1 //===-- Twine.h - Fast Temporary String Concatenation -----------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 #ifndef LLVM_ADT_TWINE_H
11 #define LLVM_ADT_TWINE_H
13 #include "llvm/ADT/StringRef.h"
14 #include "llvm/System/DataTypes.h"
20 class SmallVectorImpl;
24 /// Twine - A lightweight data structure for efficiently representing the
25 /// concatenation of temporary values as strings.
27 /// A Twine is a kind of rope, it represents a concatenated string using a
28 /// binary-tree, where the string is the preorder of the nodes. Since the
29 /// Twine can be efficiently rendered into a buffer when its result is used,
30 /// it avoids the cost of generating temporary values for intermediate string
31 /// results -- particularly in cases when the Twine result is never
32 /// required. By explicitly tracking the type of leaf nodes, we can also avoid
33 /// the creation of temporary strings for conversions operations (such as
34 /// appending an integer to a string).
36 /// A Twine is not intended for use directly and should not be stored, its
37 /// implementation relies on the ability to store pointers to temporary stack
38 /// objects which may be deallocated at the end of a statement. Twines should
39 /// only be used accepted as const references in arguments, when an API wishes
40 /// to accept possibly-concatenated strings.
42 /// Twines support a special 'null' value, which always concatenates to form
43 /// itself, and renders as an empty string. This can be returned from APIs to
44 /// effectively nullify any concatenations performed on the result.
46 /// \b Implementation \n
48 /// Given the nature of a Twine, it is not possible for the Twine's
49 /// concatenation method to construct interior nodes; the result must be
50 /// represented inside the returned value. For this reason a Twine object
51 /// actually holds two values, the left- and right-hand sides of a
52 /// concatenation. We also have nullary Twine objects, which are effectively
53 /// sentinel values that represent empty strings.
55 /// Thus, a Twine can effectively have zero, one, or two children. The \see
56 /// isNullary(), \see isUnary(), and \see isBinary() predicates exist for
57 /// testing the number of children.
59 /// We maintain a number of invariants on Twine objects (FIXME: Why):
60 /// - Nullary twines are always represented with their Kind on the left-hand
61 /// side, and the Empty kind on the right-hand side.
62 /// - Unary twines are always represented with the value on the left-hand
63 /// side, and the Empty kind on the right-hand side.
64 /// - If a Twine has another Twine as a child, that child should always be
65 /// binary (otherwise it could have been folded into the parent).
67 /// These invariants are check by \see isValid().
69 /// \b Efficiency Considerations \n
71 /// The Twine is designed to yield efficient and small code for common
72 /// situations. For this reason, the concat() method is inlined so that
73 /// concatenations of leaf nodes can be optimized into stores directly into a
74 /// single stack allocated object.
76 /// In practice, not all compilers can be trusted to optimize concat() fully,
77 /// so we provide two additional methods (and accompanying operator+
78 /// overloads) to guarantee that particularly important cases (cstring plus
79 /// StringRef) codegen as desired.
81 /// NodeKind - Represent the type of an argument.
83 /// An empty string; the result of concatenating anything with it is also
90 /// A pointer to a Twine instance.
93 /// A pointer to a C string instance.
96 /// A pointer to an std::string instance.
99 /// A pointer to a StringRef instance.
102 /// A pointer to an unsigned int value, to render as an unsigned decimal
106 /// A pointer to an int value, to render as a signed decimal integer.
109 /// A pointer to an unsigned long value, to render as an unsigned decimal
113 /// A pointer to a long value, to render as a signed decimal integer.
116 /// A pointer to an unsigned long long value, to render as an unsigned
120 /// A pointer to a long long value, to render as a signed decimal integer.
123 /// A pointer to a uint64_t value, to render as an unsigned hexadecimal
129 /// LHS - The prefix in the concatenation, which may be uninitialized for
130 /// Null or Empty kinds.
132 /// RHS - The suffix in the concatenation, which may be uninitialized for
133 /// Null or Empty kinds.
135 /// LHSKind - The NodeKind of the left hand side, \see getLHSKind().
136 unsigned char LHSKind;
137 /// RHSKind - The NodeKind of the left hand side, \see getLHSKind().
138 unsigned char RHSKind;
141 /// Construct a nullary twine; the kind must be NullKind or EmptyKind.
142 explicit Twine(NodeKind Kind)
143 : LHSKind(Kind), RHSKind(EmptyKind) {
144 assert(isNullary() && "Invalid kind!");
147 /// Construct a binary twine.
148 explicit Twine(const Twine &_LHS, const Twine &_RHS)
149 : LHS(&_LHS), RHS(&_RHS), LHSKind(TwineKind), RHSKind(TwineKind) {
150 assert(isValid() && "Invalid twine!");
153 /// Construct a twine from explicit values.
154 explicit Twine(const void *_LHS, NodeKind _LHSKind,
155 const void *_RHS, NodeKind _RHSKind)
156 : LHS(_LHS), RHS(_RHS), LHSKind(_LHSKind), RHSKind(_RHSKind) {
157 assert(isValid() && "Invalid twine!");
160 /// isNull - Check for the null twine.
161 bool isNull() const {
162 return getLHSKind() == NullKind;
165 /// isEmpty - Check for the empty twine.
166 bool isEmpty() const {
167 return getLHSKind() == EmptyKind;
170 /// isNullary - Check if this is a nullary twine (null or empty).
171 bool isNullary() const {
172 return isNull() || isEmpty();
175 /// isUnary - Check if this is a unary twine.
176 bool isUnary() const {
177 return getRHSKind() == EmptyKind && !isNullary();
180 /// isBinary - Check if this is a binary twine.
181 bool isBinary() const {
182 return getLHSKind() != NullKind && getRHSKind() != EmptyKind;
185 /// isValid - Check if this is a valid twine (satisfying the invariants on
186 /// order and number of arguments).
187 bool isValid() const {
188 // Nullary twines always have Empty on the RHS.
189 if (isNullary() && getRHSKind() != EmptyKind)
192 // Null should never appear on the RHS.
193 if (getRHSKind() == NullKind)
196 // The RHS cannot be non-empty if the LHS is empty.
197 if (getRHSKind() != EmptyKind && getLHSKind() == EmptyKind)
200 // A twine child should always be binary.
201 if (getLHSKind() == TwineKind &&
202 !static_cast<const Twine*>(LHS)->isBinary())
204 if (getRHSKind() == TwineKind &&
205 !static_cast<const Twine*>(RHS)->isBinary())
211 /// getLHSKind - Get the NodeKind of the left-hand side.
212 NodeKind getLHSKind() const { return (NodeKind) LHSKind; }
214 /// getRHSKind - Get the NodeKind of the left-hand side.
215 NodeKind getRHSKind() const { return (NodeKind) RHSKind; }
217 /// printOneChild - Print one child from a twine.
218 void printOneChild(raw_ostream &OS, const void *Ptr, NodeKind Kind) const;
220 /// printOneChildRepr - Print the representation of one child from a twine.
221 void printOneChildRepr(raw_ostream &OS, const void *Ptr,
222 NodeKind Kind) const;
225 /// @name Constructors
228 /// Construct from an empty string.
229 /*implicit*/ Twine() : LHSKind(EmptyKind), RHSKind(EmptyKind) {
230 assert(isValid() && "Invalid twine!");
233 /// Construct from a C string.
235 /// We take care here to optimize "" into the empty twine -- this will be
236 /// optimized out for string constants. This allows Twine arguments have
237 /// default "" values, without introducing unnecessary string constants.
238 /*implicit*/ Twine(const char *Str)
239 : RHSKind(EmptyKind) {
240 if (Str[0] != '\0') {
242 LHSKind = CStringKind;
246 assert(isValid() && "Invalid twine!");
249 /// Construct from an std::string.
250 /*implicit*/ Twine(const std::string &Str)
251 : LHS(&Str), LHSKind(StdStringKind), RHSKind(EmptyKind) {
252 assert(isValid() && "Invalid twine!");
255 /// Construct from a StringRef.
256 /*implicit*/ Twine(const StringRef &Str)
257 : LHS(&Str), LHSKind(StringRefKind), RHSKind(EmptyKind) {
258 assert(isValid() && "Invalid twine!");
261 /// Construct a twine to print \arg Val as an unsigned decimal integer.
262 explicit Twine(const unsigned int &Val)
263 : LHS(&Val), LHSKind(DecUIKind), RHSKind(EmptyKind) {
266 /// Construct a twine to print \arg Val as a signed decimal integer.
267 explicit Twine(const int &Val)
268 : LHS(&Val), LHSKind(DecIKind), RHSKind(EmptyKind) {
271 /// Construct a twine to print \arg Val as an unsigned decimal integer.
272 explicit Twine(const unsigned long &Val)
273 : LHS(&Val), LHSKind(DecULKind), RHSKind(EmptyKind) {
276 /// Construct a twine to print \arg Val as a signed decimal integer.
277 explicit Twine(const long &Val)
278 : LHS(&Val), LHSKind(DecLKind), RHSKind(EmptyKind) {
281 /// Construct a twine to print \arg Val as an unsigned decimal integer.
282 explicit Twine(const unsigned long long &Val)
283 : LHS(&Val), LHSKind(DecULLKind), RHSKind(EmptyKind) {
286 /// Construct a twine to print \arg Val as a signed decimal integer.
287 explicit Twine(const long long &Val)
288 : LHS(&Val), LHSKind(DecLLKind), RHSKind(EmptyKind) {
291 // FIXME: Unfortunately, to make sure this is as efficient as possible we
292 // need extra binary constructors from particular types. We can't rely on
293 // the compiler to be smart enough to fold operator+()/concat() down to the
296 /// Construct as the concatenation of a C string and a StringRef.
297 /*implicit*/ Twine(const char *_LHS, const StringRef &_RHS)
298 : LHS(_LHS), RHS(&_RHS), LHSKind(CStringKind), RHSKind(StringRefKind) {
299 assert(isValid() && "Invalid twine!");
302 /// Construct as the concatenation of a StringRef and a C string.
303 /*implicit*/ Twine(const StringRef &_LHS, const char *_RHS)
304 : LHS(&_LHS), RHS(_RHS), LHSKind(StringRefKind), RHSKind(CStringKind) {
305 assert(isValid() && "Invalid twine!");
308 /// Create a 'null' string, which is an empty string that always
309 /// concatenates to form another empty string.
310 static Twine createNull() {
311 return Twine(NullKind);
315 /// @name Numeric Conversions
318 // Construct a twine to print \arg Val as an unsigned hexadecimal integer.
319 static Twine utohexstr(const uint64_t &Val) {
320 return Twine(&Val, UHexKind, 0, EmptyKind);
324 /// @name Predicate Operations
327 /// isTriviallyEmpty - Check if this twine is trivially empty; a false
328 /// return value does not necessarily mean the twine is empty.
329 bool isTriviallyEmpty() const {
333 /// isSingleStringRef - Return true if this twine can be dynamically
334 /// accessed as a single StringRef value with getSingleStringRef().
335 bool isSingleStringRef() const {
336 if (getRHSKind() != EmptyKind) return false;
338 switch (getLHSKind()) {
350 /// @name String Operations
353 Twine concat(const Twine &Suffix) const;
356 /// @name Output & Conversion.
359 /// str - Return the twine contents as a std::string.
360 std::string str() const;
362 /// toVector - Write the concatenated string into the given SmallString or
364 void toVector(SmallVectorImpl<char> &Out) const;
366 /// getSingleStringRef - This returns the twine as a single StringRef. This
367 /// method is only valid if isSingleStringRef() is true.
368 StringRef getSingleStringRef() const {
369 assert(isSingleStringRef() &&"This cannot be had as a single stringref!");
370 switch (getLHSKind()) {
371 default: assert(0 && "Out of sync with isSingleStringRef");
372 case EmptyKind: return StringRef();
373 case CStringKind: return StringRef((const char*)LHS);
374 case StdStringKind: return StringRef(*(const std::string*)LHS);
375 case StringRefKind: return *(const StringRef*)LHS;
379 /// toStringRef - This returns the twine as a single StringRef if it can be
380 /// represented as such. Otherwise the twine is written into the given
381 /// SmallVector and a StringRef to the SmallVector's data is returned.
382 StringRef toStringRef(SmallVectorImpl<char> &Out) const;
384 /// print - Write the concatenated string represented by this twine to the
386 void print(raw_ostream &OS) const;
388 /// dump - Dump the concatenated string represented by this twine to stderr.
391 /// print - Write the representation of this twine to the stream \arg OS.
392 void printRepr(raw_ostream &OS) const;
394 /// dumpRepr - Dump the representation of this twine to stderr.
395 void dumpRepr() const;
400 /// @name Twine Inline Implementations
403 inline Twine Twine::concat(const Twine &Suffix) const {
404 // Concatenation with null is null.
405 if (isNull() || Suffix.isNull())
406 return Twine(NullKind);
408 // Concatenation with empty yields the other side.
411 if (Suffix.isEmpty())
414 // Otherwise we need to create a new node, taking care to fold in unary
416 const void *NewLHS = this, *NewRHS = &Suffix;
417 NodeKind NewLHSKind = TwineKind, NewRHSKind = TwineKind;
420 NewLHSKind = getLHSKind();
422 if (Suffix.isUnary()) {
424 NewRHSKind = Suffix.getLHSKind();
427 return Twine(NewLHS, NewLHSKind, NewRHS, NewRHSKind);
430 inline Twine operator+(const Twine &LHS, const Twine &RHS) {
431 return LHS.concat(RHS);
434 /// Additional overload to guarantee simplified codegen; this is equivalent to
437 inline Twine operator+(const char *LHS, const StringRef &RHS) {
438 return Twine(LHS, RHS);
441 /// Additional overload to guarantee simplified codegen; this is equivalent to
444 inline Twine operator+(const StringRef &LHS, const char *RHS) {
445 return Twine(LHS, RHS);
448 inline raw_ostream &operator<<(raw_ostream &OS, const Twine &RHS) {