the result 69 // formatting differs from the source formatting (in the amount of 70 // white space). If there's a difference and SourcePos is set in 71 // ConfigMode, //line comments are used in the output to restore 72 // original source positions for a reader. 73 pos token.Position // current position in AST (source) space 74 out token.Position // current position in output space 75 last token.Position // value of pos after calling writeString 76 linePtr *int // if set, record out.Line for the next token in *linePtr 77 78 // The list of all source comments, in order of appearance. 79 comments []*ast.CommentGroup // may be nil 80 useNodeComments bool // if not set, ignore lead and line comments of nodes 81 82 // Information about p.comments[p.cindex]; set up by nextComment. 83 commentInfo 84 85 // Cache of already computed node sizes. 86 nodeSizes map[ast.Node]int 87 88 // Cache of most recently computed line position. 89 cachedPos token.Pos 90 cachedLine int // line corresponding to cachedPos 91 } 92 93 func (p *printer) init(cfg *Config, fset *token.FileSet, nodeSizes map[ast.Node]int) { 94 p.Config = *cfg 95 p.fset = fset 96 p.pos = token.Position{Line: 1, Column: 1} 97 p.out = token.Position{Line: 1, Column: 1} 98 p.wsbuf = make([]whiteSpace, 0, 16) // whitespace sequences are short 99 p.nodeSizes = nodeSizes 100 p.cachedPos = -1 101 } 102 103 func (p *printer) internalError(msg ...interface{}) { 104 if debug { 105 fmt.Print(p.pos.String() + ": ") 106 fmt.Println(msg...) 107 panic("go/printer") 108 } 109 } 110 111 // commentsHaveNewline reports whether a list of comments belonging to 112 // an *ast.CommentGroup contains newlines. Because the position information 113 // may only be partially correct, we also have to read the comment text. 114 func (p *printer) commentsHaveNewline(list []*ast.Comment) bool { 115 // len(list) > 0 116 line := p.lineFor(list[0].Pos()) 117 for i, c := range list { 118 if i > 0 && p.lineFor(list[i].Pos()) != line { 119 // not all comments on the same line 120 return true 121 } 122 if t := c.Text; len(t) >= 2 && (t[1] == '/' || strings.Contains(t, "\n")) { 123 return true 124 } 125 } 126 _ = line 127 return false 128 } 129 130 func (p *printer) nextComment() { 131 for p.cindex < len(p.comments) { 132 c := p.comments[p.cindex] 133 p.cindex++ 134 if list := c.List; len(list) > 0 { 135 p.comment = c 136 p.commentOffset = p.posFor(list[0].Pos()).Offset 137 p.commentNewline = p.commentsHaveNewline(list) 138 return 139 } 140 // we should not reach here (correct ASTs don't have empty 141 // ast.CommentGroup nodes), but be conservative and try again 142 } 143 // no more comments 144 p.commentOffset = infinity 145 } 146 147 // commentBefore reports whether the current comment group occurs 148 // before the next position in the source code and printing it does 149 // not introduce implicit semicolons. 150 // 151 func (p *printer) commentBefore(next token.Position) bool { 152 return p.commentOffset < next.Offset && (!p.impliedSemi || !p.commentNewline) 153 } 154 155 // commentSizeBefore returns the estimated size of the 156 // comments on the same line before the next position. 157 // 158 func (p *printer) commentSizeBefore(next token.Position) int { 159 // save/restore current p.commentInfo (p.nextComment() modifies it) 160 defer func(info commentInfo) { 161 p.commentInfo = info 162 }(p.commentInfo) 163 164 size := 0 165 for p.commentBefore(next) { 166 for _, c := range p.comment.List { 167 size += len(c.Text) 168 } 169 p.nextComment() 170 } 171 return size 172 } 173 174 // recordLine records the output line number for the next non-whitespace 175 // token in *linePtr. It is used to compute an accurate line number for a 176 // formatted construct, independent of pending (not yet emitted) whitespace 177 // or comments. 178 // 179 func (p *printer) recordLine(linePtr *int) { 180 p.linePtr = linePtr 181 } 182 183 // linesFrom returns the number of output lines between the current 184 // output line and the line argument, ignoring any pending (not yet 185 // emitted) whitespace or comments. It is used to compute an accurate 186 // size (in number of lines) for a formatted construct. 187 // 188 func (p *printer) linesFrom(line int) int { 189 return p.out.Line - line 190 } 191 192 func (p *printer) posFor(pos token.Pos) token.Position { 193 // not used frequently enough to cache entire token.Position 194 return p.fset.Position(pos) 195 } 196 197 func (p *printer) lineFor(pos token.Pos) int { 198 if pos != p.cachedPos { 199 p.cachedPos = pos 200 p.cachedLine = p.fset.Position(pos).Line 201 } 202 return p.cachedLine 203 } 204 205 // atLineBegin emits a //line comment if necessary and prints indentation. 206 func (p *printer) atLineBegin(pos token.Position) { 207 // write a //line comment if necessary 208 if p.Config.Mode&SourcePos != 0 && 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 comment 213 p.out.Filename = pos.Filename 214 p.out.Line = pos.Line 215 } 216 217 // write indentation 218 // use "hard" htabs - indentation columns 219 // must not be discarded by the tabwriter 220 n := p.Config.Indent + p.indent // include base indentation 221 for i := 0; i < n; i++ { 222 p.output = append(p.output, '\t') 223 } 224 225 // update positions 226 p.pos.Offset += n 227 p.pos.Column += n 228 p.out.Column += n 229 } 230 231 // writeByte writes ch n times to p.output and updates p.pos. 232 func (p *printer) writeByte(ch byte, n int) { 233 if p.out.Column == 1 { 234 p.atLineBegin(p.pos) 235 } 236 237 for i := 0; i < n; i++ { 238 p.output = append(p.output, ch) 239 } 240 241 // update positions 242 p.pos.Offset += n 243 if ch == '\n' || ch == '\f' { 244 p.pos.Line += n 245 p.out.Line += n 246 p.pos.Column = 1 247 p.out.Column = 1 248 return 249 } 250 p.pos.Column += n 251 p.out.Column += n 252 } 253 254 // writeString writes the string s to p.output and updates p.pos, p.out, 255 // and p.last. If isLit is set, s is escaped w/ tabwriter.Escape characters 256 // to protect s from being interpreted by the tabwriter. 257 // 258 // Note: writeString is only used to write Go tokens, literals, and 259 // comments, all of which must be written literally. Thus, it is correct 260 // to always set isLit = true. However, setting it explicitly only when 261 // needed (i.e., when we don't know that s contains no tabs or line breaks) 262 // avoids processing extra escape characters and reduces run time of the 263 // printer benchmark by up to 10%. 264 // 265 func (p *printer) writeString(pos token.Position, s string, isLit bool) { 266 if p.out.Column == 1 { 267 p.atLineBegin(pos) 268 } 269 270 if pos.IsValid() { 271 // update p.pos (if pos is invalid, continue with existing p.pos) 272 // Note: Must do this after handling line beginnings because 273 // atLineBegin updates p.pos if there's indentation, but p.pos 274 // is the position of s. 275 p.pos = pos 276 } 277 278 if isLit { 279 // Protect s such that is passes through the tabwriter 280 // unchanged. Note that valid Go programs cannot contain 281 // tabwriter.Escape bytes since they do not appear in legal 282 // UTF-8 sequences. 283 p.output = append(p.output, tabwriter.Escape) 284 } 285 286 if debug { 287 p.output = append(p.output, fmt.Sprintf("/*%s*/", pos)...) // do not update p.pos! 288 } 289 p.output = append(p.output, s...) 290 291 // update positions 292 nlines := 0 293 var li int // index of last newline; valid if nlines > 0 294 for i := 0; i < len(s); i++ { 295 // Go tokens cannot contain '\f' - no need to look for it 296 if s[i] == '\n' { 297 nlines++ 298 li = i 299 } 300 } 301 p.pos.Offset += len(s) 302 if nlines > 0 { 303 p.pos.Line += nlines 304 p.out.Line += nlines 305 c := len(s) - li 306 p.pos.Column = c 307 p.out.Column = c 308 } else { 309 p.pos.Column += len(s) 310 p.out.Column += len(s) 311 } 312 313 if isLit { 314 p.output = append(p.output, tabwriter.Escape) 315 } 316 317 p.last = p.pos 318 } 319 320 // writeCommentPrefix writes the whitespace before a comment. 321 // If there is any pending whitespace, it consumes as much of 322 // it as is likely to help position the comment nicely. 323 // pos is the comment position, next the position of the item 324 // after all pending comments, prev is the previous comment in 325 // a group of comments (or nil), and tok is the next token. 326 // 327 func (p *printer) writeCommentPrefix(pos, next token.Position, prev, comment *ast.Comment, tok token.Token) { 328 if len(p.output) == 0 { 329 // the comment is the first item to be printed - don't write any whitespace 330 return 331 } 332 333 if pos.IsValid() && pos.Filename != p.last.Filename { 334 // comment in a different file - separate with newlines 335 p.writeByte('\f', maxNewlines) 336 return 337 } 338 339 if pos.Line == p.last.Line && (prev == nil || prev.Text[1] != '/') { 340 // comment on the same line as last item: 341 // separate with at least one separator 342 hasSep := false 343 if prev == nil { 344 // first comment of a comment group 345 j := 0 346 for i, ch := range p.wsbuf { 347 switch ch { 348 case blank: 349 // ignore any blanks before a comment 350 p.wsbuf[i] = ignore 351 continue 352 case vtab: 353 // respect existing tabs - important 354 // for proper formatting of commented structs 355 hasSep = true 356 continue 357 case indent: 358 // apply pending indentation 359 continue 360 } 361 j = i 362 break 363 } 364 p.writeWhitespace(j) 365 } 366 // make sure there is at least one separator 367 if !hasSep { 368 sep := byte('\t') 369 if pos.Line == next.Line { 370 // next item is on the same line as the comment 371 // (which must be a /*-style comment): separate 372 // with a blank instead of a tab 373 sep = ' ' 374 } 375 p.writeByte(sep, 1) 376 } 377 378 } else { 379 // comment on a different line: 380 // separate with at least one line break 381 droppedLinebreak := false 382 j := 0 383 for i, ch := range p.wsbuf { 384 switch ch { 385 case blank, vtab: 386 // ignore any horizontal whitespace before line breaks 387 p.wsbuf[i] = ignore 388 continue 389 case indent: 390 // apply pending indentation 391 continue 392 case unindent: 393 // if this is not the last unindent, apply it 394 // as it is (likely) belonging to the last 395 // construct (e.g., a multi-line expression list) 396 // and is not part of closing a block 397 if i+1 < len(p.wsbuf) && p.wsbuf[i+1] == unindent { 398 continue 399 } 400 // if the next token is not a closing }, apply the unindent 401 // if it appears that the comment is aligned with the 402 // token; otherwise assume the unindent is part of a 403 // closing block and stop (this scenario appears with 404 // comments before a case label where the comments 405 // apply to the next case instead of the current one) 406 if tok != token.RBRACE && pos.Column == next.Column { 407 continue 408 } 409 case newline, formfeed: 410 p.wsbuf[i] = ignore 411 droppedLinebreak = prev == nil // record only if first comment of a group 412 } 413 j = i 414 break 415 } 416 p.writeWhitespace(j) 417 418 // determine number of linebreaks before the comment 419 n := 0 420 if pos.IsValid() && p.last.IsValid() { 421 n = pos.Line - p.last.Line 422 if n < 0 { // should never happen 423 n = 0 424 } 425 } 426 427 // at the package scope level only (p.indent == 0), 428 // add an extra newline if we dropped one before: 429 // this preserves a blank line before documentation 430 // comments at the package scope level (issue 2570) 431 if p.indent == 0 && droppedLinebreak { 432 n++ 433 } 434 435 // make sure there is at least one line break 436 // if the previous comment was a line comment 437 if n == 0 && prev != nil && prev.Text[1] == '/' { 438 n = 1 439 } 440 441 if n > 0 { 442 // use formfeeds to break columns before a comment; 443 // this is analogous to using formfeeds to separate 444 // individual lines of /*-style comments 445 p.writeByte('\f', nlimit(n)) 446 } 447 } 448 } 449 450 // Returns true if s contains only white space 451 // (only tabs and blanks can appear in the printer's context). 452 // 453 func isBlank(s string) bool { 454 for i := 0; i < len(s); i++ { 455 if s[i] > ' ' { 456 return false 457 } 458 } 459 return true 460 } 461 462 // commonPrefix returns the common prefix of a and b. 463 func commonPrefix(a, b string) string { 464 i := 0 465 for i < len(a) && i < len(b) && a[i] == b[i] && (a[i] <= ' ' || a[i] == '*') { 466 i++ 467 } 468 return a[0:i] 469 } 470 471 // trimRight returns s with trailing whitespace removed. 472 func trimRight(s string) string { 473 return strings.TrimRightFunc(s, unicode.IsSpace) 474 } 475 476 // stripCommonPrefix removes a common prefix from /*-style comment lines (unless no 477 // comment line is indented, all but the first line have some form of space prefix). 478 // The prefix is computed using heuristics such that is likely that the comment 479 // contents are nicely laid out after re-printing each line using the printer's 480 // current indentation. 481 // 482 func stripCommonPrefix(lines []string) { 483 if len(lines) <= 1 { 484 return // at most one line - nothing to do 485 } 486 // len(lines) > 1 487 488 // The heuristic in this function tries to handle a few 489 // common patterns of /*-style comments: Comments where 490 // the opening /* and closing */ are aligned and the 491 // rest of the comment text is aligned and indented with 492 // blanks or tabs, cases with a vertical "line of stars" 493 // on the left, and cases where the closing */ is on the 494 // same line as the last comment text. 495 496 // Compute maximum common white prefix of all but the first, 497 // last, and blank lines, and replace blank lines with empty 498 // lines (the first line starts with /* and has no prefix). 499 // In cases where only the first and last lines are not blank, 500 // such as two-line comments, or comments where all inner lines 501 // are blank, consider the last line for the prefix computation 502 // since otherwise the prefix would be empty. 503 // 504 // Note that the first and last line are never empty (they 505 // contain the opening /* and closing */ respectively) and 506 // thus they can be ignored by the blank line check. 507 prefix := "" 508 prefixSet := false 509 if len(lines) > 2 { 510 for i, line := range lines[1 : len(lines)-1] { 511 if isBlank(line) { 512 lines[1+i] = "" // range starts with lines[1] 513 } else { 514 if !prefixSet { 515 prefix = line 516 prefixSet = true 517 } 518 prefix = commonPrefix(prefix, line) 519 } 520 521 } 522 } 523 // If we don't have a prefix yet, consider the last line. 524 if !prefixSet { 525 line := lines[len(lines)-1] 526 prefix = commonPrefix(line, line) 527 } 528 529 /* 530 * Check for vertical "line of stars" and correct prefix accordingly. 531 */ 532 lineOfStars := false 533 if i := strings.Index(prefix, "*"); i >= 0 { 534 // Line of stars present. 535 if i > 0 && prefix[i-1] == ' ' { 536 i-- // remove trailing blank from prefix so stars remain aligned 537 } 538 prefix = prefix[0:i] 539 lineOfStars = true 540 } else { 541 // No line of stars present. 542 // Determine the white space on the first line after the /* 543 // and before the beginning of the comment text, assume two 544 // blanks instead of the /* unless the first character after 545 // the /* is a tab. If the first comment line is empty but 546 // for the opening /*, assume up to 3 blanks or a tab. This 547 // whitespace may be found as suffix in the common prefix. 548 first := lines[0] 549 if isBlank(first[2:]) { 550 // no comment text on the first line: 551 // reduce prefix by up to 3 blanks or a tab 552 // if present - this keeps comment text indented 553 // relative to the /* and */'s if it was indented 554 // in the first place 555 i := len(prefix) 556 for n := 0; n < 3 && i > 0 && prefix[i-1] == ' '; n++ { 557 i-- 558 } 559 if i == len(prefix) && i > 0 && prefix[i-1] == '\t' { 560 i-- 561 } 562 prefix = prefix[0:i] 563 } else { 564 // comment text on the first line 565 suffix := make([]byte, len(first)) 566 n := 2 // start after opening /* 567 for n < len(first) && first[n] <= ' ' { 568 suffix[n] = first[n] 569 n++ 570 } 571 if n > 2 && suffix[2] == '\t' { 572 // assume the '\t' compensates for the /* 573 suffix = suffix[2:n] 574 } else { 575 // otherwise assume two blanks 576 suffix[0], suffix[1] = ' ', ' ' 577 suffix = suffix[0:n] 578 } 579 // Shorten the computed common prefix by the length of 580 // suffix, if it is found as suffix of the prefix. 581 prefix = strings.TrimSuffix(prefix, string(suffix)) 582 } 583 } 584 585 // Handle last line: If it only contains a closing */, align it 586 // with the opening /*, otherwise align the text with the other 587 // lines. 588 last := lines[len(lines)-1] 589 closing := "*/" 590 i := strings.Index(last, closing) // i >= 0 (closing is always present) 591 if isBlank(last[0:i]) { 592 // last line only contains closing */ 593 if lineOfStars { 594 closing = " */" // add blank to align final star 595 } 596 lines[len(lines)-1] = prefix + closing 597 } else { 598 // last line contains more comment text - assume 599 // it is aligned like the other lines and include 600 // in prefix computation 601 prefix = commonPrefix(prefix, last) 602 } 603 604 // Remove the common prefix from all but the first and empty lines. 605 for i, line := range lines { 606 if i > 0 && line != "" { 607 lines[i] = line[len(prefix):] 608 } 609 } 610 } 611 612 func (p *printer) writeComment(comment *ast.Comment) { 613 text := comment.Text 614 pos := p.posFor(comment.Pos()) 615 616 const linePrefix = "//line " 617 if strings.HasPrefix(text, linePrefix) && (!pos.IsValid() || pos.Column == 1) { 618 // possibly a line directive 619 ldir := strings.TrimSpace(text[len(linePrefix):]) 620 if i := strings.LastIndex(ldir, ":"); i >= 0 { 621 if line, err := strconv.Atoi(ldir[i+1:]); err == nil && line > 0 { 622 // The line directive we are about to print changed 623 // the Filename and Line number used for subsequent 624 // tokens. We have to update our AST-space position 625 // accordingly and suspend indentation temporarily. 626 indent := p.indent 627 p.indent = 0 628 defer func() { 629 p.pos.Filename = ldir[:i] 630 p.pos.Line = line 631 p.pos.Column = 1 632 p.indent = indent 633 }() 634 } 635 } 636 } 637 638 // shortcut common case of //-style comments 639 if text[1] == '/' { 640 p.writeString(pos, trimRight(text), true) 641 return 642 } 643 644 // for /*-style comments, print line by line and let the 645 // write function take care of the proper indentation 646 lines := strings.Split(text, "\n") 647 648 // The comment started in the first column but is going 649 // to be indented. For an idempotent result, add indentation 650 // to all lines such that they look like they were indented 651 // before - this will make sure the common prefix computation 652 // is the same independent of how many times formatting is 653 // applied (was issue 1835). 654 if pos.IsValid() && pos.Column == 1 && p.indent > 0 { 655 for i, line := range lines[1:] { 656 lines[1+i] = " " + line 657 } 658 } 659 660 stripCommonPrefix(lines) 661 662 // write comment lines, separated by formfeed, 663 // without a line break after the last line 664 for i, line := range lines { 665 if i > 0 { 666 p.writeByte('\f', 1) 667 pos = p.pos 668 } 669 if len(line) > 0 { 670 p.writeString(pos, trimRight(line), true) 671 } 672 } 673 } 674 675 // writeCommentSuffix writes a line break after a comment if indicated 676 // and processes any leftover indentation information. If a line break 677 // is needed, the kind of break (newline vs formfeed) depends on the 678 // pending whitespace. The writeCommentSuffix result indicates if a 679 // newline was written or if a formfeed was dropped from the whitespace 680 // buffer. 681 // 682 func (p *printer) writeCommentSuffix(needsLinebreak bool) (wroteNewline, droppedFF bool) { 683 for i, ch := range p.wsbuf { 684 switch ch { 685 case blank, vtab: 686 // ignore trailing whitespace 687 p.wsbuf[i] = ignore 688 case indent, unindent: 689 // don't lose indentation information 690 case newline, formfeed: 691 // if we need a line break, keep exactly one 692 // but remember if we dropped any formfeeds 693 if needsLinebreak { 694 needsLinebreak = false 695 wroteNewline = true 696 } else { 697 if ch == formfeed { 698 droppedFF = true 699 } 700 p.wsbuf[i] = ignore 701 } 702 } 703 } 704 p.writeWhitespace(len(p.wsbuf)) 705 706 // make sure we have a line break 707 if needsLinebreak { 708 p.writeByte('\n', 1) 709 wroteNewline = true 710 } 711 712 return 713 } 714 715 // intersperseComments consumes all comments that appear before the next token 716 // tok and prints it together with the buffered whitespace (i.e., the whitespace 717 // that needs to be written before the next token). A heuristic is used to mix 718 // the comments and whitespace. The intersperseComments result indicates if a 719 // newline was written or if a formfeed was dropped from the whitespace buffer. 720 // 721 func (p *printer) intersperseComments(next token.Position, tok token.Token) (wroteNewline, droppedFF bool) { 722 var last *ast.Comment 723 for p.commentBefore(next) { 724 for _, c := range p.comment.List { 725 p.writeCommentPrefix(p.posFor(c.Pos()), next, last, c, tok) 726 p.writeComment(c) 727 last = c 728 } 729 p.nextComment() 730 } 731 732 if last != nil { 733 // if the last comment is a /*-style comment and the next item 734 // follows on the same line but is not a comma, and not a "closing" 735 // token immediately following its corresponding "opening" token, 736 // add an extra blank for separation unless explicitly disabled 737 if p.mode&noExtraBlank == 0 && 738 last.Text[1] == '*' && p.lineFor(last.Pos()) == next.Line && 739 tok != token.COMMA && 740 (tok != token.RPAREN || p.prevOpen == token.LPAREN) && 741 (tok != token.RBRACK || p.prevOpen == token.LBRACK) { 742 p.writeByte(' ', 1) 743 } 744 // ensure that there is a line break after a //-style comment, 745 // before a closing '}' unless explicitly disabled, or at eof 746 needsLinebreak := 747 last.Text[1] == '/' || 748 tok == token.RBRACE && p.mode&noExtraLinebreak == 0 || 749 tok == token.EOF 750 return p.writeCommentSuffix(needsLinebreak) 751 } 752 753 // no comment was written - we should never reach here since 754 // intersperseComments should not be called in that case 755 p.internalError("intersperseComments called without pending comments") 756 return 757 } 758 759 // whiteWhitespace writes the first n whitespace entries. 760 func (p *printer) writeWhitespace(n int) { 761 // write entries 762 for i := 0; i < n; i++ { 763 switch ch := p.wsbuf[i]; ch { 764 case ignore: 765 // ignore! 766 case indent: 767 p.indent++ 768 case unindent: 769 p.indent-- 770 if p.indent < 0 { 771 p.internalError("negative indentation:", p.indent) 772 p.indent = 0 773 } 774 case newline, formfeed: 775 // A line break immediately followed by a "correcting" 776 // unindent is swapped with the unindent - this permits 777 // proper label positioning. If a comment is between 778 // the line break and the label, the unindent is not 779 // part of the comment whitespace prefix and the comment 780 // will be positioned correctly indented. 781 if i+1 < n && p.wsbuf[i+1] == unindent { 782 // Use a formfeed to terminate the current section. 783 // Otherwise, a long label name on the next line leading 784 // to a wide column may increase the indentation column 785 // of lines before the label; effectively leading to wrong 786 // indentation. 787 p.wsbuf[i], p.wsbuf[i+1] = unindent, formfeed 788 i-- // do it again 789 continue 790 } 791 fallthrough 792 default: 793 p.writeByte(byte(ch), 1) 794 } 795 } 796 797 // shift remaining entries down 798 l := copy(p.wsbuf, p.wsbuf[n:]) 799 p.wsbuf = p.wsbuf[:l] 800 } 801 802 // ---------------------------------------------------------------------------- 803 // Printing interface 804 805 // nlines limits n to maxNewlines. 806 func nlimit(n int) int { 807 if n > maxNewlines { 808 n = maxNewlines 809 } 810 return n 811 } 812 813 func mayCombine(prev token.Token, next byte) (b bool) { 814 switch prev { 815 case token.INT: 816 b = next == '.' // 1. 817 case token.ADD: 818 b = next == '+' // ++ 819 case token.SUB: 820 b = next == '-' // -- 821 case token.QUO: 822 b = next == '*' // /* 823 case token.LSS: 824 b = next == '-' || next == '<' // <- or << 825 case token.AND: 826 b = next == '&' || next == '^' // && or &^ 827 } 828 return 829 } 830 831 // print prints a list of "items" (roughly corresponding to syntactic 832 // tokens, but also including whitespace and formatting information). 833 // It is the only print function that should be called directly from 834 // any of the AST printing functions in nodes.go. 835 // 836 // Whitespace is accumulated until a non-whitespace token appears. Any 837 // comments that need to appear before that token are printed first, 838 // taking into account the amount and structure of any pending white- 839 // space for best comment placement. Then, any leftover whitespace is 840 // printed, followed by the actual token. 841 // 842 func (p *printer) print(args ...interface{}) { 843 for _, arg := range args { 844 // information about the current arg 845 var data string 846 var isLit bool 847 var impliedSemi bool // value for p.impliedSemi after this arg 848 849 // record previous opening token, if any 850 switch p.lastTok { 851 case token.ILLEGAL: 852 // ignore (white space) 853 case token.LPAREN, token.LBRACK: 854 p.prevOpen = p.lastTok 855 default: 856 // other tokens followed any opening token 857 p.prevOpen = token.ILLEGAL 858 } 859 860 switch x := arg.(type) { 861 case pmode: 862 // toggle printer mode 863 p.mode ^= x 864 continue 865 866 case whiteSpace: 867 if x == ignore { 868 // don't add ignore's to the buffer; they 869 // may screw up "correcting" unindents (see 870 // LabeledStmt) 871 continue 872 } 873 i := len(p.wsbuf) 874 if i == cap(p.wsbuf) { 875 // Whitespace sequences are very short so this should 876 // never happen. Handle gracefully (but possibly with 877 // bad comment placement) if it does happen. 878 p.writeWhitespace(i) 879 i = 0 880 } 881 p.wsbuf = p.wsbuf[0 : i+1] 882 p.wsbuf[i] = x 883 if x == newline || x == formfeed { 884 // newlines affect the current state (p.impliedSemi) 885 // and not the state after printing arg (impliedSemi) 886 // because comments can be interspersed before the arg 887 // in this case 888 p.impliedSemi = false 889 } 890 p.lastTok = token.ILLEGAL 891 continue 892 893 case *ast.Ident: 894 data = x.Name 895 impliedSemi = true 896 p.lastTok = token.IDENT 897 898 case *ast.BasicLit: 899 data = x.Value 900 isLit = true 901 impliedSemi = true 902 p.lastTok = x.Kind 903 904 case token.Token: 905 s := x.String() 906 if mayCombine(p.lastTok, s[0]) { 907 // the previous and the current token must be 908 // separated by a blank otherwise they combine 909 // into a different incorrect token sequence 910 // (except for token.INT followed by a '.' this 911 // should never happen because it is taken care 912 // of via binary expression formatting) 913 if len(p.wsbuf) != 0 { 914 p.internalError("whitespace buffer not empty") 915 } 916 p.wsbuf = p.wsbuf[0:1] 917 p.wsbuf[0] = ' ' 918 } 919 data = s 920 // some keywords followed by a newline imply a semicolon 921 switch x { 922 case token.BREAK, token.CONTINUE, token.FALLTHROUGH, token.RETURN, 923 token.INC, token.DEC, token.RPAREN, token.RBRACK, token.RBRACE: 924 impliedSemi = true 925 } 926 p.lastTok = x 927 928 case token.Pos: 929 if x.IsValid() { 930 p.pos = p.posFor(x) // accurate position of next item 931 } 932 continue 933 934 case string: 935 // incorrect AST - print error message 936 data = x 937 isLit = true 938 impliedSemi = true 939 p.lastTok = token.STRING 940 941 default: 942 fmt.Fprintf(os.Stderr, "print: unsupported argument %v (%T)\n", arg, arg) 943 panic("go/printer type") 944 } 945 // data != "" 946 947 next := p.pos // estimated/accurate position of next item 948 wroteNewline, droppedFF := p.flush(next, p.lastTok) 949 950 // intersperse extra newlines if present in the source and 951 // if they don't cause extra semicolons (don't do this in 952 // flush as it will cause extra newlines at the end of a file) 953 if !p.impliedSemi { 954 n := nlimit(next.Line - p.pos.Line) 955 // don't exceed maxNewlines if we already wrote one 956 if wroteNewline && n == maxNewlines { 957 n = maxNewlines - 1 958 } 959 if n > 0 { 960 ch := byte('\n') 961 if droppedFF { 962 ch = '\f' // use formfeed since we dropped one before 963 } 964 p.writeByte(ch, n) 965 impliedSemi = false 966 } 967 } 968 969 // the next token starts now - record its line number if requested 970 if p.linePtr != nil { 971 *p.linePtr = p.out.Line 972 p.linePtr = nil 973 } 974 975 p.writeString(next, data, isLit) 976 p.impliedSemi = impliedSemi 977 } 978 } 979 980 // flush prints any pending comments and whitespace occurring textually 981 // before the position of the next token tok. The flush result indicates 982 // if a newline was written or if a formfeed was dropped from the whitespace 983 // buffer. 984 // 985 func (p *printer) flush(next token.Position, tok token.Token) (wroteNewline, droppedFF bool) { 986 if p.commentBefore(next) { 987 // if there are comments before the next item, intersperse them 988 wroteNewline, droppedFF = p.intersperseComments(next, tok) 989 } else { 990 // otherwise, write any leftover whitespace 991 p.writeWhitespace(len(p.wsbuf)) 992 } 993 return 994 } 995 996 // getNode returns the ast.CommentGroup associated with n, if any. 997 func getDoc(n ast.Node) *ast.CommentGroup { 998 switch n := n.(type) { 999 case *ast.Field: 1000 return n.Doc 1001 case *ast.ImportSpec: 1002 return n.Doc 1003 case *ast.ValueSpec: 1004 return n.Doc 1005 case *ast.TypeSpec: 1006 return n.Doc 1007 case *ast.GenDecl: 1008 return n.Doc 1009 case *ast.FuncDecl: 1010 return n.Doc 1011 case *ast.File: 1012 return n.Doc 1013 } 1014 return nil 1015 } 1016 1017 func (p *printer) printNode(node interface{}) error { 1018 // unpack *CommentedNode, if any 1019 var comments []*ast.CommentGroup 1020 if cnode, ok := node.(*CommentedNode); ok { 1021 node = cnode.Node 1022 comments = cnode.Comments 1023 } 1024 1025 if comments != nil { 1026 // commented node - restrict comment list to relevant range 1027 n, ok := node.(ast.Node) 1028 if !ok { 1029 goto unsupported 1030 } 1031 beg := n.Pos() 1032 end := n.End() 1033 // if the node has associated documentation, 1034 // include that commentgroup in the range 1035 // (the comment list is sorted in the order 1036 // of the comment appearance in the source code) 1037 if doc := getDoc(n); doc != nil { 1038 beg = doc.Pos() 1039 } 1040 // token.Pos values are global offsets, we can 1041 // compare them directly 1042 i := 0 1043 for i < len(comments) && comments[i].End() < beg { 1044 i++ 1045 } 1046 j := i 1047 for j < len(comments) && comments[j].Pos() < end { 1048 j++ 1049 } 1050 if i < j { 1051 p.comments = comments[i:j] 1052 } 1053 } else if n, ok := node.(*ast.File); ok { 1054 // use ast.File comments, if any 1055 p.comments = n.Comments 1056 } 1057 1058 // if there are no comments, use node comments 1059 p.useNodeComments = p.comments == nil 1060 1061 // get comments ready for use 1062 p.nextComment() 1063 1064 // format node 1065 switch n := node.(type) { 1066 case ast.Expr: 1067 p.expr(n) 1068 case ast.Stmt: 1069 // A labeled statement will un-indent to position the label. 1070 // Set p.indent to 1 so we don't get indent "underflow". 1071 if _, ok := n.(*ast.LabeledStmt); ok { 1072 p.indent = 1 1073 } 1074 p.stmt(n, false) 1075 case ast.Decl: 1076 p.decl(n) 1077 case ast.Spec: 1078 p.spec(n, 1, false) 1079 case []ast.Stmt: 1080 // A labeled statement will un-indent to position the label. 1081 // Set p.indent to 1 so we don't get indent "underflow". 1082 for _, s := range n { 1083 if _, ok := s.(*ast.LabeledStmt); ok { 1084 p.indent = 1 1085 } 1086 } 1087 p.stmtList(n, 0, false) 1088 case []ast.Decl: 1089 p.declList(n) 1090 case *ast.File: 1091 p.file(n) 1092 default: 1093 goto unsupported 1094 } 1095 1096 return nil 1097 1098 unsupported: 1099 return fmt.Errorf("go/printer: unsupported node type %T", node) 1100 } 1101 1102 // ---------------------------------------------------------------------------- 1103 // Trimmer 1104 1105 // A trimmer is an io.Writer filter for stripping tabwriter.Escape 1106 // characters, trailing blanks and tabs, and for converting formfeed 1107 // and vtab characters into newlines and htabs (in case no tabwriter 1108 // is used). Text bracketed by tabwriter.Escape characters is passed 1109 // through unchanged. 1110 // 1111 type trimmer struct { 1112 output io.Writer 1113 state int 1114 space []byte 1115 } 1116 1117 // trimmer is implemented as a state machine. 1118 // It can be in one of the following states: 1119 const ( 1120 inSpace = iota // inside space 1121 inEscape // inside text bracketed by tabwriter.Escapes 1122 inText // inside text 1123 ) 1124 1125 func (p *trimmer) resetSpace() { 1126 p.state = inSpace 1127 p.space = p.space[0:0] 1128 } 1129 1130 // Design note: It is tempting to eliminate extra blanks occurring in 1131 // whitespace in this function as it could simplify some 1132 // of the blanks logic in the node printing functions. 1133 // However, this would mess up any formatting done by 1134 // the tabwriter. 1135 1136 var aNewline = []byte("\n") 1137 1138 func (p *trimmer) Write(data []byte) (n int, err error) { 1139 // invariants: 1140 // p.state == inSpace: 1141 // p.space is unwritten 1142 // p.state == inEscape, inText: 1143 // data[m:n] is unwritten 1144 m := 0 1145 var b byte 1146 for n, b = range data { 1147 if b == '\v' { 1148 b = '\t' // convert to htab 1149 } 1150 switch p.state { 1151 case inSpace: 1152 switch b { 1153 case '\t', ' ': 1154 p.space = append(p.space, b) 1155 case '\n', '\f': 1156 p.resetSpace() // discard trailing space 1157 _, err = p.output.Write(aNewline) 1158 case tabwriter.Escape: 1159 _, err = p.output.Write(p.space) 1160 p.state = inEscape 1161 m = n + 1 // +1: skip tabwriter.Escape 1162 default: 1163 _, err = p.output.Write(p.space) 1164 p.state = inText 1165 m = n 1166 } 1167 case inEscape: 1168 if b == tabwriter.Escape { 1169 _, err = p.output.Write(data[m:n]) 1170 p.resetSpace() 1171 } 1172 case inText: 1173 switch b { 1174 case '\t', ' ': 1175 _, err = p.output.Write(data[m:n]) 1176 p.resetSpace() 1177 p.space = append(p.space, b) 1178 case '\n', '\f': 1179 _, err = p.output.Write(data[m:n]) 1180 p.resetSpace() 1181 if err == nil { 1182 _, err = p.output.Write(aNewline) 1183 } 1184 case tabwriter.Escape: 1185 _, err = p.output.Write(data[m:n]) 1186 p.state = inEscape 1187 m = n + 1 // +1: skip tabwriter.Escape 1188 } 1189 default: 1190 panic("unreachable") 1191 } 1192 if err != nil { 1193 return 1194 } 1195 } 1196 n = len(data) 1197 1198 switch p.state { 1199 case inEscape, inText: 1200 _, err = p.output.Write(data[m:n]) 1201 p.resetSpace() 1202 } 1203 1204 return 1205 } 1206 1207 // ---------------------------------------------------------------------------- 1208 // Public interface 1209 1210 // A Mode value is a set of flags (or 0). They control printing. 1211 type Mode uint 1212 1213 const ( 1214 RawFormat Mode = 1 << iota // do not use a tabwriter; if set, UseSpaces is ignored 1215 TabIndent // use tabs for indentation independent of UseSpaces 1216 UseSpaces // use spaces instead of tabs for alignment 1217 SourcePos // emit //line comments to preserve original source positions 1218 ) 1219 1220 // A Config node controls the output of Fprint. 1221 type Config struct { 1222 Mode Mode // default: 0 1223 Tabwidth int // default: 8 1224 Indent int // default: 0 (all code is indented at least by this much) 1225 } 1226 1227 // fprint implements Fprint and takes a nodesSizes map for setting up the printer state. 1228 func (cfg *Config) fprint(output io.Writer, fset *token.FileSet, node interface{}, nodeSizes map[ast.Node]int) (err error) { 1229 // print node 1230 var p printer 1231 p.init(cfg, fset, nodeSizes) 1232 if err = p.printNode(node); err != nil { 1233 return 1234 } 1235 // print outstanding comments 1236 p.impliedSemi = false // EOF acts like a newline 1237 p.flush(token.Position{Offset: infinity, Line: infinity}, token.EOF) 1238 1239 // redirect output through a trimmer to eliminate trailing whitespace 1240 // (Input to a tabwriter must be untrimmed since trailing tabs provide 1241 // formatting information. The tabwriter could provide trimming 1242 // functionality but no tabwriter is used when RawFormat is set.) 1243 output = &trimmer{output: output} 1244 1245 // redirect output through a tabwriter if necessary 1246 if cfg.Mode&RawFormat == 0 { 1247 minwidth := cfg.Tabwidth 1248 1249 padchar := byte('\t') 1250 if cfg.Mode&UseSpaces != 0 { 1251 padchar = ' ' 1252 } 1253 1254 twmode := tabwriter.DiscardEmptyColumns 1255 if cfg.Mode&TabIndent != 0 { 1256 minwidth = 0 1257 twmode |= tabwriter.TabIndent 1258 } 1259 1260 output = tabwriter.NewWriter(output, minwidth, cfg.Tabwidth, 1, padchar, twmode) 1261 } 1262 1263 // write printer result via tabwriter/trimmer to output 1264 if _, err = output.Write(p.output); err != nil { 1265 return 1266 } 1267 1268 // flush tabwriter, if any 1269 if tw, _ := output.(*tabwriter.Writer); tw != nil { 1270 err = tw.Flush() 1271 } 1272 1273 return 1274 } 1275 1276 // A CommentedNode bundles an AST node and corresponding comments. 1277 // It may be provided as argument to any of the Fprint functions. 1278 // 1279 type CommentedNode struct { 1280 Node interface{} // *ast.File, or ast.Expr, ast.Decl, ast.Spec, or ast.Stmt 1281 Comments []*ast.CommentGroup 1282 } 1283 1284 // Fprint "pretty-prints" an AST node to output for a given configuration cfg. 1285 // Position information is interpreted relative to the file set fset. 1286 // The node type must be *ast.File, *CommentedNode, []ast.Decl, []ast.Stmt, 1287 // or assignment-compatible to ast.Expr, ast.Decl, ast.Spec, or ast.Stmt. 1288 // 1289 func (cfg *Config) Fprint(output io.Writer, fset *token.FileSet, node interface{}) error { 1290 return cfg.fprint(output, fset, node, make(map[ast.Node]int)) 1291 } 1292 1293 // Fprint "pretty-prints" an AST node to output. 1294 // It calls Config.Fprint with default settings. 1295 // 1296 func Fprint(output io.Writer, fset *token.FileSet, node interface{}) error { 1297 return (&Config{Tabwidth: 8}).Fprint(output, fset, node) 1298 }