clang 19.0.0git
FormatToken.cpp
Go to the documentation of this file.
1//===--- FormatToken.cpp - Format C++ code --------------------------------===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8///
9/// \file
10/// This file implements specific functions of \c FormatTokens and their
11/// roles.
12///
13//===----------------------------------------------------------------------===//
14
15#include "FormatToken.h"
17#include "llvm/ADT/SmallVector.h"
18#include "llvm/Support/Debug.h"
19#include <climits>
20
21namespace clang {
22namespace format {
23
25 static const char *const TokNames[] = {
26#define TYPE(X) #X,
28#undef TYPE
29 nullptr};
30
32 return TokNames[Type];
33 llvm_unreachable("unknown TokenType");
34 return nullptr;
35}
36
37// Sorted common C++ non-keyword types.
39 "clock_t", "int16_t", "int32_t", "int64_t", "int8_t",
40 "intptr_t", "ptrdiff_t", "size_t", "time_t", "uint16_t",
41 "uint32_t", "uint64_t", "uint8_t", "uintptr_t",
42};
43
44bool FormatToken::isTypeName(const LangOptions &LangOpts) const {
45 const bool IsCpp = LangOpts.CXXOperatorNames;
46 return is(TT_TypeName) || Tok.isSimpleTypeSpecifier(LangOpts) ||
47 (IsCpp && is(tok::identifier) &&
48 std::binary_search(CppNonKeywordTypes.begin(),
50}
51
52bool FormatToken::isTypeOrIdentifier(const LangOptions &LangOpts) const {
53 return isTypeName(LangOpts) || isOneOf(tok::kw_auto, tok::identifier);
54}
55
57 assert(is(tok::r_brace));
58 if (!Style.Cpp11BracedListStyle ||
59 Style.AlignAfterOpenBracket != FormatStyle::BAS_BlockIndent) {
60 return false;
61 }
62 const auto *LBrace = MatchingParen;
63 assert(LBrace && LBrace->is(tok::l_brace));
64 if (LBrace->is(BK_BracedInit))
65 return true;
66 if (LBrace->Previous && LBrace->Previous->is(tok::equal))
67 return true;
68 return false;
69}
70
72 // C# Does not indent object initialisers as continuations.
73 if (is(tok::l_brace) && getBlockKind() == BK_BracedInit && Style.isCSharp())
74 return true;
75 if (is(TT_TemplateString) && opensScope())
76 return true;
77 return is(TT_ArrayInitializerLSquare) || is(TT_ProtoExtensionLSquare) ||
78 (is(tok::l_brace) &&
79 (getBlockKind() == BK_Block || is(TT_DictLiteral) ||
80 (!Style.Cpp11BracedListStyle && NestingLevel == 0))) ||
81 (is(tok::less) && Style.isProto());
82}
83
85
87
90 bool DryRun) {
91 if (!State.NextToken || !State.NextToken->Previous)
92 return 0;
93
94 if (Formats.size() <= 1)
95 return 0; // Handled by formatFromToken (1) or avoid severe penalty (0).
96
97 // Ensure that we start on the opening brace.
98 const FormatToken *LBrace =
99 State.NextToken->Previous->getPreviousNonComment();
100 if (!LBrace || !LBrace->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||
101 LBrace->is(BK_Block) || LBrace->is(TT_DictLiteral) ||
102 LBrace->Next->is(TT_DesignatedInitializerPeriod)) {
103 return 0;
104 }
105
106 // Calculate the number of code points we have to format this list. As the
107 // first token is already placed, we have to subtract it.
108 unsigned RemainingCodePoints =
109 Style.ColumnLimit - State.Column + State.NextToken->Previous->ColumnWidth;
110
111 // Find the best ColumnFormat, i.e. the best number of columns to use.
112 const ColumnFormat *Format = getColumnFormat(RemainingCodePoints);
113
114 // If no ColumnFormat can be used, the braced list would generally be
115 // bin-packed. Add a severe penalty to this so that column layouts are
116 // preferred if possible.
117 if (!Format)
118 return 10'000;
119
120 // Format the entire list.
121 unsigned Penalty = 0;
122 unsigned Column = 0;
123 unsigned Item = 0;
124 while (State.NextToken != LBrace->MatchingParen) {
125 bool NewLine = false;
126 unsigned ExtraSpaces = 0;
127
128 // If the previous token was one of our commas, we are now on the next item.
129 if (Item < Commas.size() && State.NextToken->Previous == Commas[Item]) {
130 if (!State.NextToken->isTrailingComment()) {
131 ExtraSpaces += Format->ColumnSizes[Column] - ItemLengths[Item];
132 ++Column;
133 }
134 ++Item;
135 }
136
137 if (Column == Format->Columns || State.NextToken->MustBreakBefore) {
138 Column = 0;
139 NewLine = true;
140 }
141
142 // Place token using the continuation indenter and store the penalty.
143 Penalty += Indenter->addTokenToState(State, NewLine, DryRun, ExtraSpaces);
144 }
145 return Penalty;
146}
147
150 bool DryRun) {
151 // Formatting with 1 Column isn't really a column layout, so we don't need the
152 // special logic here. We can just avoid bin packing any of the parameters.
153 if (Formats.size() == 1 || HasNestedBracedList)
154 State.Stack.back().AvoidBinPacking = true;
155 return 0;
156}
157
158// Returns the lengths in code points between Begin and End (both included),
159// assuming that the entire sequence is put on a single line.
160static unsigned CodePointsBetween(const FormatToken *Begin,
161 const FormatToken *End) {
162 assert(End->TotalLength >= Begin->TotalLength);
163 return End->TotalLength - Begin->TotalLength + Begin->ColumnWidth;
164}
165
167 // FIXME: At some point we might want to do this for other lists, too.
168 if (!Token->MatchingParen ||
169 !Token->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare)) {
170 return;
171 }
172
173 // In C++11 braced list style, we should not format in columns unless they
174 // have many items (20 or more) or we allow bin-packing of function call
175 // arguments.
177 Commas.size() < 19) {
178 return;
179 }
180
181 // Limit column layout for JavaScript array initializers to 20 or more items
182 // for now to introduce it carefully. We can become more aggressive if this
183 // necessary.
184 if (Token->is(TT_ArrayInitializerLSquare) && Commas.size() < 19)
185 return;
186
187 // Column format doesn't really make sense if we don't align after brackets.
189 return;
190
191 FormatToken *ItemBegin = Token->Next;
192 while (ItemBegin->isTrailingComment())
193 ItemBegin = ItemBegin->Next;
194 SmallVector<bool, 8> MustBreakBeforeItem;
195
196 // The lengths of an item if it is put at the end of the line. This includes
197 // trailing comments which are otherwise ignored for column alignment.
198 SmallVector<unsigned, 8> EndOfLineItemLength;
199 MustBreakBeforeItem.reserve(Commas.size() + 1);
200 EndOfLineItemLength.reserve(Commas.size() + 1);
201 ItemLengths.reserve(Commas.size() + 1);
202
203 bool HasSeparatingComment = false;
204 for (unsigned i = 0, e = Commas.size() + 1; i != e; ++i) {
205 assert(ItemBegin);
206 // Skip comments on their own line.
207 while (ItemBegin->HasUnescapedNewline && ItemBegin->isTrailingComment()) {
208 ItemBegin = ItemBegin->Next;
209 HasSeparatingComment = i > 0;
210 }
211
212 MustBreakBeforeItem.push_back(ItemBegin->MustBreakBefore);
213 if (ItemBegin->is(tok::l_brace))
214 HasNestedBracedList = true;
215 const FormatToken *ItemEnd = nullptr;
216 if (i == Commas.size()) {
217 ItemEnd = Token->MatchingParen;
218 const FormatToken *NonCommentEnd = ItemEnd->getPreviousNonComment();
219 ItemLengths.push_back(CodePointsBetween(ItemBegin, NonCommentEnd));
221 !ItemEnd->Previous->isTrailingComment()) {
222 // In Cpp11 braced list style, the } and possibly other subsequent
223 // tokens will need to stay on a line with the last element.
224 while (ItemEnd->Next && !ItemEnd->Next->CanBreakBefore)
225 ItemEnd = ItemEnd->Next;
226 } else {
227 // In other braced lists styles, the "}" can be wrapped to the new line.
228 ItemEnd = Token->MatchingParen->Previous;
229 }
230 } else {
231 ItemEnd = Commas[i];
232 // The comma is counted as part of the item when calculating the length.
233 ItemLengths.push_back(CodePointsBetween(ItemBegin, ItemEnd));
234
235 // Consume trailing comments so the are included in EndOfLineItemLength.
236 if (ItemEnd->Next && !ItemEnd->Next->HasUnescapedNewline &&
237 ItemEnd->Next->isTrailingComment()) {
238 ItemEnd = ItemEnd->Next;
239 }
240 }
241 EndOfLineItemLength.push_back(CodePointsBetween(ItemBegin, ItemEnd));
242 // If there is a trailing comma in the list, the next item will start at the
243 // closing brace. Don't create an extra item for this.
244 if (ItemEnd->getNextNonComment() == Token->MatchingParen)
245 break;
246 ItemBegin = ItemEnd->Next;
247 }
248
249 // Don't use column layout for lists with few elements and in presence of
250 // separating comments.
251 if (Commas.size() < 5 || HasSeparatingComment)
252 return;
253
254 if (Token->NestingLevel != 0 && Token->is(tok::l_brace) && Commas.size() < 19)
255 return;
256
257 // We can never place more than ColumnLimit / 3 items in a row (because of the
258 // spaces and the comma).
259 unsigned MaxItems = Style.ColumnLimit / 3;
260 SmallVector<unsigned> MinSizeInColumn;
261 MinSizeInColumn.reserve(MaxItems);
262 for (unsigned Columns = 1; Columns <= MaxItems; ++Columns) {
263 ColumnFormat Format;
264 Format.Columns = Columns;
265 Format.ColumnSizes.resize(Columns);
266 MinSizeInColumn.assign(Columns, UINT_MAX);
267 Format.LineCount = 1;
268 bool HasRowWithSufficientColumns = false;
269 unsigned Column = 0;
270 for (unsigned i = 0, e = ItemLengths.size(); i != e; ++i) {
271 assert(i < MustBreakBeforeItem.size());
272 if (MustBreakBeforeItem[i] || Column == Columns) {
273 ++Format.LineCount;
274 Column = 0;
275 }
276 if (Column == Columns - 1)
277 HasRowWithSufficientColumns = true;
278 unsigned Length =
279 (Column == Columns - 1) ? EndOfLineItemLength[i] : ItemLengths[i];
280 Format.ColumnSizes[Column] = std::max(Format.ColumnSizes[Column], Length);
281 MinSizeInColumn[Column] = std::min(MinSizeInColumn[Column], Length);
282 ++Column;
283 }
284 // If all rows are terminated early (e.g. by trailing comments), we don't
285 // need to look further.
286 if (!HasRowWithSufficientColumns)
287 break;
288 Format.TotalWidth = Columns - 1; // Width of the N-1 spaces.
289
290 for (unsigned i = 0; i < Columns; ++i)
291 Format.TotalWidth += Format.ColumnSizes[i];
292
293 // Don't use this Format, if the difference between the longest and shortest
294 // element in a column exceeds a threshold to avoid excessive spaces.
295 if ([&] {
296 for (unsigned i = 0; i < Columns - 1; ++i)
297 if (Format.ColumnSizes[i] - MinSizeInColumn[i] > 10)
298 return true;
299 return false;
300 }()) {
301 continue;
302 }
303
304 // Ignore layouts that are bound to violate the column limit.
305 if (Format.TotalWidth > Style.ColumnLimit && Columns > 1)
306 continue;
307
308 Formats.push_back(Format);
309 }
310}
311
312const CommaSeparatedList::ColumnFormat *
313CommaSeparatedList::getColumnFormat(unsigned RemainingCharacters) const {
314 const ColumnFormat *BestFormat = nullptr;
315 for (const ColumnFormat &Format : llvm::reverse(Formats)) {
316 if (Format.TotalWidth <= RemainingCharacters || Format.Columns == 1) {
317 if (BestFormat && Format.LineCount > BestFormat->LineCount)
318 break;
319 BestFormat = &Format;
320 }
321 }
322 return BestFormat;
323}
324
325} // namespace format
326} // namespace clang
This file implements an indenter that manages the indentation of continuations.
This file contains the declaration of the FormatToken, a wrapper around Token with additional informa...
#define LIST_TOKEN_TYPES
Definition: FormatToken.h:27
static constexpr bool isOneOf()
static const char *const TokNames[]
Definition: TokenKinds.cpp:17
SourceLocation Begin
ContinuationIndenter * Indenter
Keeps track of the various options that can be enabled, which controls the dialect of C or C++ that i...
Definition: LangOptions.h:461
Token - This structure provides full information about a lexed token.
Definition: Token.h:36
bool is(tok::TokenKind K) const
is/isNot - Predicates to check if this token is a specific kind, as in "if (Tok.is(tok::l_brace)) {....
Definition: Token.h:99
bool isOneOf(tok::TokenKind K1, tok::TokenKind K2) const
Definition: Token.h:101
bool isSimpleTypeSpecifier(const LangOptions &LangOpts) const
Determine whether the token kind starts a simple-type-specifier.
Definition: Lexer.cpp:78
The base class of the type hierarchy.
Definition: Type.h:1826
unsigned formatFromToken(LineState &State, ContinuationIndenter *Indenter, bool DryRun) override
Apply the special formatting that the given role demands.
unsigned formatAfterToken(LineState &State, ContinuationIndenter *Indenter, bool DryRun) override
Same as formatFromToken, but assumes that the first token has already been set thereby deciding on th...
Definition: FormatToken.cpp:88
void precomputeFormattingInfos(const FormatToken *Token) override
After the TokenAnnotator has finished annotating all the tokens, this function precomputes required i...
virtual void precomputeFormattingInfos(const FormatToken *Token)
After the TokenAnnotator has finished annotating all the tokens, this function precomputes required i...
Definition: FormatToken.cpp:86
const FormatStyle & Style
Definition: FormatToken.h:960
#define UINT_MAX
Definition: limits.h:64
const char * getTokenTypeName(TokenType Type)
Determines the name of a token type.
Definition: FormatToken.cpp:24
static SmallVector< StringRef > CppNonKeywordTypes
Definition: FormatToken.cpp:38
static unsigned CodePointsBetween(const FormatToken *Begin, const FormatToken *End)
TokenType
Determines the semantic type of a syntactic token, e.g.
Definition: FormatToken.h:206
The JSON file list parser is used to communicate input to InstallAPI.
if(T->getSizeExpr()) TRY_TO(TraverseStmt(const_cast< Expr * >(T -> getSizeExpr())))
The FormatStyle is used to configure the formatting to follow specific guidelines.
Definition: Format.h:55
bool Cpp11BracedListStyle
If true, format braced lists as best suited for C++11 braced lists.
Definition: Format.h:2459
@ BAS_DontAlign
Don't align, instead use ContinuationIndentWidth, e.g.:
Definition: Format.h:78
@ BAS_BlockIndent
Always break after an open bracket, if the parameters don't fit on a single line.
Definition: Format.h:99
bool BinPackArguments
If false, a function call's arguments will either be all on the same line or will have one line each.
Definition: Format.h:1193
BracketAlignmentStyle AlignAfterOpenBracket
If true, horizontally aligns arguments after an open bracket.
Definition: Format.h:107
unsigned ColumnLimit
The column limit.
Definition: Format.h:2337
A wrapper around a Token storing information about the whitespace characters preceding it.
Definition: FormatToken.h:291
unsigned NestingLevel
The nesting level of this token, i.e.
Definition: FormatToken.h:514
bool isTypeName(const LangOptions &LangOpts) const
Definition: FormatToken.cpp:44
unsigned CanBreakBefore
true if it is allowed to break before this token.
Definition: FormatToken.h:346
StringRef TokenText
The raw text of the token.
Definition: FormatToken.h:311
bool opensScope() const
Returns whether Tok is ([{ or an opening < of a template or in protos.
Definition: FormatToken.h:699
FormatToken * getNextNonComment() const
Returns the next token ignoring comments.
Definition: FormatToken.h:839
FormatToken * getPreviousNonComment() const
Returns the previous token ignoring comments.
Definition: FormatToken.h:831
BraceBlockKind getBlockKind() const
Definition: FormatToken.h:383
FormatToken * Next
The next token in the unwrapped line.
Definition: FormatToken.h:563
unsigned MustBreakBefore
Whether there must be a line break before this token.
Definition: FormatToken.h:336
unsigned HasUnescapedNewline
Whether there is at least one unescaped newline before the Token.
Definition: FormatToken.h:324
bool isBlockIndentedInitRBrace(const FormatStyle &Style) const
Returns true if this token ends a block indented initializer list.
Definition: FormatToken.cpp:56
bool is(tok::TokenKind Kind) const
Definition: FormatToken.h:603
bool opensBlockOrBlockTypeList(const FormatStyle &Style) const
Returns true if this tokens starts a block-type list, i.e.
Definition: FormatToken.cpp:71
bool isOneOf(A K1, B K2) const
Definition: FormatToken.h:615
bool isTrailingComment() const
Definition: FormatToken.h:773
bool isTypeOrIdentifier(const LangOptions &LangOpts) const
Definition: FormatToken.cpp:52
FormatToken * MatchingParen
If this is a bracket, this points to the matching one.
Definition: FormatToken.h:557
FormatToken * Previous
The previous token in the unwrapped line.
Definition: FormatToken.h:560
The current state when indenting a unwrapped line.