1//===- Twine.h - Fast Temporary String Concatenation ------------*- C++ -*-===//
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#ifndef LLVM_ADT_TWINE_H
11#define LLVM_ADT_TWINE_H
12
13#include "llvm/ADT/SmallVector.h"
14#include "llvm/ADT/StringRef.h"
15#include "llvm/Support/ErrorHandling.h"
16#include <cassert>
17#include <cstdint>
18#include <string>
19
20namespace llvm {
21
22 class formatv_object_base;
23 class raw_ostream;
24
25 /// Twine - A lightweight data structure for efficiently representing the
26 /// concatenation of temporary values as strings.
27 ///
28 /// A Twine is a kind of rope, it represents a concatenated string using a
29 /// binary-tree, where the string is the preorder of the nodes. Since the
30 /// Twine can be efficiently rendered into a buffer when its result is used,
31 /// it avoids the cost of generating temporary values for intermediate string
32 /// results -- particularly in cases when the Twine result is never
33 /// required. By explicitly tracking the type of leaf nodes, we can also avoid
34 /// the creation of temporary strings for conversions operations (such as
35 /// appending an integer to a string).
36 ///
37 /// A Twine is not intended for use directly and should not be stored, its
38 /// implementation relies on the ability to store pointers to temporary stack
39 /// objects which may be deallocated at the end of a statement. Twines should
40 /// only be used accepted as const references in arguments, when an API wishes
41 /// to accept possibly-concatenated strings.
42 ///
43 /// Twines support a special 'null' value, which always concatenates to form
44 /// itself, and renders as an empty string. This can be returned from APIs to
45 /// effectively nullify any concatenations performed on the result.
46 ///
47 /// \b Implementation
48 ///
49 /// Given the nature of a Twine, it is not possible for the Twine's
50 /// concatenation method to construct interior nodes; the result must be
51 /// represented inside the returned value. For this reason a Twine object
52 /// actually holds two values, the left- and right-hand sides of a
53 /// concatenation. We also have nullary Twine objects, which are effectively
54 /// sentinel values that represent empty strings.
55 ///
56 /// Thus, a Twine can effectively have zero, one, or two children. The \see
57 /// isNullary(), \see isUnary(), and \see isBinary() predicates exist for
58 /// testing the number of children.
59 ///
60 /// We maintain a number of invariants on Twine objects (FIXME: Why):
61 /// - Nullary twines are always represented with their Kind on the left-hand
62 /// side, and the Empty kind on the right-hand side.
63 /// - Unary twines are always represented with the value on the left-hand
64 /// side, and the Empty kind on the right-hand side.
65 /// - If a Twine has another Twine as a child, that child should always be
66 /// binary (otherwise it could have been folded into the parent).
67 ///
68 /// These invariants are check by \see isValid().
69 ///
70 /// \b Efficiency Considerations
71 ///
72 /// The Twine is designed to yield efficient and small code for common
73 /// situations. For this reason, the concat() method is inlined so that
74 /// concatenations of leaf nodes can be optimized into stores directly into a
75 /// single stack allocated object.
76 ///
77 /// In practice, not all compilers can be trusted to optimize concat() fully,
78 /// so we provide two additional methods (and accompanying operator+
79 /// overloads) to guarantee that particularly important cases (cstring plus
80 /// StringRef) codegen as desired.
81 class Twine {
82 /// NodeKind - Represent the type of an argument.
83 enum NodeKind : unsigned char {
84 /// An empty string; the result of concatenating anything with it is also
85 /// empty.
86 NullKind,
87
88 /// The empty string.
89 EmptyKind,
90
91 /// A pointer to a Twine instance.
92 TwineKind,
93
94 /// A pointer to a C string instance.
95 CStringKind,
96
97 /// A pointer to an std::string instance.
98 StdStringKind,
99
100 /// A pointer to a StringRef instance.
101 StringRefKind,
102
103 /// A pointer to a SmallString instance.
104 SmallStringKind,
105
106 /// A pointer to a formatv_object_base instance.
107 FormatvObjectKind,
108
109 /// A char value, to render as a character.
110 CharKind,
111
112 /// An unsigned int value, to render as an unsigned decimal integer.
113 DecUIKind,
114
115 /// An int value, to render as a signed decimal integer.
116 DecIKind,
117
118 /// A pointer to an unsigned long value, to render as an unsigned decimal
119 /// integer.
120 DecULKind,
121
122 /// A pointer to a long value, to render as a signed decimal integer.
123 DecLKind,
124
125 /// A pointer to an unsigned long long value, to render as an unsigned
126 /// decimal integer.
127 DecULLKind,
128
129 /// A pointer to a long long value, to render as a signed decimal integer.
130 DecLLKind,
131
132 /// A pointer to a uint64_t value, to render as an unsigned hexadecimal
133 /// integer.
134 UHexKind
135 };
136
137 union Child
138 {
139 const Twine *twine;
140 const char *cString;
141 const std::string *stdString;
142 const StringRef *stringRef;
143 const SmallVectorImpl<char> *smallString;
144 const formatv_object_base *formatvObject;
145 char character;
146 unsigned int decUI;
147 int decI;
148 const unsigned long *decUL;
149 const long *decL;
150 const unsigned long long *decULL;
151 const long long *decLL;
152 const uint64_t *uHex;
153 };
154
155 /// LHS - The prefix in the concatenation, which may be uninitialized for
156 /// Null or Empty kinds.
157 Child LHS;
158
159 /// RHS - The suffix in the concatenation, which may be uninitialized for
160 /// Null or Empty kinds.
161 Child RHS;
162
163 /// LHSKind - The NodeKind of the left hand side, \see getLHSKind().
164 NodeKind LHSKind = EmptyKind;
165
166 /// RHSKind - The NodeKind of the right hand side, \see getRHSKind().
167 NodeKind RHSKind = EmptyKind;
168
169 /// Construct a nullary twine; the kind must be NullKind or EmptyKind.
170 explicit Twine(NodeKind Kind) : LHSKind(Kind) {
171 assert(isNullary() && "Invalid kind!");
172 }
173
174 /// Construct a binary twine.
175 explicit Twine(const Twine &LHS, const Twine &RHS)
176 : LHSKind(TwineKind), RHSKind(TwineKind) {
177 this->LHS.twine = &LHS;
178 this->RHS.twine = &RHS;
179 assert(isValid() && "Invalid twine!");
180 }
181
182 /// Construct a twine from explicit values.
183 explicit Twine(Child LHS, NodeKind LHSKind, Child RHS, NodeKind RHSKind)
184 : LHS(LHS), RHS(RHS), LHSKind(LHSKind), RHSKind(RHSKind) {
185 assert(isValid() && "Invalid twine!");
186 }
187
188 /// Check for the null twine.
189 bool isNull() const {
190 return getLHSKind() == NullKind;
191 }
192
193 /// Check for the empty twine.
194 bool isEmpty() const {
195 return getLHSKind() == EmptyKind;
196 }
197
198 /// Check if this is a nullary twine (null or empty).
199 bool isNullary() const {
200 return isNull() || isEmpty();
201 }
202
203 /// Check if this is a unary twine.
204 bool isUnary() const {
205 return getRHSKind() == EmptyKind && !isNullary();
206 }
207
208 /// Check if this is a binary twine.
209 bool isBinary() const {
210 return getLHSKind() != NullKind && getRHSKind() != EmptyKind;
211 }
212
213 /// Check if this is a valid twine (satisfying the invariants on
214 /// order and number of arguments).
215 bool isValid() const {
216 // Nullary twines always have Empty on the RHS.
217 if (isNullary() && getRHSKind() != EmptyKind)
218 return false;
219
220 // Null should never appear on the RHS.
221 if (getRHSKind() == NullKind)
222 return false;
223
224 // The RHS cannot be non-empty if the LHS is empty.
225 if (getRHSKind() != EmptyKind && getLHSKind() == EmptyKind)
226 return false;
227
228 // A twine child should always be binary.
229 if (getLHSKind() == TwineKind &&
230 !LHS.twine->isBinary())
231 return false;
232 if (getRHSKind() == TwineKind &&
233 !RHS.twine->isBinary())
234 return false;
235
236 return true;
237 }
238
239 /// Get the NodeKind of the left-hand side.
240 NodeKind getLHSKind() const { return LHSKind; }
241
242 /// Get the NodeKind of the right-hand side.
243 NodeKind getRHSKind() const { return RHSKind; }
244
245 /// Print one child from a twine.
246 void printOneChild(raw_ostream &OS, Child Ptr, NodeKind Kind) const;
247
248 /// Print the representation of one child from a twine.
249 void printOneChildRepr(raw_ostream &OS, Child Ptr,
250 NodeKind Kind) const;
251
252 public:
253 /// @name Constructors
254 /// @{
255
256 /// Construct from an empty string.
257 /*implicit*/ Twine() {
258 assert(isValid() && "Invalid twine!");
259 }
260
261 Twine(const Twine &) = default;
262
263 /// Construct from a C string.
264 ///
265 /// We take care here to optimize "" into the empty twine -- this will be
266 /// optimized out for string constants. This allows Twine arguments have
267 /// default "" values, without introducing unnecessary string constants.
268 /*implicit*/ Twine(const char *Str) {
269 if (Str[0] != '\0') {
270 LHS.cString = Str;
271 LHSKind = CStringKind;
272 } else
273 LHSKind = EmptyKind;
274
275 assert(isValid() && "Invalid twine!");
276 }
277
278 /// Construct from an std::string.
279 /*implicit*/ Twine(const std::string &Str) : LHSKind(StdStringKind) {
280 LHS.stdString = &Str;
281 assert(isValid() && "Invalid twine!");
282 }
283
284 /// Construct from a StringRef.
285 /*implicit*/ Twine(const StringRef &Str) : LHSKind(StringRefKind) {
286 LHS.stringRef = &Str;
287 assert(isValid() && "Invalid twine!");
288 }
289
290 /// Construct from a SmallString.
291 /*implicit*/ Twine(const SmallVectorImpl<char> &Str)
292 : LHSKind(SmallStringKind) {
293 LHS.smallString = &Str;
294 assert(isValid() && "Invalid twine!");
295 }
296
297 /// Construct from a formatv_object_base.
298 /*implicit*/ Twine(const formatv_object_base &Fmt)
299 : LHSKind(FormatvObjectKind) {
300 LHS.formatvObject = &Fmt;
301 assert(isValid() && "Invalid twine!");
302 }
303
304 /// Construct from a char.
305 explicit Twine(char Val) : LHSKind(CharKind) {
306 LHS.character = Val;
307 }
308
309 /// Construct from a signed char.
310 explicit Twine(signed char Val) : LHSKind(CharKind) {
311 LHS.character = static_cast<char>(Val);
312 }
313
314 /// Construct from an unsigned char.
315 explicit Twine(unsigned char Val) : LHSKind(CharKind) {
316 LHS.character = static_cast<char>(Val);
317 }
318
319 /// Construct a twine to print \p Val as an unsigned decimal integer.
320 explicit Twine(unsigned Val) : LHSKind(DecUIKind) {
321 LHS.decUI = Val;
322 }
323
324 /// Construct a twine to print \p Val as a signed decimal integer.
325 explicit Twine(int Val) : LHSKind(DecIKind) {
326 LHS.decI = Val;
327 }
328
329 /// Construct a twine to print \p Val as an unsigned decimal integer.
330 explicit Twine(const unsigned long &Val) : LHSKind(DecULKind) {
331 LHS.decUL = &Val;
332 }
333
334 /// Construct a twine to print \p Val as a signed decimal integer.
335 explicit Twine(const long &Val) : LHSKind(DecLKind) {
336 LHS.decL = &Val;
337 }
338
339 /// Construct a twine to print \p Val as an unsigned decimal integer.
340 explicit Twine(const unsigned long long &Val) : LHSKind(DecULLKind) {
341 LHS.decULL = &Val;
342 }
343
344 /// Construct a twine to print \p Val as a signed decimal integer.
345 explicit Twine(const long long &Val) : LHSKind(DecLLKind) {
346 LHS.decLL = &Val;
347 }
348
349 // FIXME: Unfortunately, to make sure this is as efficient as possible we
350 // need extra binary constructors from particular types. We can't rely on
351 // the compiler to be smart enough to fold operator+()/concat() down to the
352 // right thing. Yet.
353
354 /// Construct as the concatenation of a C string and a StringRef.
355 /*implicit*/ Twine(const char *LHS, const StringRef &RHS)
356 : LHSKind(CStringKind), RHSKind(StringRefKind) {
357 this->LHS.cString = LHS;
358 this->RHS.stringRef = &RHS;
359 assert(isValid() && "Invalid twine!");
360 }
361
362 /// Construct as the concatenation of a StringRef and a C string.
363 /*implicit*/ Twine(const StringRef &LHS, const char *RHS)
364 : LHSKind(StringRefKind), RHSKind(CStringKind) {
365 this->LHS.stringRef = &LHS;
366 this->RHS.cString = RHS;
367 assert(isValid() && "Invalid twine!");
368 }
369
370 /// Since the intended use of twines is as temporary objects, assignments
371 /// when concatenating might cause undefined behavior or stack corruptions
372 Twine &operator=(const Twine &) = delete;
373
374 /// Create a 'null' string, which is an empty string that always
375 /// concatenates to form another empty string.
376 static Twine createNull() {
377 return Twine(NullKind);
378 }
379
380 /// @}
381 /// @name Numeric Conversions
382 /// @{
383
384 // Construct a twine to print \p Val as an unsigned hexadecimal integer.
385 static Twine utohexstr(const uint64_t &Val) {
386 Child LHS, RHS;
387 LHS.uHex = &Val;
388 RHS.twine = nullptr;
389 return Twine(LHS, UHexKind, RHS, EmptyKind);
390 }
391
392 /// @}
393 /// @name Predicate Operations
394 /// @{
395
396 /// Check if this twine is trivially empty; a false return value does not
397 /// necessarily mean the twine is empty.
398 bool isTriviallyEmpty() const {
399 return isNullary();
400 }
401
402 /// Return true if this twine can be dynamically accessed as a single
403 /// StringRef value with getSingleStringRef().
404 bool isSingleStringRef() const {
405 if (getRHSKind() != EmptyKind) return false;
406
407 switch (getLHSKind()) {
408 case EmptyKind:
409 case CStringKind:
410 case StdStringKind:
411 case StringRefKind:
412 case SmallStringKind:
413 return true;
414 default:
415 return false;
416 }
417 }
418
419 /// @}
420 /// @name String Operations
421 /// @{
422
423 Twine concat(const Twine &Suffix) const;
424
425 /// @}
426 /// @name Output & Conversion.
427 /// @{
428
429 /// Return the twine contents as a std::string.
430 std::string str() const;
431
432 /// Append the concatenated string into the given SmallString or SmallVector.
433 void toVector(SmallVectorImpl<char> &Out) const;
434
435 /// This returns the twine as a single StringRef. This method is only valid
436 /// if isSingleStringRef() is true.
437 StringRef getSingleStringRef() const {
438 assert(isSingleStringRef() &&"This cannot be had as a single stringref!");
439 switch (getLHSKind()) {
440 default: llvm_unreachable("Out of sync with isSingleStringRef");
441 case EmptyKind: return StringRef();
442 case CStringKind: return StringRef(LHS.cString);
443 case StdStringKind: return StringRef(*LHS.stdString);
444 case StringRefKind: return *LHS.stringRef;
445 case SmallStringKind:
446 return StringRef(LHS.smallString->data(), LHS.smallString->size());
447 }
448 }
449
450 /// This returns the twine as a single StringRef if it can be
451 /// represented as such. Otherwise the twine is written into the given
452 /// SmallVector and a StringRef to the SmallVector's data is returned.
453 StringRef toStringRef(SmallVectorImpl<char> &Out) const {
454 if (isSingleStringRef())
455 return getSingleStringRef();
456 toVector(Out);
457 return StringRef(Out.data(), Out.size());
458 }
459
460 /// This returns the twine as a single null terminated StringRef if it
461 /// can be represented as such. Otherwise the twine is written into the
462 /// given SmallVector and a StringRef to the SmallVector's data is returned.
463 ///
464 /// The returned StringRef's size does not include the null terminator.
465 StringRef toNullTerminatedStringRef(SmallVectorImpl<char> &Out) const;
466
467 /// Write the concatenated string represented by this twine to the
468 /// stream \p OS.
469 void print(raw_ostream &OS) const;
470
471 /// Dump the concatenated string represented by this twine to stderr.
472 void dump() const;
473
474 /// Write the representation of this twine to the stream \p OS.
475 void printRepr(raw_ostream &OS) const;
476
477 /// Dump the representation of this twine to stderr.
478 void dumpRepr() const;
479
480 /// @}
481 };
482
483 /// @name Twine Inline Implementations
484 /// @{
485
486 inline Twine Twine::concat(const Twine &Suffix) const {
487 // Concatenation with null is null.
488 if (isNull() || Suffix.isNull())
489 return Twine(NullKind);
490
491 // Concatenation with empty yields the other side.
492 if (isEmpty())
493 return Suffix;
494 if (Suffix.isEmpty())
495 return *this;
496
497 // Otherwise we need to create a new node, taking care to fold in unary
498 // twines.
499 Child NewLHS, NewRHS;
500 NewLHS.twine = this;
501 NewRHS.twine = &Suffix;
502 NodeKind NewLHSKind = TwineKind, NewRHSKind = TwineKind;
503 if (isUnary()) {
504 NewLHS = LHS;
505 NewLHSKind = getLHSKind();
506 }
507 if (Suffix.isUnary()) {
508 NewRHS = Suffix.LHS;
509 NewRHSKind = Suffix.getLHSKind();
510 }
511
512 return Twine(NewLHS, NewLHSKind, NewRHS, NewRHSKind);
513 }
514
515 inline Twine operator+(const Twine &LHS, const Twine &RHS) {
516 return LHS.concat(RHS);
517 }
518
519 /// Additional overload to guarantee simplified codegen; this is equivalent to
520 /// concat().
521
522 inline Twine operator+(const char *LHS, const StringRef &RHS) {
523 return Twine(LHS, RHS);
524 }
525
526 /// Additional overload to guarantee simplified codegen; this is equivalent to
527 /// concat().
528
529 inline Twine operator+(const StringRef &LHS, const char *RHS) {
530 return Twine(LHS, RHS);
531 }
532
533 inline raw_ostream &operator<<(raw_ostream &OS, const Twine &RHS) {
534 RHS.print(OS);
535 return OS;
536 }
537
538 /// @}
539
540} // end namespace llvm
541
542#endif // LLVM_ADT_TWINE_H
543