Daniel Jasper | 0df5093 | 2014-12-10 19:00:42 +0000 | [diff] [blame] | 1 | //===--- UnwrappedLineFormatter.cpp - Format C++ code ---------------------===// |
| 2 | // |
| 3 | // The LLVM Compiler Infrastructure |
| 4 | // |
| 5 | // This file is distributed under the University of Illinois Open Source |
| 6 | // License. See LICENSE.TXT for details. |
| 7 | // |
| 8 | //===----------------------------------------------------------------------===// |
| 9 | |
| 10 | #include "UnwrappedLineFormatter.h" |
| 11 | #include "WhitespaceManager.h" |
| 12 | #include "llvm/Support/Debug.h" |
| 13 | |
| 14 | #define DEBUG_TYPE "format-formatter" |
| 15 | |
| 16 | namespace clang { |
| 17 | namespace format { |
| 18 | |
| 19 | namespace { |
| 20 | |
| 21 | bool startsExternCBlock(const AnnotatedLine &Line) { |
| 22 | const FormatToken *Next = Line.First->getNextNonComment(); |
| 23 | const FormatToken *NextNext = Next ? Next->getNextNonComment() : nullptr; |
| 24 | return Line.First->is(tok::kw_extern) && Next && Next->isStringLiteral() && |
| 25 | NextNext && NextNext->is(tok::l_brace); |
| 26 | } |
| 27 | |
| 28 | class LineJoiner { |
| 29 | public: |
Nico Weber | fac2371 | 2015-02-04 15:26:27 +0000 | [diff] [blame] | 30 | LineJoiner(const FormatStyle &Style, const AdditionalKeywords &Keywords) |
| 31 | : Style(Style), Keywords(Keywords) {} |
Daniel Jasper | 0df5093 | 2014-12-10 19:00:42 +0000 | [diff] [blame] | 32 | |
| 33 | /// \brief Calculates how many lines can be merged into 1 starting at \p I. |
| 34 | unsigned |
| 35 | tryFitMultipleLinesInOne(unsigned Indent, |
| 36 | SmallVectorImpl<AnnotatedLine *>::const_iterator I, |
| 37 | SmallVectorImpl<AnnotatedLine *>::const_iterator E) { |
| 38 | // We can never merge stuff if there are trailing line comments. |
| 39 | const AnnotatedLine *TheLine = *I; |
| 40 | if (TheLine->Last->is(TT_LineComment)) |
| 41 | return 0; |
| 42 | |
| 43 | if (Style.ColumnLimit > 0 && Indent > Style.ColumnLimit) |
| 44 | return 0; |
| 45 | |
| 46 | unsigned Limit = |
| 47 | Style.ColumnLimit == 0 ? UINT_MAX : Style.ColumnLimit - Indent; |
| 48 | // If we already exceed the column limit, we set 'Limit' to 0. The different |
| 49 | // tryMerge..() functions can then decide whether to still do merging. |
| 50 | Limit = TheLine->Last->TotalLength > Limit |
| 51 | ? 0 |
| 52 | : Limit - TheLine->Last->TotalLength; |
| 53 | |
| 54 | if (I + 1 == E || I[1]->Type == LT_Invalid || I[1]->First->MustBreakBefore) |
| 55 | return 0; |
| 56 | |
| 57 | // FIXME: TheLine->Level != 0 might or might not be the right check to do. |
| 58 | // If necessary, change to something smarter. |
| 59 | bool MergeShortFunctions = |
| 60 | Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_All || |
| 61 | (Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty && |
| 62 | I[1]->First->is(tok::r_brace)) || |
| 63 | (Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Inline && |
| 64 | TheLine->Level != 0); |
| 65 | |
| 66 | if (TheLine->Last->is(TT_FunctionLBrace) && |
| 67 | TheLine->First != TheLine->Last) { |
| 68 | return MergeShortFunctions ? tryMergeSimpleBlock(I, E, Limit) : 0; |
| 69 | } |
| 70 | if (TheLine->Last->is(tok::l_brace)) { |
| 71 | return Style.BreakBeforeBraces == FormatStyle::BS_Attach |
| 72 | ? tryMergeSimpleBlock(I, E, Limit) |
| 73 | : 0; |
| 74 | } |
| 75 | if (I[1]->First->is(TT_FunctionLBrace) && |
| 76 | Style.BreakBeforeBraces != FormatStyle::BS_Attach) { |
| 77 | if (I[1]->Last->is(TT_LineComment)) |
| 78 | return 0; |
| 79 | |
| 80 | // Check for Limit <= 2 to account for the " {". |
| 81 | if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(TheLine))) |
| 82 | return 0; |
| 83 | Limit -= 2; |
| 84 | |
| 85 | unsigned MergedLines = 0; |
| 86 | if (MergeShortFunctions) { |
| 87 | MergedLines = tryMergeSimpleBlock(I + 1, E, Limit); |
| 88 | // If we managed to merge the block, count the function header, which is |
| 89 | // on a separate line. |
| 90 | if (MergedLines > 0) |
| 91 | ++MergedLines; |
| 92 | } |
| 93 | return MergedLines; |
| 94 | } |
| 95 | if (TheLine->First->is(tok::kw_if)) { |
| 96 | return Style.AllowShortIfStatementsOnASingleLine |
| 97 | ? tryMergeSimpleControlStatement(I, E, Limit) |
| 98 | : 0; |
| 99 | } |
| 100 | if (TheLine->First->isOneOf(tok::kw_for, tok::kw_while)) { |
| 101 | return Style.AllowShortLoopsOnASingleLine |
| 102 | ? tryMergeSimpleControlStatement(I, E, Limit) |
| 103 | : 0; |
| 104 | } |
| 105 | if (TheLine->First->isOneOf(tok::kw_case, tok::kw_default)) { |
| 106 | return Style.AllowShortCaseLabelsOnASingleLine |
| 107 | ? tryMergeShortCaseLabels(I, E, Limit) |
| 108 | : 0; |
| 109 | } |
| 110 | if (TheLine->InPPDirective && |
| 111 | (TheLine->First->HasUnescapedNewline || TheLine->First->IsFirst)) { |
| 112 | return tryMergeSimplePPDirective(I, E, Limit); |
| 113 | } |
| 114 | return 0; |
| 115 | } |
| 116 | |
| 117 | private: |
| 118 | unsigned |
| 119 | tryMergeSimplePPDirective(SmallVectorImpl<AnnotatedLine *>::const_iterator I, |
| 120 | SmallVectorImpl<AnnotatedLine *>::const_iterator E, |
| 121 | unsigned Limit) { |
| 122 | if (Limit == 0) |
| 123 | return 0; |
| 124 | if (!I[1]->InPPDirective || I[1]->First->HasUnescapedNewline) |
| 125 | return 0; |
| 126 | if (I + 2 != E && I[2]->InPPDirective && !I[2]->First->HasUnescapedNewline) |
| 127 | return 0; |
| 128 | if (1 + I[1]->Last->TotalLength > Limit) |
| 129 | return 0; |
| 130 | return 1; |
| 131 | } |
| 132 | |
| 133 | unsigned tryMergeSimpleControlStatement( |
| 134 | SmallVectorImpl<AnnotatedLine *>::const_iterator I, |
| 135 | SmallVectorImpl<AnnotatedLine *>::const_iterator E, unsigned Limit) { |
| 136 | if (Limit == 0) |
| 137 | return 0; |
| 138 | if ((Style.BreakBeforeBraces == FormatStyle::BS_Allman || |
| 139 | Style.BreakBeforeBraces == FormatStyle::BS_GNU) && |
| 140 | (I[1]->First->is(tok::l_brace) && !Style.AllowShortBlocksOnASingleLine)) |
| 141 | return 0; |
| 142 | if (I[1]->InPPDirective != (*I)->InPPDirective || |
| 143 | (I[1]->InPPDirective && I[1]->First->HasUnescapedNewline)) |
| 144 | return 0; |
| 145 | Limit = limitConsideringMacros(I + 1, E, Limit); |
| 146 | AnnotatedLine &Line = **I; |
| 147 | if (Line.Last->isNot(tok::r_paren)) |
| 148 | return 0; |
| 149 | if (1 + I[1]->Last->TotalLength > Limit) |
| 150 | return 0; |
| 151 | if (I[1]->First->isOneOf(tok::semi, tok::kw_if, tok::kw_for, |
| 152 | tok::kw_while, TT_LineComment)) |
| 153 | return 0; |
| 154 | // Only inline simple if's (no nested if or else). |
| 155 | if (I + 2 != E && Line.First->is(tok::kw_if) && |
| 156 | I[2]->First->is(tok::kw_else)) |
| 157 | return 0; |
| 158 | return 1; |
| 159 | } |
| 160 | |
| 161 | unsigned tryMergeShortCaseLabels( |
| 162 | SmallVectorImpl<AnnotatedLine *>::const_iterator I, |
| 163 | SmallVectorImpl<AnnotatedLine *>::const_iterator E, unsigned Limit) { |
| 164 | if (Limit == 0 || I + 1 == E || |
| 165 | I[1]->First->isOneOf(tok::kw_case, tok::kw_default)) |
| 166 | return 0; |
| 167 | unsigned NumStmts = 0; |
| 168 | unsigned Length = 0; |
| 169 | bool InPPDirective = I[0]->InPPDirective; |
| 170 | for (; NumStmts < 3; ++NumStmts) { |
| 171 | if (I + 1 + NumStmts == E) |
| 172 | break; |
| 173 | const AnnotatedLine *Line = I[1 + NumStmts]; |
| 174 | if (Line->InPPDirective != InPPDirective) |
| 175 | break; |
| 176 | if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace)) |
| 177 | break; |
| 178 | if (Line->First->isOneOf(tok::kw_if, tok::kw_for, tok::kw_switch, |
| 179 | tok::kw_while, tok::comment)) |
| 180 | return 0; |
| 181 | Length += I[1 + NumStmts]->Last->TotalLength + 1; // 1 for the space. |
| 182 | } |
| 183 | if (NumStmts == 0 || NumStmts == 3 || Length > Limit) |
| 184 | return 0; |
| 185 | return NumStmts; |
| 186 | } |
| 187 | |
| 188 | unsigned |
| 189 | tryMergeSimpleBlock(SmallVectorImpl<AnnotatedLine *>::const_iterator I, |
| 190 | SmallVectorImpl<AnnotatedLine *>::const_iterator E, |
| 191 | unsigned Limit) { |
| 192 | AnnotatedLine &Line = **I; |
| 193 | |
| 194 | // Don't merge ObjC @ keywords and methods. |
Nico Weber | 33381f5 | 2015-02-07 01:57:32 +0000 | [diff] [blame] | 195 | // FIXME: If an option to allow short exception handling clauses on a single |
| 196 | // line is added, change this to not return for @try and friends. |
Daniel Jasper | 0df5093 | 2014-12-10 19:00:42 +0000 | [diff] [blame] | 197 | if (Style.Language != FormatStyle::LK_Java && |
| 198 | Line.First->isOneOf(tok::at, tok::minus, tok::plus)) |
| 199 | return 0; |
| 200 | |
| 201 | // Check that the current line allows merging. This depends on whether we |
| 202 | // are in a control flow statements as well as several style flags. |
| 203 | if (Line.First->isOneOf(tok::kw_else, tok::kw_case)) |
| 204 | return 0; |
| 205 | if (Line.First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_do, tok::kw_try, |
Nico Weber | fac2371 | 2015-02-04 15:26:27 +0000 | [diff] [blame] | 206 | tok::kw___try, tok::kw_catch, tok::kw___finally, |
| 207 | tok::kw_for, tok::r_brace) || |
| 208 | Line.First->is(Keywords.kw___except)) { |
Daniel Jasper | 0df5093 | 2014-12-10 19:00:42 +0000 | [diff] [blame] | 209 | if (!Style.AllowShortBlocksOnASingleLine) |
| 210 | return 0; |
| 211 | if (!Style.AllowShortIfStatementsOnASingleLine && |
| 212 | Line.First->is(tok::kw_if)) |
| 213 | return 0; |
| 214 | if (!Style.AllowShortLoopsOnASingleLine && |
| 215 | Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for)) |
| 216 | return 0; |
| 217 | // FIXME: Consider an option to allow short exception handling clauses on |
| 218 | // a single line. |
Nico Weber | fac2371 | 2015-02-04 15:26:27 +0000 | [diff] [blame] | 219 | // FIXME: This isn't covered by tests. |
| 220 | // FIXME: For catch, __except, __finally the first token on the line |
| 221 | // is '}', so this isn't correct here. |
| 222 | if (Line.First->isOneOf(tok::kw_try, tok::kw___try, tok::kw_catch, |
| 223 | Keywords.kw___except, tok::kw___finally)) |
Daniel Jasper | 0df5093 | 2014-12-10 19:00:42 +0000 | [diff] [blame] | 224 | return 0; |
| 225 | } |
| 226 | |
| 227 | FormatToken *Tok = I[1]->First; |
| 228 | if (Tok->is(tok::r_brace) && !Tok->MustBreakBefore && |
| 229 | (Tok->getNextNonComment() == nullptr || |
| 230 | Tok->getNextNonComment()->is(tok::semi))) { |
| 231 | // We merge empty blocks even if the line exceeds the column limit. |
| 232 | Tok->SpacesRequiredBefore = 0; |
| 233 | Tok->CanBreakBefore = true; |
| 234 | return 1; |
| 235 | } else if (Limit != 0 && Line.First->isNot(tok::kw_namespace) && |
| 236 | !startsExternCBlock(Line)) { |
| 237 | // We don't merge short records. |
| 238 | if (Line.First->isOneOf(tok::kw_class, tok::kw_union, tok::kw_struct)) |
| 239 | return 0; |
| 240 | |
| 241 | // Check that we still have three lines and they fit into the limit. |
| 242 | if (I + 2 == E || I[2]->Type == LT_Invalid) |
| 243 | return 0; |
| 244 | Limit = limitConsideringMacros(I + 2, E, Limit); |
| 245 | |
| 246 | if (!nextTwoLinesFitInto(I, Limit)) |
| 247 | return 0; |
| 248 | |
| 249 | // Second, check that the next line does not contain any braces - if it |
| 250 | // does, readability declines when putting it into a single line. |
| 251 | if (I[1]->Last->is(TT_LineComment)) |
| 252 | return 0; |
| 253 | do { |
| 254 | if (Tok->is(tok::l_brace) && Tok->BlockKind != BK_BracedInit) |
| 255 | return 0; |
| 256 | Tok = Tok->Next; |
| 257 | } while (Tok); |
| 258 | |
| 259 | // Last, check that the third line starts with a closing brace. |
| 260 | Tok = I[2]->First; |
| 261 | if (Tok->isNot(tok::r_brace)) |
| 262 | return 0; |
| 263 | |
| 264 | return 2; |
| 265 | } |
| 266 | return 0; |
| 267 | } |
| 268 | |
| 269 | /// Returns the modified column limit for \p I if it is inside a macro and |
| 270 | /// needs a trailing '\'. |
| 271 | unsigned |
| 272 | limitConsideringMacros(SmallVectorImpl<AnnotatedLine *>::const_iterator I, |
| 273 | SmallVectorImpl<AnnotatedLine *>::const_iterator E, |
| 274 | unsigned Limit) { |
| 275 | if (I[0]->InPPDirective && I + 1 != E && |
| 276 | !I[1]->First->HasUnescapedNewline && !I[1]->First->is(tok::eof)) { |
| 277 | return Limit < 2 ? 0 : Limit - 2; |
| 278 | } |
| 279 | return Limit; |
| 280 | } |
| 281 | |
| 282 | bool nextTwoLinesFitInto(SmallVectorImpl<AnnotatedLine *>::const_iterator I, |
| 283 | unsigned Limit) { |
| 284 | if (I[1]->First->MustBreakBefore || I[2]->First->MustBreakBefore) |
| 285 | return false; |
| 286 | return 1 + I[1]->Last->TotalLength + 1 + I[2]->Last->TotalLength <= Limit; |
| 287 | } |
| 288 | |
| 289 | bool containsMustBreak(const AnnotatedLine *Line) { |
| 290 | for (const FormatToken *Tok = Line->First; Tok; Tok = Tok->Next) { |
| 291 | if (Tok->MustBreakBefore) |
| 292 | return true; |
| 293 | } |
| 294 | return false; |
| 295 | } |
| 296 | |
| 297 | const FormatStyle &Style; |
Nico Weber | fac2371 | 2015-02-04 15:26:27 +0000 | [diff] [blame] | 298 | const AdditionalKeywords &Keywords; |
Daniel Jasper | 0df5093 | 2014-12-10 19:00:42 +0000 | [diff] [blame] | 299 | }; |
| 300 | |
| 301 | class NoColumnLimitFormatter { |
| 302 | public: |
| 303 | NoColumnLimitFormatter(ContinuationIndenter *Indenter) : Indenter(Indenter) {} |
| 304 | |
| 305 | /// \brief Formats the line starting at \p State, simply keeping all of the |
| 306 | /// input's line breaking decisions. |
| 307 | void format(unsigned FirstIndent, const AnnotatedLine *Line) { |
| 308 | LineState State = |
| 309 | Indenter->getInitialState(FirstIndent, Line, /*DryRun=*/false); |
| 310 | while (State.NextToken) { |
| 311 | bool Newline = |
| 312 | Indenter->mustBreak(State) || |
| 313 | (Indenter->canBreak(State) && State.NextToken->NewlinesBefore > 0); |
| 314 | Indenter->addTokenToState(State, Newline, /*DryRun=*/false); |
| 315 | } |
| 316 | } |
| 317 | |
| 318 | private: |
| 319 | ContinuationIndenter *Indenter; |
| 320 | }; |
| 321 | |
Daniel Jasper | d1c1373 | 2015-01-23 19:37:25 +0000 | [diff] [blame] | 322 | |
| 323 | static void markFinalized(FormatToken *Tok) { |
| 324 | for (; Tok; Tok = Tok->Next) { |
| 325 | Tok->Finalized = true; |
| 326 | for (AnnotatedLine *Child : Tok->Children) |
| 327 | markFinalized(Child->First); |
| 328 | } |
| 329 | } |
| 330 | |
Daniel Jasper | 0df5093 | 2014-12-10 19:00:42 +0000 | [diff] [blame] | 331 | } // namespace |
| 332 | |
| 333 | unsigned |
| 334 | UnwrappedLineFormatter::format(const SmallVectorImpl<AnnotatedLine *> &Lines, |
| 335 | bool DryRun, int AdditionalIndent, |
| 336 | bool FixBadIndentation) { |
Nico Weber | fac2371 | 2015-02-04 15:26:27 +0000 | [diff] [blame] | 337 | LineJoiner Joiner(Style, Keywords); |
Daniel Jasper | 0df5093 | 2014-12-10 19:00:42 +0000 | [diff] [blame] | 338 | |
| 339 | // Try to look up already computed penalty in DryRun-mode. |
| 340 | std::pair<const SmallVectorImpl<AnnotatedLine *> *, unsigned> CacheKey( |
| 341 | &Lines, AdditionalIndent); |
| 342 | auto CacheIt = PenaltyCache.find(CacheKey); |
| 343 | if (DryRun && CacheIt != PenaltyCache.end()) |
| 344 | return CacheIt->second; |
| 345 | |
| 346 | assert(!Lines.empty()); |
| 347 | unsigned Penalty = 0; |
| 348 | std::vector<int> IndentForLevel; |
| 349 | for (unsigned i = 0, e = Lines[0]->Level; i != e; ++i) |
| 350 | IndentForLevel.push_back(Style.IndentWidth * i + AdditionalIndent); |
| 351 | const AnnotatedLine *PreviousLine = nullptr; |
| 352 | for (SmallVectorImpl<AnnotatedLine *>::const_iterator I = Lines.begin(), |
| 353 | E = Lines.end(); |
| 354 | I != E; ++I) { |
| 355 | const AnnotatedLine &TheLine = **I; |
| 356 | const FormatToken *FirstTok = TheLine.First; |
| 357 | int Offset = getIndentOffset(*FirstTok); |
| 358 | |
| 359 | // Determine indent and try to merge multiple unwrapped lines. |
| 360 | unsigned Indent; |
| 361 | if (TheLine.InPPDirective) { |
| 362 | Indent = TheLine.Level * Style.IndentWidth; |
| 363 | } else { |
| 364 | while (IndentForLevel.size() <= TheLine.Level) |
| 365 | IndentForLevel.push_back(-1); |
| 366 | IndentForLevel.resize(TheLine.Level + 1); |
| 367 | Indent = getIndent(IndentForLevel, TheLine.Level); |
| 368 | } |
| 369 | unsigned LevelIndent = Indent; |
| 370 | if (static_cast<int>(Indent) + Offset >= 0) |
| 371 | Indent += Offset; |
| 372 | |
| 373 | // Merge multiple lines if possible. |
| 374 | unsigned MergedLines = Joiner.tryFitMultipleLinesInOne(Indent, I, E); |
| 375 | if (MergedLines > 0 && Style.ColumnLimit == 0) { |
| 376 | // Disallow line merging if there is a break at the start of one of the |
| 377 | // input lines. |
| 378 | for (unsigned i = 0; i < MergedLines; ++i) { |
| 379 | if (I[i + 1]->First->NewlinesBefore > 0) |
| 380 | MergedLines = 0; |
| 381 | } |
| 382 | } |
| 383 | if (!DryRun) { |
| 384 | for (unsigned i = 0; i < MergedLines; ++i) { |
| 385 | join(*I[i], *I[i + 1]); |
| 386 | } |
| 387 | } |
| 388 | I += MergedLines; |
| 389 | |
| 390 | bool FixIndentation = |
| 391 | FixBadIndentation && (LevelIndent != FirstTok->OriginalColumn); |
| 392 | if (TheLine.First->is(tok::eof)) { |
| 393 | if (PreviousLine && PreviousLine->Affected && !DryRun) { |
| 394 | // Remove the file's trailing whitespace. |
| 395 | unsigned Newlines = std::min(FirstTok->NewlinesBefore, 1u); |
| 396 | Whitespaces->replaceWhitespace(*TheLine.First, Newlines, |
| 397 | /*IndentLevel=*/0, /*Spaces=*/0, |
| 398 | /*TargetColumn=*/0); |
| 399 | } |
| 400 | } else if (TheLine.Type != LT_Invalid && |
| 401 | (TheLine.Affected || FixIndentation)) { |
| 402 | if (FirstTok->WhitespaceRange.isValid()) { |
| 403 | if (!DryRun) |
| 404 | formatFirstToken(*TheLine.First, PreviousLine, TheLine.Level, Indent, |
| 405 | TheLine.InPPDirective); |
| 406 | } else { |
| 407 | Indent = LevelIndent = FirstTok->OriginalColumn; |
| 408 | } |
| 409 | |
| 410 | // If everything fits on a single line, just put it there. |
| 411 | unsigned ColumnLimit = Style.ColumnLimit; |
| 412 | if (I + 1 != E) { |
| 413 | AnnotatedLine *NextLine = I[1]; |
| 414 | if (NextLine->InPPDirective && !NextLine->First->HasUnescapedNewline) |
| 415 | ColumnLimit = getColumnLimit(TheLine.InPPDirective); |
| 416 | } |
| 417 | |
| 418 | if (TheLine.Last->TotalLength + Indent <= ColumnLimit || |
| 419 | TheLine.Type == LT_ImportStatement) { |
| 420 | LineState State = Indenter->getInitialState(Indent, &TheLine, DryRun); |
| 421 | while (State.NextToken) { |
Daniel Jasper | 47b35ae | 2015-01-29 10:47:14 +0000 | [diff] [blame] | 422 | formatChildren(State, /*Newline=*/false, DryRun, Penalty); |
Daniel Jasper | 0df5093 | 2014-12-10 19:00:42 +0000 | [diff] [blame] | 423 | Indenter->addTokenToState(State, /*Newline=*/false, DryRun); |
| 424 | } |
| 425 | } else if (Style.ColumnLimit == 0) { |
| 426 | // FIXME: Implement nested blocks for ColumnLimit = 0. |
| 427 | NoColumnLimitFormatter Formatter(Indenter); |
| 428 | if (!DryRun) |
| 429 | Formatter.format(Indent, &TheLine); |
| 430 | } else { |
| 431 | Penalty += format(TheLine, Indent, DryRun); |
| 432 | } |
| 433 | |
| 434 | if (!TheLine.InPPDirective) |
| 435 | IndentForLevel[TheLine.Level] = LevelIndent; |
| 436 | } else if (TheLine.ChildrenAffected) { |
| 437 | format(TheLine.Children, DryRun); |
| 438 | } else { |
| 439 | // Format the first token if necessary, and notify the WhitespaceManager |
| 440 | // about the unchanged whitespace. |
| 441 | for (FormatToken *Tok = TheLine.First; Tok; Tok = Tok->Next) { |
| 442 | if (Tok == TheLine.First && (Tok->NewlinesBefore > 0 || Tok->IsFirst)) { |
| 443 | unsigned LevelIndent = Tok->OriginalColumn; |
| 444 | if (!DryRun) { |
| 445 | // Remove trailing whitespace of the previous line. |
| 446 | if ((PreviousLine && PreviousLine->Affected) || |
| 447 | TheLine.LeadingEmptyLinesAffected) { |
| 448 | formatFirstToken(*Tok, PreviousLine, TheLine.Level, LevelIndent, |
| 449 | TheLine.InPPDirective); |
| 450 | } else { |
| 451 | Whitespaces->addUntouchableToken(*Tok, TheLine.InPPDirective); |
| 452 | } |
| 453 | } |
| 454 | |
| 455 | if (static_cast<int>(LevelIndent) - Offset >= 0) |
| 456 | LevelIndent -= Offset; |
| 457 | if (Tok->isNot(tok::comment) && !TheLine.InPPDirective) |
| 458 | IndentForLevel[TheLine.Level] = LevelIndent; |
| 459 | } else if (!DryRun) { |
| 460 | Whitespaces->addUntouchableToken(*Tok, TheLine.InPPDirective); |
| 461 | } |
| 462 | } |
| 463 | } |
Daniel Jasper | d1c1373 | 2015-01-23 19:37:25 +0000 | [diff] [blame] | 464 | if (!DryRun) |
| 465 | markFinalized(TheLine.First); |
Daniel Jasper | 0df5093 | 2014-12-10 19:00:42 +0000 | [diff] [blame] | 466 | PreviousLine = *I; |
| 467 | } |
| 468 | PenaltyCache[CacheKey] = Penalty; |
| 469 | return Penalty; |
| 470 | } |
| 471 | |
| 472 | unsigned UnwrappedLineFormatter::format(const AnnotatedLine &Line, |
| 473 | unsigned FirstIndent, bool DryRun) { |
| 474 | LineState State = Indenter->getInitialState(FirstIndent, &Line, DryRun); |
| 475 | |
| 476 | // If the ObjC method declaration does not fit on a line, we should format |
| 477 | // it with one arg per line. |
| 478 | if (State.Line->Type == LT_ObjCMethodDecl) |
| 479 | State.Stack.back().BreakBeforeParameter = true; |
| 480 | |
| 481 | // Find best solution in solution space. |
| 482 | return analyzeSolutionSpace(State, DryRun); |
| 483 | } |
| 484 | |
| 485 | void UnwrappedLineFormatter::formatFirstToken(FormatToken &RootToken, |
| 486 | const AnnotatedLine *PreviousLine, |
| 487 | unsigned IndentLevel, |
| 488 | unsigned Indent, |
| 489 | bool InPPDirective) { |
| 490 | unsigned Newlines = |
| 491 | std::min(RootToken.NewlinesBefore, Style.MaxEmptyLinesToKeep + 1); |
| 492 | // Remove empty lines before "}" where applicable. |
| 493 | if (RootToken.is(tok::r_brace) && |
| 494 | (!RootToken.Next || |
| 495 | (RootToken.Next->is(tok::semi) && !RootToken.Next->Next))) |
| 496 | Newlines = std::min(Newlines, 1u); |
| 497 | if (Newlines == 0 && !RootToken.IsFirst) |
| 498 | Newlines = 1; |
| 499 | if (RootToken.IsFirst && !RootToken.HasUnescapedNewline) |
| 500 | Newlines = 0; |
| 501 | |
| 502 | // Remove empty lines after "{". |
| 503 | if (!Style.KeepEmptyLinesAtTheStartOfBlocks && PreviousLine && |
| 504 | PreviousLine->Last->is(tok::l_brace) && |
| 505 | PreviousLine->First->isNot(tok::kw_namespace) && |
| 506 | !startsExternCBlock(*PreviousLine)) |
| 507 | Newlines = 1; |
| 508 | |
| 509 | // Insert extra new line before access specifiers. |
| 510 | if (PreviousLine && PreviousLine->Last->isOneOf(tok::semi, tok::r_brace) && |
| 511 | RootToken.isAccessSpecifier() && RootToken.NewlinesBefore == 1) |
| 512 | ++Newlines; |
| 513 | |
| 514 | // Remove empty lines after access specifiers. |
Daniel Jasper | ac5c97e3 | 2015-03-09 08:13:55 +0000 | [diff] [blame^] | 515 | if (PreviousLine && PreviousLine->First->isAccessSpecifier() && |
| 516 | (!PreviousLine->InPPDirective || !RootToken.HasUnescapedNewline)) |
Daniel Jasper | 0df5093 | 2014-12-10 19:00:42 +0000 | [diff] [blame] | 517 | Newlines = std::min(1u, Newlines); |
| 518 | |
| 519 | Whitespaces->replaceWhitespace(RootToken, Newlines, IndentLevel, Indent, |
| 520 | Indent, InPPDirective && |
| 521 | !RootToken.HasUnescapedNewline); |
| 522 | } |
| 523 | |
| 524 | /// \brief Get the indent of \p Level from \p IndentForLevel. |
| 525 | /// |
| 526 | /// \p IndentForLevel must contain the indent for the level \c l |
| 527 | /// at \p IndentForLevel[l], or a value < 0 if the indent for |
| 528 | /// that level is unknown. |
| 529 | unsigned UnwrappedLineFormatter::getIndent(ArrayRef<int> IndentForLevel, |
| 530 | unsigned Level) { |
| 531 | if (IndentForLevel[Level] != -1) |
| 532 | return IndentForLevel[Level]; |
| 533 | if (Level == 0) |
| 534 | return 0; |
| 535 | return getIndent(IndentForLevel, Level - 1) + Style.IndentWidth; |
| 536 | } |
| 537 | |
| 538 | void UnwrappedLineFormatter::join(AnnotatedLine &A, const AnnotatedLine &B) { |
| 539 | assert(!A.Last->Next); |
| 540 | assert(!B.First->Previous); |
| 541 | if (B.Affected) |
| 542 | A.Affected = true; |
| 543 | A.Last->Next = B.First; |
| 544 | B.First->Previous = A.Last; |
| 545 | B.First->CanBreakBefore = true; |
| 546 | unsigned LengthA = A.Last->TotalLength + B.First->SpacesRequiredBefore; |
| 547 | for (FormatToken *Tok = B.First; Tok; Tok = Tok->Next) { |
| 548 | Tok->TotalLength += LengthA; |
| 549 | A.Last = Tok; |
| 550 | } |
| 551 | } |
| 552 | |
| 553 | unsigned UnwrappedLineFormatter::analyzeSolutionSpace(LineState &InitialState, |
| 554 | bool DryRun) { |
| 555 | std::set<LineState *, CompareLineStatePointers> Seen; |
| 556 | |
| 557 | // Increasing count of \c StateNode items we have created. This is used to |
| 558 | // create a deterministic order independent of the container. |
| 559 | unsigned Count = 0; |
| 560 | QueueType Queue; |
| 561 | |
| 562 | // Insert start element into queue. |
| 563 | StateNode *Node = |
| 564 | new (Allocator.Allocate()) StateNode(InitialState, false, nullptr); |
| 565 | Queue.push(QueueItem(OrderedPenalty(0, Count), Node)); |
| 566 | ++Count; |
| 567 | |
| 568 | unsigned Penalty = 0; |
| 569 | |
| 570 | // While not empty, take first element and follow edges. |
| 571 | while (!Queue.empty()) { |
| 572 | Penalty = Queue.top().first.first; |
| 573 | StateNode *Node = Queue.top().second; |
| 574 | if (!Node->State.NextToken) { |
| 575 | DEBUG(llvm::dbgs() << "\n---\nPenalty for line: " << Penalty << "\n"); |
| 576 | break; |
| 577 | } |
| 578 | Queue.pop(); |
| 579 | |
| 580 | // Cut off the analysis of certain solutions if the analysis gets too |
| 581 | // complex. See description of IgnoreStackForComparison. |
| 582 | if (Count > 10000) |
| 583 | Node->State.IgnoreStackForComparison = true; |
| 584 | |
| 585 | if (!Seen.insert(&Node->State).second) |
| 586 | // State already examined with lower penalty. |
| 587 | continue; |
| 588 | |
| 589 | FormatDecision LastFormat = Node->State.NextToken->Decision; |
| 590 | if (LastFormat == FD_Unformatted || LastFormat == FD_Continue) |
| 591 | addNextStateToQueue(Penalty, Node, /*NewLine=*/false, &Count, &Queue); |
| 592 | if (LastFormat == FD_Unformatted || LastFormat == FD_Break) |
| 593 | addNextStateToQueue(Penalty, Node, /*NewLine=*/true, &Count, &Queue); |
| 594 | } |
| 595 | |
| 596 | if (Queue.empty()) { |
| 597 | // We were unable to find a solution, do nothing. |
| 598 | // FIXME: Add diagnostic? |
| 599 | DEBUG(llvm::dbgs() << "Could not find a solution.\n"); |
| 600 | return 0; |
| 601 | } |
| 602 | |
| 603 | // Reconstruct the solution. |
| 604 | if (!DryRun) |
| 605 | reconstructPath(InitialState, Queue.top().second); |
| 606 | |
| 607 | DEBUG(llvm::dbgs() << "Total number of analyzed states: " << Count << "\n"); |
| 608 | DEBUG(llvm::dbgs() << "---\n"); |
| 609 | |
| 610 | return Penalty; |
| 611 | } |
| 612 | |
NAKAMURA Takumi | e9ac8b4 | 2014-12-17 14:46:56 +0000 | [diff] [blame] | 613 | #ifndef NDEBUG |
Daniel Jasper | 11a0ac6 | 2014-12-12 09:40:58 +0000 | [diff] [blame] | 614 | static void printLineState(const LineState &State) { |
| 615 | llvm::dbgs() << "State: "; |
| 616 | for (const ParenState &P : State.Stack) { |
| 617 | llvm::dbgs() << P.Indent << "|" << P.LastSpace << "|" << P.NestedBlockIndent |
| 618 | << " "; |
| 619 | } |
| 620 | llvm::dbgs() << State.NextToken->TokenText << "\n"; |
| 621 | } |
NAKAMURA Takumi | e9ac8b4 | 2014-12-17 14:46:56 +0000 | [diff] [blame] | 622 | #endif |
Daniel Jasper | 11a0ac6 | 2014-12-12 09:40:58 +0000 | [diff] [blame] | 623 | |
Daniel Jasper | 0df5093 | 2014-12-10 19:00:42 +0000 | [diff] [blame] | 624 | void UnwrappedLineFormatter::reconstructPath(LineState &State, |
| 625 | StateNode *Current) { |
| 626 | std::deque<StateNode *> Path; |
| 627 | // We do not need a break before the initial token. |
| 628 | while (Current->Previous) { |
| 629 | Path.push_front(Current); |
| 630 | Current = Current->Previous; |
| 631 | } |
| 632 | for (std::deque<StateNode *>::iterator I = Path.begin(), E = Path.end(); |
| 633 | I != E; ++I) { |
| 634 | unsigned Penalty = 0; |
| 635 | formatChildren(State, (*I)->NewLine, /*DryRun=*/false, Penalty); |
| 636 | Penalty += Indenter->addTokenToState(State, (*I)->NewLine, false); |
| 637 | |
| 638 | DEBUG({ |
Daniel Jasper | 11a0ac6 | 2014-12-12 09:40:58 +0000 | [diff] [blame] | 639 | printLineState((*I)->Previous->State); |
Daniel Jasper | 0df5093 | 2014-12-10 19:00:42 +0000 | [diff] [blame] | 640 | if ((*I)->NewLine) { |
| 641 | llvm::dbgs() << "Penalty for placing " |
| 642 | << (*I)->Previous->State.NextToken->Tok.getName() << ": " |
| 643 | << Penalty << "\n"; |
| 644 | } |
| 645 | }); |
| 646 | } |
| 647 | } |
| 648 | |
| 649 | void UnwrappedLineFormatter::addNextStateToQueue(unsigned Penalty, |
| 650 | StateNode *PreviousNode, |
| 651 | bool NewLine, unsigned *Count, |
| 652 | QueueType *Queue) { |
| 653 | if (NewLine && !Indenter->canBreak(PreviousNode->State)) |
| 654 | return; |
| 655 | if (!NewLine && Indenter->mustBreak(PreviousNode->State)) |
| 656 | return; |
| 657 | |
| 658 | StateNode *Node = new (Allocator.Allocate()) |
| 659 | StateNode(PreviousNode->State, NewLine, PreviousNode); |
| 660 | if (!formatChildren(Node->State, NewLine, /*DryRun=*/true, Penalty)) |
| 661 | return; |
| 662 | |
| 663 | Penalty += Indenter->addTokenToState(Node->State, NewLine, true); |
| 664 | |
| 665 | Queue->push(QueueItem(OrderedPenalty(Penalty, *Count), Node)); |
| 666 | ++(*Count); |
| 667 | } |
| 668 | |
| 669 | bool UnwrappedLineFormatter::formatChildren(LineState &State, bool NewLine, |
| 670 | bool DryRun, unsigned &Penalty) { |
Daniel Jasper | 0df5093 | 2014-12-10 19:00:42 +0000 | [diff] [blame] | 671 | const FormatToken *LBrace = State.NextToken->getPreviousNonComment(); |
Daniel Jasper | 47b35ae | 2015-01-29 10:47:14 +0000 | [diff] [blame] | 672 | FormatToken &Previous = *State.NextToken->Previous; |
Daniel Jasper | 0df5093 | 2014-12-10 19:00:42 +0000 | [diff] [blame] | 673 | if (!LBrace || LBrace->isNot(tok::l_brace) || LBrace->BlockKind != BK_Block || |
| 674 | Previous.Children.size() == 0) |
| 675 | // The previous token does not open a block. Nothing to do. We don't |
| 676 | // assert so that we can simply call this function for all tokens. |
| 677 | return true; |
| 678 | |
| 679 | if (NewLine) { |
Daniel Jasper | 11a0ac6 | 2014-12-12 09:40:58 +0000 | [diff] [blame] | 680 | int AdditionalIndent = State.Stack.back().Indent - |
| 681 | Previous.Children[0]->Level * Style.IndentWidth; |
Daniel Jasper | 0df5093 | 2014-12-10 19:00:42 +0000 | [diff] [blame] | 682 | |
| 683 | Penalty += format(Previous.Children, DryRun, AdditionalIndent, |
| 684 | /*FixBadIndentation=*/true); |
| 685 | return true; |
| 686 | } |
| 687 | |
| 688 | if (Previous.Children[0]->First->MustBreakBefore) |
| 689 | return false; |
| 690 | |
| 691 | // Cannot merge multiple statements into a single line. |
| 692 | if (Previous.Children.size() > 1) |
| 693 | return false; |
| 694 | |
| 695 | // Cannot merge into one line if this line ends on a comment. |
| 696 | if (Previous.is(tok::comment)) |
| 697 | return false; |
| 698 | |
| 699 | // We can't put the closing "}" on a line with a trailing comment. |
| 700 | if (Previous.Children[0]->Last->isTrailingComment()) |
| 701 | return false; |
| 702 | |
| 703 | // If the child line exceeds the column limit, we wouldn't want to merge it. |
| 704 | // We add +2 for the trailing " }". |
| 705 | if (Style.ColumnLimit > 0 && |
| 706 | Previous.Children[0]->Last->TotalLength + State.Column + 2 > |
| 707 | Style.ColumnLimit) |
| 708 | return false; |
| 709 | |
| 710 | if (!DryRun) { |
| 711 | Whitespaces->replaceWhitespace( |
| 712 | *Previous.Children[0]->First, |
| 713 | /*Newlines=*/0, /*IndentLevel=*/0, /*Spaces=*/1, |
| 714 | /*StartOfTokenColumn=*/State.Column, State.Line->InPPDirective); |
| 715 | } |
| 716 | Penalty += format(*Previous.Children[0], State.Column + 1, DryRun); |
| 717 | |
| 718 | State.Column += 1 + Previous.Children[0]->Last->TotalLength; |
| 719 | return true; |
| 720 | } |
| 721 | |
| 722 | } // namespace format |
| 723 | } // namespace clang |