// stringpool.h -- a string pool for gold -*- C++ -*-
-// Copyright 2006, 2007 Free Software Foundation, Inc.
+// Copyright (C) 2006-2019 Free Software Foundation, Inc.
// Written by Ian Lance Taylor <iant@google.com>.
// This file is part of gold.
#include <string>
#include <list>
+#include <vector>
#ifndef GOLD_STRINGPOOL_H
#define GOLD_STRINGPOOL_H
class Output_file;
+// Return the length of a string in units of Char_type.
+
+template<typename Char_type>
+inline size_t
+string_length(const Char_type* p)
+{
+ size_t len = 0;
+ for (; *p != 0; ++p)
+ ++len;
+ return len;
+}
+
+// Specialize string_length for char. Maybe we could just use
+// std::char_traits<>::length?
+
+template<>
+inline size_t
+string_length(const char* p)
+{
+ return strlen(p);
+}
+
// A Stringpool is a pool of unique strings. It provides the
// following features:
// single zero byte, as required for SHT_STRTAB sections. This
// conversion is only permitted after all strings have been added to
// the Stringpool. After doing this conversion, you can ask for the
-// offset of any string in the stringpool in the string table, and you
-// can write the resulting string table to an output file.
+// offset of any string (or any key) in the stringpool in the string
+// table, and you can write the resulting string table to an output
+// file.
// When a Stringpool is turned into a string table, then as an
// optimization it will reuse string suffixes to avoid duplicating
// string "abc" will be stored, and "bc" will be represented by an
// offset into the middle of the string "abc".
+
+// A simple chunked vector class--this is a subset of std::vector
+// which stores memory in chunks. We don't provide iterators, because
+// we don't need them.
+
+template<typename Element>
+class Chunked_vector
+{
+ public:
+ Chunked_vector()
+ : chunks_(), size_(0)
+ { }
+
+ // Clear the elements.
+ void
+ clear()
+ {
+ this->chunks_.clear();
+ this->size_ = 0;
+ }
+
+ // Reserve elements.
+ void
+ reserve(unsigned int n)
+ {
+ if (n > this->chunks_.size() * chunk_size)
+ {
+ this->chunks_.resize((n + chunk_size - 1) / chunk_size);
+ // We need to call reserve() of all chunks since changing
+ // this->chunks_ causes Element_vectors to be copied. The
+ // reserved capacity of an Element_vector may be lost in copying.
+ for (size_t i = 0; i < this->chunks_.size(); ++i)
+ this->chunks_[i].reserve(chunk_size);
+ }
+ }
+
+ // Get the number of elements.
+ size_t
+ size() const
+ { return this->size_; }
+
+ // Push a new element on the back of the vector.
+ void
+ push_back(const Element& element)
+ {
+ size_t chunk_index = this->size_ / chunk_size;
+ if (chunk_index >= this->chunks_.size())
+ {
+ this->chunks_.push_back(Element_vector());
+ this->chunks_.back().reserve(chunk_size);
+ gold_assert(chunk_index < this->chunks_.size());
+ }
+ this->chunks_[chunk_index].push_back(element);
+ this->size_++;
+ }
+
+ // Return a reference to an entry in the vector.
+ Element&
+ operator[](size_t i)
+ { return this->chunks_[i / chunk_size][i % chunk_size]; }
+
+ const Element&
+ operator[](size_t i) const
+ { return this->chunks_[i / chunk_size][i % chunk_size]; }
+
+ private:
+ static const unsigned int chunk_size = 8192;
+
+ typedef std::vector<Element> Element_vector;
+ typedef std::vector<Element_vector> Chunk_vector;
+
+ Chunk_vector chunks_;
+ size_t size_;
+};
+
+
// Stringpools are implemented in terms of Stringpool_template, which
// is generalized on the type of character used for the strings. Most
// uses will want the Stringpool type which uses char. Other cases
typedef size_t Key;
// Create a Stringpool.
- Stringpool_template();
+ Stringpool_template(uint64_t addralign = 1);
~Stringpool_template();
void
clear();
+ // Hint to the stringpool class that you intend to insert n additional
+ // elements. The stringpool class can use this info however it likes;
+ // in practice it will resize its internal hashtables to make room.
+ void
+ reserve(unsigned int n);
+
// Indicate that we should not reserve offset 0 to hold the empty
// string when converting the stringpool to a string table. This
// should not be called for a proper ELF SHT_STRTAB section.
void
set_no_zero_null()
- { this->zero_null_ = false; }
+ {
+ gold_assert(this->string_set_.empty()
+ && this->offset_ == sizeof(Stringpool_char));
+ this->zero_null_ = false;
+ this->offset_ = 0;
+ }
+
+ // Indicate that this string pool should be optimized, even if not
+ // running with -O2.
+ void
+ set_optimize()
+ { this->optimize_ = true; }
// Add the string S to the pool. This returns a canonical permanent
// pointer to the string in the pool. If COPY is true, the string
const Stringpool_char*
add(const Stringpool_char* s, bool copy, Key* pkey);
- // Add the prefix of length LEN of string S to the pool.
+ // Add the string S to the pool.
const Stringpool_char*
- add_prefix(const Stringpool_char* s, size_t len, Key* pkey);
+ add(const std::basic_string<Stringpool_char>& s, bool copy, Key* pkey)
+ { return this->add_with_length(s.data(), s.size(), copy, pkey); }
+
+ // Add string S of length LEN characters to the pool. If COPY is
+ // true, S need not be null terminated.
+ const Stringpool_char*
+ add_with_length(const Stringpool_char* s, size_t len, bool copy, Key* pkey);
// If the string S is present in the pool, return the canonical
// string pointer. Otherwise, return NULL. If PKEY is not NULL,
// Get the offset of the string S in the string table. This returns
// the offset in bytes, not in units of Stringpool_char. This may
// only be called after set_string_offsets has been called.
- off_t
+ section_offset_type
get_offset(const Stringpool_char* s) const;
// Get the offset of the string S in the string table.
- off_t
+ section_offset_type
get_offset(const std::basic_string<Stringpool_char>& s) const
- { return this->get_offset(s.c_str()); }
+ { return this->get_offset_with_length(s.c_str(), s.size()); }
+
+ // Get the offset of string S, with length LENGTH characters, in the
+ // string table.
+ section_offset_type
+ get_offset_with_length(const Stringpool_char* s, size_t length) const;
+
+ // Get the offset of the string with key K.
+ section_offset_type
+ get_offset_from_key(Key k) const
+ {
+ gold_assert(k <= this->key_to_offset_.size());
+ return this->key_to_offset_[k - 1];
+ }
// Get the size of the string table. This returns the number of
// bytes, not in units of Stringpool_char.
- off_t
+ section_size_type
get_strtab_size() const
{
gold_assert(this->strtab_size_ != 0);
// specified size. buffer_size should be at least
// get_strtab_size().
void
- write_to_buffer(unsigned char* buffer, size_t buffer_size);
+ write_to_buffer(unsigned char* buffer, section_size_type buffer_size);
// Dump statistical information to stderr.
void
Stringpool_template(const Stringpool_template&);
Stringpool_template& operator=(const Stringpool_template&);
- // Return the length of a string in units of Stringpool_char.
- static size_t
- string_length(const Stringpool_char*);
-
// Return whether two strings are equal.
static bool
string_equal(const Stringpool_char*, const Stringpool_char*);
size_t len;
// Allocated size of buffer.
size_t alc;
- // Buffer index.
- unsigned int index;
// Buffer.
char data[1];
};
+ // Add a new key offset entry.
+ void
+ new_key_offset(size_t);
+
// Copy a string into the buffers, returning a canonical string.
const Stringpool_char*
- add_string(const Stringpool_char*, size_t, Key*);
+ add_string(const Stringpool_char*, size_t);
// Return whether s1 is a suffix of s2.
static bool
// Note that these constructors are relatively expensive, because
// they compute the hash code.
- Hashkey(const Stringpool_char* s)
+ explicit Hashkey(const Stringpool_char* s)
: string(s), length(string_length(s)), hash_code(string_hash(s, length))
{ }
operator()(const Hashkey&, const Hashkey&) const;
};
- // The hash table is a map from strings to a pair of Key and string
- // table offsets. We only use the offsets if we turn this into an
- // string table section.
+ // The hash table is a map from strings to Keys.
- typedef std::pair<Key, off_t> Hashval;
+ typedef Key Hashval;
typedef Unordered_map<Hashkey, Hashval, Stringpool_hash,
Stringpool_eq> String_set_type;
operator()(const Stringpool_sort_info&, const Stringpool_sort_info&) const;
};
+ // Keys map to offsets via a Chunked_vector. We only use the
+ // offsets if we turn this into an string table section.
+ typedef Chunked_vector<section_offset_type> Key_to_offset;
+
// List of Stringdata structures.
typedef std::list<Stringdata*> Stringdata_list;
// Mapping from const char* to namepool entry.
String_set_type string_set_;
+ // Mapping from Key to string table offset.
+ Key_to_offset key_to_offset_;
// List of buffers.
Stringdata_list strings_;
// Size of string table.
- off_t strtab_size_;
- // Next Stringdata index.
- unsigned int next_index_;
- // Next key value for a string we don't copy.
- int next_uncopied_key_;
+ section_size_type strtab_size_;
// Whether to reserve offset 0 to hold the null string.
bool zero_null_;
+ // Whether to optimize the string table.
+ bool optimize_;
+ // offset of the next string.
+ section_offset_type offset_;
+ // The alignment of strings in the stringpool.
+ uint64_t addralign_;
};
// The most common type of Stringpool.