suffixarray.go

     1  // Copyright 2010 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  // Package suffixarray implements substring search in logarithmic time using
     6  // an in-memory suffix array.
     7  //
     8  // Example use:
     9  //
    10  //	// create index for some data
    11  //	index := suffixarray.New(data)
    12  //
    13  //	// lookup byte slice s
    14  //	offsets1 := index.Lookup(s, -1) // the list of all indices where s occurs in data
    15  //	offsets2 := index.Lookup(s, 3)  // the list of at most 3 indices where s occurs in data
    16  //
    17  package suffixarray
    18  
    19  import (
    20  	"bytes"
    21  	"encoding/binary"
    22  	"errors"
    23  	"io"
    24  	"math"
    25  	"regexp"
    26  	"sort"
    27  )
    28  
    29  // Can change for testing
    30  var maxData32 int = realMaxData32
    31  
    32  const realMaxData32 = math.MaxInt32
    33  
    34  // Index implements a suffix array for fast substring search.
    35  type Index struct {
    36  	data []byte
    37  	sa   ints // suffix array for data; sa.len() == len(data)
    38  }
    39  
    40  // An ints is either an []int32 or an []int64.
    41  // That is, one of them is empty, and one is the real data.
    42  // The int64 form is used when len(data) > maxData32
    43  type ints struct {
    44  	int32 []int32
    45  	int64 []int64
    46  }
    47  
    48  func (a *ints) len() int {
    49  	return len(a.int32) + len(a.int64)
    50  }
    51  
    52  func (a *ints) get(i int) int64 {
    53  	if a.int32 != nil {
    54  		return int64(a.int32[i])
    55  	}
    56  	return a.int64[i]
    57  }
    58  
    59  func (a *ints) set(i int, v int64) {
    60  	if a.int32 != nil {
    61  		a.int32[i] = int32(v)
    62  	} else {
    63  		a.int64[i] = v
    64  	}
    65  }
    66  
    67  func (a *ints) slice(i, j int) ints {
    68  	if a.int32 != nil {
    69  		return ints{a.int32[i:j], nil}
    70  	}
    71  	return ints{nil, a.int64[i:j]}
    72  }
    73  
    74  // New creates a new Index for data.
    75  // Index creation time is O(N) for N = len(data).
    76  func New(data []byte) *Index {
    77  	ix := &Index{data: data}
    78  	if len(data) <= maxData32 {
    79  		ix.sa.int32 = make([]int32, len(data))
    80  		text_32(data, ix.sa.int32)
    81  	} else {
    82  		ix.sa.int64 = make([]int64, len(data))
    83  		text_64(data, ix.sa.int64)
    84  	}
    85  	return ix
    86  }
    87  
    88  // writeInt writes an int x to w using buf to buffer the write.
    89  func writeInt(w io.Writer, buf []byte, x int) error {
    90  	binary.PutVarint(buf, int64(x))
    91  	_, err := w.Write(buf[0:binary.MaxVarintLen64])
    92  	return err
    93  }
    94  
    95  // readInt reads an int x from r using buf to buffer the read and returns x.
    96  func readInt(r io.Reader, buf []byte) (int64, error) {
    97  	_, err := io.ReadFull(r, buf[0:binary.MaxVarintLen64]) // ok to continue with error
    98  	x, _ := binary.Varint(buf)
    99  	return x, err
   100  }
   101  
   102  // writeSlice writes data[:n] to w and returns n.
   103  // It uses buf to buffer the write.
   104  func writeSlice(w io.Writer, buf []byte, data ints) (n int, err error) {
   105  	// encode as many elements as fit into buf
   106  	p := binary.MaxVarintLen64
   107  	m := data.len()
   108  	for ; n < m && p+binary.MaxVarintLen64 <= len(buf); n++ {
   109  		p += binary.PutUvarint(buf[p:], uint64(data.get(n)))
   110  	}
   111  
   112  	// update buffer size
   113  	binary.PutVarint(buf, int64(p))
   114  
   115  	// write buffer
   116  	_, err = w.Write(buf[0:p])
   117  	return
   118  }
   119  
   120  var errTooBig = errors.New("suffixarray: data too large")
   121  
   122  // readSlice reads data[:n] from r and returns n.
   123  // It uses buf to buffer the read.
   124  func readSlice(r io.Reader, buf []byte, data ints) (n int, err error) {
   125  	// read buffer size
   126  	var size64 int64
   127  	size64, err = readInt(r, buf)
   128  	if err != nil {
   129  		return
   130  	}
   131  	if int64(int(size64)) != size64 || int(size64) < 0 {
   132  		// We never write chunks this big anyway.
   133  		return 0, errTooBig
   134  	}
   135  	size := int(size64)
   136  
   137  	// read buffer w/o the size
   138  	if _, err = io.ReadFull(r, buf[binary.MaxVarintLen64:size]); err != nil {
   139  		return
   140  	}
   141  
   142  	// decode as many elements as present in buf
   143  	for p := binary.MaxVarintLen64; p < size; n++ {
   144  		x, w := binary.Uvarint(buf[p:])
   145  		data.set(n, int64(x))
   146  		p += w
   147  	}
   148  
   149  	return
   150  }
   151  
   152  const bufSize = 16 << 10 // reasonable for BenchmarkSaveRestore
   153  
   154  // Read reads the index from r into x; x must not be nil.
   155  func (x *Index) Read(r io.Reader) error {
   156  	// buffer for all reads
   157  	buf := make([]byte, bufSize)
   158  
   159  	// read length
   160  	n64, err := readInt(r, buf)
   161  	if err != nil {
   162  		return err
   163  	}
   164  	if int64(int(n64)) != n64 || int(n64) < 0 {
   165  		return errTooBig
   166  	}
   167  	n := int(n64)
   168  
   169  	// allocate space
   170  	if 2*n < cap(x.data) || cap(x.data) < n || x.sa.int32 != nil && n > maxData32 || x.sa.int64 != nil && n <= maxData32 {
   171  		// new data is significantly smaller or larger than
   172  		// existing buffers - allocate new ones
   173  		x.data = make([]byte, n)
   174  		x.sa.int32 = nil
   175  		x.sa.int64 = nil
   176  		if n <= maxData32 {
   177  			x.sa.int32 = make([]int32, n)
   178  		} else {
   179  			x.sa.int64 = make([]int64, n)
   180  		}
   181  	} else {
   182  		// re-use existing buffers
   183  		x.data = x.data[0:n]
   184  		x.sa = x.sa.slice(0, n)
   185  	}
   186  
   187  	// read data
   188  	if _, err := io.ReadFull(r, x.data); err != nil {
   189  		return err
   190  	}
   191  
   192  	// read index
   193  	sa := x.sa
   194  	for sa.len() > 0 {
   195  		n, err := readSlice(r, buf, sa)
   196  		if err != nil {
   197  			return err
   198  		}
   199  		sa = sa.slice(n, sa.len())
   200  	}
   201  	return nil
   202  }
   203  
   204  // Write writes the index x to w.
   205  func (x *Index) Write(w io.Writer) error {
   206  	// buffer for all writes
   207  	buf := make([]byte, bufSize)
   208  
   209  	// write length
   210  	if err := writeInt(w, buf, len(x.data)); err != nil {
   211  		return err
   212  	}
   213  
   214  	// write data
   215  	if _, err := w.Write(x.data); err != nil {
   216  		return err
   217  	}
   218  
   219  	// write index
   220  	sa := x.sa
   221  	for sa.len() > 0 {
   222  		n, err := writeSlice(w, buf, sa)
   223  		if err != nil {
   224  			return err
   225  		}
   226  		sa = sa.slice(n, sa.len())
   227  	}
   228  	return nil
   229  }
   230  
   231  // Bytes returns the data over which the index was created.
   232  // It must not be modified.
   233  //
   234  func (x *Index) Bytes() []byte {
   235  	return x.data
   236  }
   237  
   238  func (x *Index) at(i int) []byte {
   239  	return x.data[x.sa.get(i):]
   240  }
   241  
   242  // lookupAll returns a slice into the matching region of the index.
   243  // The runtime is O(log(N)*len(s)).
   244  func (x *Index) lookupAll(s []byte) ints {
   245  	// find matching suffix index range [i:j]
   246  	// find the first index where s would be the prefix
   247  	i := sort.Search(x.sa.len(), func(i int) bool { return bytes.Compare(x.at(i), s) >= 0 })
   248  	// starting at i, find the first index at which s is not a prefix
   249  	j := i + sort.Search(x.sa.len()-i, func(j int) bool { return !bytes.HasPrefix(x.at(j+i), s) })
   250  	return x.sa.slice(i, j)
   251  }
   252  
   253  // Lookup returns an unsorted list of at most n indices where the byte string s
   254  // occurs in the indexed data. If n < 0, all occurrences are returned.
   255  // The result is nil if s is empty, s is not found, or n == 0.
   256  // Lookup time is O(log(N)*len(s) + len(result)) where N is the
   257  // size of the indexed data.
   258  //
   259  func (x *Index) Lookup(s []byte, n int) (result []int) {
   260  	if len(s) > 0 && n != 0 {
   261  		matches := x.lookupAll(s)
   262  		count := matches.len()
   263  		if n < 0 || count < n {
   264  			n = count
   265  		}
   266  		// 0 <= n <= count
   267  		if n > 0 {
   268  			result = make([]int, n)
   269  			if matches.int32 != nil {
   270  				for i := range result {
   271  					result[i] = int(matches.int32[i])
   272  				}
   273  			} else {
   274  				for i := range result {
   275  					result[i] = int(matches.int64[i])
   276  				}
   277  			}
   278  		}
   279  	}
   280  	return
   281  }
   282  
   283  // FindAllIndex returns a sorted list of non-overlapping matches of the
   284  // regular expression r, where a match is a pair of indices specifying
   285  // the matched slice of x.Bytes(). If n < 0, all matches are returned
   286  // in successive order. Otherwise, at most n matches are returned and
   287  // they may not be successive. The result is nil if there are no matches,
   288  // or if n == 0.
   289  //
   290  func (x *Index) FindAllIndex(r *regexp.Regexp, n int) (result [][]int) {
   291  	// a non-empty literal prefix is used to determine possible
   292  	// match start indices with Lookup
   293  	prefix, complete := r.LiteralPrefix()
   294  	lit := []byte(prefix)
   295  
   296  	// worst-case scenario: no literal prefix
   297  	if prefix == "" {
   298  		return r.FindAllIndex(x.data, n)
   299  	}
   300  
   301  	// if regexp is a literal just use Lookup and convert its
   302  	// result into match pairs
   303  	if complete {
   304  		// Lookup returns indices that may belong to overlapping matches.
   305  		// After eliminating them, we may end up with fewer than n matches.
   306  		// If we don't have enough at the end, redo the search with an
   307  		// increased value n1, but only if Lookup returned all the requested
   308  		// indices in the first place (if it returned fewer than that then
   309  		// there cannot be more).
   310  		for n1 := n; ; n1 += 2 * (n - len(result)) /* overflow ok */ {
   311  			indices := x.Lookup(lit, n1)
   312  			if len(indices) == 0 {
   313  				return
   314  			}
   315  			sort.Ints(indices)
   316  			pairs := make([]int, 2*len(indices))
   317  			result = make([][]int, len(indices))
   318  			count := 0
   319  			prev := 0
   320  			for _, i := range indices {
   321  				if count == n {
   322  					break
   323  				}
   324  				// ignore indices leading to overlapping matches
   325  				if prev <= i {
   326  					j := 2 * count
   327  					pairs[j+0] = i
   328  					pairs[j+1] = i + len(lit)
   329  					result[count] = pairs[j : j+2]
   330  					count++
   331  					prev = i + len(lit)
   332  				}
   333  			}
   334  			result = result[0:count]
   335  			if len(result) >= n || len(indices) != n1 {
   336  				// found all matches or there's no chance to find more
   337  				// (n and n1 can be negative)
   338  				break
   339  			}
   340  		}
   341  		if len(result) == 0 {
   342  			result = nil
   343  		}
   344  		return
   345  	}
   346  
   347  	// regexp has a non-empty literal prefix; Lookup(lit) computes
   348  	// the indices of possible complete matches; use these as starting
   349  	// points for anchored searches
   350  	// (regexp "^" matches beginning of input, not beginning of line)
   351  	r = regexp.MustCompile("^" + r.String()) // compiles because r compiled
   352  
   353  	// same comment about Lookup applies here as in the loop above
   354  	for n1 := n; ; n1 += 2 * (n - len(result)) /* overflow ok */ {
   355  		indices := x.Lookup(lit, n1)
   356  		if len(indices) == 0 {
   357  			return
   358  		}
   359  		sort.Ints(indices)
   360  		result = result[0:0]
   361  		prev := 0
   362  		for _, i := range indices {
   363  			if len(result) == n {
   364  				break
   365  			}
   366  			m := r.FindIndex(x.data[i:]) // anchored search - will not run off
   367  			// ignore indices leading to overlapping matches
   368  			if m != nil && prev <= i {
   369  				m[0] = i // correct m
   370  				m[1] += i
   371  				result = append(result, m)
   372  				prev = m[1]
   373  			}
   374  		}
   375  		if len(result) >= n || len(indices) != n1 {
   376  			// found all matches or there's no chance to find more
   377  			// (n and n1 can be negative)
   378  			break
   379  		}
   380  	}
   381  	if len(result) == 0 {
   382  		result = nil
   383  	}
   384  	return
   385  }
   386
View as plain text