1 // stringpool.h -- a string pool for gold -*- C++ -*-
3 // Copyright (C) 2006-2019 Free Software Foundation, Inc.
4 // Written by Ian Lance Taylor <iant@google.com>.
6 // This file is part of gold.
8 // This program is free software; you can redistribute it and/or modify
9 // it under the terms of the GNU General Public License as published by
10 // the Free Software Foundation; either version 3 of the License, or
11 // (at your option) any later version.
13 // This program is distributed in the hope that it will be useful,
14 // but WITHOUT ANY WARRANTY; without even the implied warranty of
15 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 // GNU General Public License for more details.
18 // You should have received a copy of the GNU General Public License
19 // along with this program; if not, write to the Free Software
20 // Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston,
21 // MA 02110-1301, USA.
27 #ifndef GOLD_STRINGPOOL_H
28 #define GOLD_STRINGPOOL_H
35 // Return the length of a string in units of Char_type.
37 template<typename Char_type
>
39 string_length(const Char_type
* p
)
47 // Specialize string_length for char. Maybe we could just use
48 // std::char_traits<>::length?
52 string_length(const char* p
)
57 // A Stringpool is a pool of unique strings. It provides the
58 // following features:
60 // Every string in the pool is unique. Thus, if you have two strings
61 // in the Stringpool, you can compare them for equality by using
62 // pointer comparison rather than string comparison.
64 // There is a key associated with every string in the pool. If you
65 // add strings to the Stringpool in the same order, then the key for
66 // each string will always be the same for any run of the linker.
67 // This is not true of the string pointers themselves, as they may
68 // change due to address space randomization. Some parts of the
69 // linker (e.g., the symbol table) use the key value instead of the
70 // string pointer so that repeated runs of the linker will generate
71 // precisely the same output.
73 // When you add a string to a Stringpool, Stringpool will optionally
74 // make a copy of it. Thus there is no requirement to keep a copy
77 // A Stringpool can be turned into a string table, a sequential series
78 // of null terminated strings. The first string may optionally be a
79 // single zero byte, as required for SHT_STRTAB sections. This
80 // conversion is only permitted after all strings have been added to
81 // the Stringpool. After doing this conversion, you can ask for the
82 // offset of any string (or any key) in the stringpool in the string
83 // table, and you can write the resulting string table to an output
86 // When a Stringpool is turned into a string table, then as an
87 // optimization it will reuse string suffixes to avoid duplicating
88 // strings. That is, given the strings "abc" and "bc", only the
89 // string "abc" will be stored, and "bc" will be represented by an
90 // offset into the middle of the string "abc".
93 // A simple chunked vector class--this is a subset of std::vector
94 // which stores memory in chunks. We don't provide iterators, because
95 // we don't need them.
97 template<typename Element
>
102 : chunks_(), size_(0)
105 // Clear the elements.
109 this->chunks_
.clear();
115 reserve(unsigned int n
)
117 if (n
> this->chunks_
.size() * chunk_size
)
119 this->chunks_
.resize((n
+ chunk_size
- 1) / chunk_size
);
120 // We need to call reserve() of all chunks since changing
121 // this->chunks_ causes Element_vectors to be copied. The
122 // reserved capacity of an Element_vector may be lost in copying.
123 for (size_t i
= 0; i
< this->chunks_
.size(); ++i
)
124 this->chunks_
[i
].reserve(chunk_size
);
128 // Get the number of elements.
131 { return this->size_
; }
133 // Push a new element on the back of the vector.
135 push_back(const Element
& element
)
137 size_t chunk_index
= this->size_
/ chunk_size
;
138 if (chunk_index
>= this->chunks_
.size())
140 this->chunks_
.push_back(Element_vector());
141 this->chunks_
.back().reserve(chunk_size
);
142 gold_assert(chunk_index
< this->chunks_
.size());
144 this->chunks_
[chunk_index
].push_back(element
);
148 // Return a reference to an entry in the vector.
151 { return this->chunks_
[i
/ chunk_size
][i
% chunk_size
]; }
154 operator[](size_t i
) const
155 { return this->chunks_
[i
/ chunk_size
][i
% chunk_size
]; }
158 static const unsigned int chunk_size
= 8192;
160 typedef std::vector
<Element
> Element_vector
;
161 typedef std::vector
<Element_vector
> Chunk_vector
;
163 Chunk_vector chunks_
;
168 // Stringpools are implemented in terms of Stringpool_template, which
169 // is generalized on the type of character used for the strings. Most
170 // uses will want the Stringpool type which uses char. Other cases
171 // are used for merging wide string constants.
173 template<typename Stringpool_char
>
174 class Stringpool_template
177 // The type of a key into the stringpool. As described above, a key
178 // value will always be the same during any run of the linker. Zero
179 // is never a valid key value.
182 // Create a Stringpool.
183 Stringpool_template(uint64_t addralign
= 1);
185 ~Stringpool_template();
187 // Clear all the data from the stringpool.
191 // Hint to the stringpool class that you intend to insert n additional
192 // elements. The stringpool class can use this info however it likes;
193 // in practice it will resize its internal hashtables to make room.
195 reserve(unsigned int n
);
197 // Indicate that we should not reserve offset 0 to hold the empty
198 // string when converting the stringpool to a string table. This
199 // should not be called for a proper ELF SHT_STRTAB section.
203 gold_assert(this->string_set_
.empty()
204 && this->offset_
== sizeof(Stringpool_char
));
205 this->zero_null_
= false;
209 // Indicate that this string pool should be optimized, even if not
213 { this->optimize_
= true; }
215 // Add the string S to the pool. This returns a canonical permanent
216 // pointer to the string in the pool. If COPY is true, the string
217 // is copied into permanent storage. If PKEY is not NULL, this sets
218 // *PKEY to the key for the string.
219 const Stringpool_char
*
220 add(const Stringpool_char
* s
, bool copy
, Key
* pkey
);
222 // Add the string S to the pool.
223 const Stringpool_char
*
224 add(const std::basic_string
<Stringpool_char
>& s
, bool copy
, Key
* pkey
)
225 { return this->add_with_length(s
.data(), s
.size(), copy
, pkey
); }
227 // Add string S of length LEN characters to the pool. If COPY is
228 // true, S need not be null terminated.
229 const Stringpool_char
*
230 add_with_length(const Stringpool_char
* s
, size_t len
, bool copy
, Key
* pkey
);
232 // If the string S is present in the pool, return the canonical
233 // string pointer. Otherwise, return NULL. If PKEY is not NULL,
234 // set *PKEY to the key.
235 const Stringpool_char
*
236 find(const Stringpool_char
* s
, Key
* pkey
) const;
238 // Turn the stringpool into a string table: determine the offsets of
239 // all the strings. After this is called, no more strings may be
240 // added to the stringpool.
242 set_string_offsets();
244 // Get the offset of the string S in the string table. This returns
245 // the offset in bytes, not in units of Stringpool_char. This may
246 // only be called after set_string_offsets has been called.
248 get_offset(const Stringpool_char
* s
) const;
250 // Get the offset of the string S in the string table.
252 get_offset(const std::basic_string
<Stringpool_char
>& s
) const
253 { return this->get_offset_with_length(s
.c_str(), s
.size()); }
255 // Get the offset of string S, with length LENGTH characters, in the
258 get_offset_with_length(const Stringpool_char
* s
, size_t length
) const;
260 // Get the offset of the string with key K.
262 get_offset_from_key(Key k
) const
264 gold_assert(k
<= this->key_to_offset_
.size());
265 return this->key_to_offset_
[k
- 1];
268 // Get the size of the string table. This returns the number of
269 // bytes, not in units of Stringpool_char.
271 get_strtab_size() const
273 gold_assert(this->strtab_size_
!= 0);
274 return this->strtab_size_
;
277 // Write the string table into the output file at the specified
280 write(Output_file
*, off_t offset
);
282 // Write the string table into the specified buffer, of the
283 // specified size. buffer_size should be at least
284 // get_strtab_size().
286 write_to_buffer(unsigned char* buffer
, section_size_type buffer_size
);
288 // Dump statistical information to stderr.
290 print_stats(const char*) const;
293 Stringpool_template(const Stringpool_template
&);
294 Stringpool_template
& operator=(const Stringpool_template
&);
296 // Return whether two strings are equal.
298 string_equal(const Stringpool_char
*, const Stringpool_char
*);
300 // Compute a hash code for a string. LENGTH is the length of the
301 // string in characters.
303 string_hash(const Stringpool_char
*, size_t length
);
305 // We store the actual data in a list of these buffers.
308 // Length of data in buffer.
310 // Allocated size of buffer.
316 // Add a new key offset entry.
318 new_key_offset(size_t);
320 // Copy a string into the buffers, returning a canonical string.
321 const Stringpool_char
*
322 add_string(const Stringpool_char
*, size_t);
324 // Return whether s1 is a suffix of s2.
326 is_suffix(const Stringpool_char
* s1
, size_t len1
,
327 const Stringpool_char
* s2
, size_t len2
);
329 // The hash table key includes the string, the length of the string,
330 // and the hash code for the string. We put the hash code
331 // explicitly into the key so that we can do a find()/insert()
332 // sequence without having to recompute the hash. Computing the
333 // hash code is a significant user of CPU time in the linker.
336 const Stringpool_char
* string
;
337 // Length is in characters, not bytes.
341 // This goes in an STL container, so we need a default
344 : string(NULL
), length(0), hash_code(0)
347 // Note that these constructors are relatively expensive, because
348 // they compute the hash code.
349 explicit Hashkey(const Stringpool_char
* s
)
350 : string(s
), length(string_length(s
)), hash_code(string_hash(s
, length
))
353 Hashkey(const Stringpool_char
* s
, size_t len
)
354 : string(s
), length(len
), hash_code(string_hash(s
, len
))
358 // Hash function. This is trivial, since we have already computed
360 struct Stringpool_hash
363 operator()(const Hashkey
& hk
) const
364 { return hk
.hash_code
; }
367 // Equality comparison function for hash table.
371 operator()(const Hashkey
&, const Hashkey
&) const;
374 // The hash table is a map from strings to Keys.
378 typedef Unordered_map
<Hashkey
, Hashval
, Stringpool_hash
,
379 Stringpool_eq
> String_set_type
;
381 // Comparison routine used when sorting into a string table.
383 typedef typename
String_set_type::iterator Stringpool_sort_info
;
385 struct Stringpool_sort_comparison
388 operator()(const Stringpool_sort_info
&, const Stringpool_sort_info
&) const;
391 // Keys map to offsets via a Chunked_vector. We only use the
392 // offsets if we turn this into an string table section.
393 typedef Chunked_vector
<section_offset_type
> Key_to_offset
;
395 // List of Stringdata structures.
396 typedef std::list
<Stringdata
*> Stringdata_list
;
398 // Mapping from const char* to namepool entry.
399 String_set_type string_set_
;
400 // Mapping from Key to string table offset.
401 Key_to_offset key_to_offset_
;
403 Stringdata_list strings_
;
404 // Size of string table.
405 section_size_type strtab_size_
;
406 // Whether to reserve offset 0 to hold the null string.
408 // Whether to optimize the string table.
410 // offset of the next string.
411 section_offset_type offset_
;
412 // The alignment of strings in the stringpool.
416 // The most common type of Stringpool.
417 typedef Stringpool_template
<char> Stringpool
;
419 } // End namespace gold.
421 #endif // !defined(GOLD_STRINGPOOL_H)