2 * Copyright (c) 1993-1994 by Xerox Corporation. All rights reserved.
4 * THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
5 * OR IMPLIED. ANY USE IS AT YOUR OWN RISK.
7 * Permission is hereby granted to use or copy this program
8 * for any purpose, provided the above notices are retained on all copies.
9 * Permission to modify the code and to distribute modified code is granted,
10 * provided the above notices are retained, and a notice that the code was
11 * modified is included with the above copyright notice.
13 * Author: Hans-J. Boehm (boehm@parc.xerox.com)
17 * Cords are immutable character strings. A number of operations
18 * on long cords are much more efficient than their strings.h counterpart.
19 * In particular, concatenation takes constant time independent of the length
20 * of the arguments. (Cords are represented as trees, with internal
21 * nodes representing concatenation and leaves consisting of either C
22 * strings or a functional description of the string.)
24 * The following are reasonable applications of cords. They would perform
25 * unacceptably if C strings were used:
26 * - A compiler that produces assembly language output by repeatedly
27 * concatenating instructions onto a cord representing the output file.
28 * - A text editor that converts the input file to a cord, and then
29 * performs editing operations by producing a new cord representing
30 * the file after each character change (and keeping the old ones in an
33 * For optimal performance, cords should be built by
34 * concatenating short sections.
35 * This interface is designed for maximum compatibility with C strings.
36 * ASCII NUL characters may be embedded in cords using CORD_from_fn.
37 * This is handled correctly, but CORD_to_char_star will produce a string
38 * with embedded NULs when given such a cord.
40 * This interface is fairly big, largely for performance reasons.
41 * The most basic constants and functions:
43 * CORD - the type of a cord;
44 * CORD_EMPTY - empty cord;
45 * CORD_len(cord) - length of a cord;
46 * CORD_cat(cord1,cord2) - concatenation of two cords;
47 * CORD_substr(cord, start, len) - substring (or subcord);
48 * CORD_pos i; CORD_FOR(i, cord) { ... CORD_pos_fetch(i) ... } -
49 * examine each character in a cord. CORD_pos_fetch(i) is the char.
50 * CORD_fetch(int i) - Retrieve i'th character (slowly).
51 * CORD_cmp(cord1, cord2) - compare two cords.
52 * CORD_from_file(FILE * f) - turn a read-only file into a cord.
53 * CORD_to_char_star(cord) - convert to C string.
54 * (Non-NULL C constant strings are cords.)
55 * CORD_printf (etc.) - cord version of printf. Use %r for cords.
62 /* Cords have type const char *. This is cheating quite a bit, and not */
63 /* 100% portable. But it means that nonempty character string */
64 /* constants may be used as cords directly, provided the string is */
65 /* never modified in place. The empty cord is represented by, and */
66 /* can be written as, 0. */
68 typedef const char * CORD;
70 /* An empty cord is always represented as nil */
73 /* Is a nonempty cord represented as a C string? */
74 #define CORD_IS_STRING(s) (*(s) != '\0')
76 /* Concatenate two cords. If the arguments are C strings, they may */
77 /* not be subsequently altered. */
78 CORD CORD_cat(CORD x, CORD y);
80 /* Concatenate a cord and a C string with known length. Except for the */
81 /* empty string case, this is a special case of CORD_cat. Since the */
82 /* length is known, it can be faster. */
83 /* The string y is shared with the resulting CORD. Hence it should */
84 /* not be altered by the caller. */
85 CORD CORD_cat_char_star(CORD x, const char * y, size_t leny);
87 /* Compute the length of a cord */
88 size_t CORD_len(CORD x);
90 /* Cords may be represented by functions defining the ith character */
91 typedef char (* CORD_fn)(size_t i, void * client_data);
93 /* Turn a functional description into a cord. */
94 CORD CORD_from_fn(CORD_fn fn, void * client_data, size_t len);
96 /* Return the substring (subcord really) of x with length at most n, */
97 /* starting at position i. (The initial character has position 0.) */
98 CORD CORD_substr(CORD x, size_t i, size_t n);
100 /* Return the argument, but rebalanced to allow more efficient */
101 /* character retrieval, substring operations, and comparisons. */
102 /* This is useful only for cords that were built using repeated */
103 /* concatenation. Guarantees log time access to the result, unless */
104 /* x was obtained through a large number of repeated substring ops */
105 /* or the embedded functional descriptions take longer to evaluate. */
106 /* May reallocate significant parts of the cord. The argument is not */
107 /* modified; only the result is balanced. */
108 CORD CORD_balance(CORD x);
110 /* The following traverse a cord by applying a function to each */
111 /* character. This is occasionally appropriate, especially where */
112 /* speed is crucial. But, since C doesn't have nested functions, */
113 /* clients of this sort of traversal are clumsy to write. Consider */
114 /* the functions that operate on cord positions instead. */
116 /* Function to iteratively apply to individual characters in cord. */
117 typedef int (* CORD_iter_fn)(char c, void * client_data);
119 /* Function to apply to substrings of a cord. Each substring is a */
120 /* a C character string, not a general cord. */
121 typedef int (* CORD_batched_iter_fn)(const char * s, void * client_data);
122 #define CORD_NO_FN ((CORD_batched_iter_fn)0)
124 /* Apply f1 to each character in the cord, in ascending order, */
125 /* starting at position i. If */
126 /* f2 is not CORD_NO_FN, then multiple calls to f1 may be replaced by */
127 /* a single call to f2. The parameter f2 is provided only to allow */
128 /* some optimization by the client. This terminates when the right */
129 /* end of this string is reached, or when f1 or f2 return != 0. In the */
130 /* latter case CORD_iter returns != 0. Otherwise it returns 0. */
131 /* The specified value of i must be < CORD_len(x). */
132 int CORD_iter5(CORD x, size_t i, CORD_iter_fn f1,
133 CORD_batched_iter_fn f2, void * client_data);
135 /* A simpler version that starts at 0, and without f2: */
136 int CORD_iter(CORD x, CORD_iter_fn f1, void * client_data);
137 #define CORD_iter(x, f1, cd) CORD_iter5(x, 0, f1, CORD_NO_FN, cd)
139 /* Similar to CORD_iter5, but end-to-beginning. No provisions for */
140 /* CORD_batched_iter_fn. */
141 int CORD_riter4(CORD x, size_t i, CORD_iter_fn f1, void * client_data);
143 /* A simpler version that starts at the end: */
144 int CORD_riter(CORD x, CORD_iter_fn f1, void * client_data);
146 /* Functions that operate on cord positions. The easy way to traverse */
147 /* cords. A cord position is logically a pair consisting of a cord */
148 /* and an index into that cord. But it is much faster to retrieve a */
149 /* character based on a position than on an index. Unfortunately, */
150 /* positions are big (order of a few 100 bytes), so allocate them with */
152 /* Things in cord_pos.h should be treated as opaque, except as */
153 /* described below. Also note that */
154 /* CORD_pos_fetch, CORD_next and CORD_prev have both macro and function */
155 /* definitions. The former may evaluate their argument more than once. */
156 #include "private/cord_pos.h"
159 Visible definitions from above:
161 typedef <OPAQUE but fairly big> CORD_pos[1];
163 * Extract the cord from a position:
164 CORD CORD_pos_to_cord(CORD_pos p);
166 * Extract the current index from a position:
167 size_t CORD_pos_to_index(CORD_pos p);
169 * Fetch the character located at the given position:
170 char CORD_pos_fetch(CORD_pos p);
172 * Initialize the position to refer to the given cord and index.
173 * Note that this is the most expensive function on positions:
174 void CORD_set_pos(CORD_pos p, CORD x, size_t i);
176 * Advance the position to the next character.
177 * P must be initialized and valid.
178 * Invalidates p if past end:
179 void CORD_next(CORD_pos p);
181 * Move the position to the preceding character.
182 * P must be initialized and valid.
183 * Invalidates p if past beginning:
184 void CORD_prev(CORD_pos p);
186 * Is the position valid, i.e. inside the cord?
187 int CORD_pos_valid(CORD_pos p);
189 #define CORD_FOR(pos, cord) \
190 for (CORD_set_pos(pos, cord, 0); CORD_pos_valid(pos); CORD_next(pos))
193 /* An out of memory handler to call. May be supplied by client. */
194 /* Must not return. */
195 extern void (* CORD_oom_fn)(void);
197 /* Dump the representation of x to stdout in an implementation defined */
198 /* manner. Intended for debugging only. */
199 void CORD_dump(CORD x);
201 /* The following could easily be implemented by the client. They are */
202 /* provided in cordxtra.c for convenience. */
204 /* Concatenate a character to the end of a cord. */
205 CORD CORD_cat_char(CORD x, char c);
207 /* Concatenate n cords. */
208 CORD CORD_catn(int n, /* CORD */ ...);
210 /* Return the character in CORD_substr(x, i, 1) */
211 char CORD_fetch(CORD x, size_t i);
213 /* Return < 0, 0, or > 0, depending on whether x < y, x = y, x > y */
214 int CORD_cmp(CORD x, CORD y);
216 /* A generalization that takes both starting positions for the */
217 /* comparison, and a limit on the number of characters to be compared. */
218 int CORD_ncmp(CORD x, size_t x_start, CORD y, size_t y_start, size_t len);
220 /* Find the first occurrence of s in x at position start or later. */
221 /* Return the position of the first character of s in x, or */
222 /* CORD_NOT_FOUND if there is none. */
223 size_t CORD_str(CORD x, size_t start, CORD s);
225 /* Return a cord consisting of i copies of (possibly NUL) c. Dangerous */
226 /* in conjunction with CORD_to_char_star. */
227 /* The resulting representation takes constant space, independent of i. */
228 CORD CORD_chars(char c, size_t i);
229 #define CORD_nul(i) CORD_chars('\0', (i))
231 /* Turn a file into cord. The file must be seekable. Its contents */
232 /* must remain constant. The file may be accessed as an immediate */
233 /* result of this call and/or as a result of subsequent accesses to */
234 /* the cord. Short files are likely to be immediately read, but */
235 /* long files are likely to be read on demand, possibly relying on */
236 /* stdio for buffering. */
237 /* We must have exclusive access to the descriptor f, i.e. we may */
238 /* read it at any time, and expect the file pointer to be */
239 /* where we left it. Normally this should be invoked as */
240 /* CORD_from_file(fopen(...)) */
241 /* CORD_from_file arranges to close the file descriptor when it is no */
242 /* longer needed (e.g. when the result becomes inaccessible). */
243 /* The file f must be such that ftell reflects the actual character */
244 /* position in the file, i.e. the number of characters that can be */
245 /* or were read with fread. On UNIX systems this is always true. On */
246 /* MS Windows systems, f must be opened in binary mode. */
247 CORD CORD_from_file(FILE * f);
249 /* Equivalent to the above, except that the entire file will be read */
250 /* and the file pointer will be closed immediately. */
251 /* The binary mode restriction from above does not apply. */
252 CORD CORD_from_file_eager(FILE * f);
254 /* Equivalent to the above, except that the file will be read on demand.*/
255 /* The binary mode restriction applies. */
256 CORD CORD_from_file_lazy(FILE * f);
258 /* Turn a cord into a C string. The result shares no structure with */
259 /* x, and is thus modifiable. */
260 char * CORD_to_char_star(CORD x);
262 /* Turn a C string into a CORD. The C string is copied, and so may */
263 /* subsequently be modified. */
264 CORD CORD_from_char_star(const char *s);
266 /* Identical to the above, but the result may share structure with */
267 /* the argument and is thus not modifiable. */
268 const char * CORD_to_const_char_star(CORD x);
270 /* Write a cord to a file, starting at the current position. No */
271 /* trailing NULs are newlines are added. */
272 /* Returns EOF if a write error occurs, 1 otherwise. */
273 int CORD_put(CORD x, FILE * f);
275 /* "Not found" result for the following two functions. */
276 #define CORD_NOT_FOUND ((size_t)(-1))
278 /* A vague analog of strchr. Returns the position (an integer, not */
279 /* a pointer) of the first occurrence of (char) c inside x at position */
280 /* i or later. The value i must be < CORD_len(x). */
281 size_t CORD_chr(CORD x, size_t i, int c);
283 /* A vague analog of strrchr. Returns index of the last occurrence */
284 /* of (char) c inside x at position i or earlier. The value i */
285 /* must be < CORD_len(x). */
286 size_t CORD_rchr(CORD x, size_t i, int c);
289 /* The following are also not primitive, but are implemented in */
290 /* cordprnt.c. They provide functionality similar to the ANSI C */
291 /* functions with corresponding names, but with the following */
292 /* additions and changes: */
293 /* 1. A %r conversion specification specifies a CORD argument. Field */
294 /* width, precision, etc. have the same semantics as for %s. */
295 /* (Note that %c, %C, and %S were already taken.) */
296 /* 2. The format string is represented as a CORD. */
297 /* 3. CORD_sprintf and CORD_vsprintf assign the result through the 1st */ /* argument. Unlike their ANSI C versions, there is no need to guess */
298 /* the correct buffer size. */
299 /* 4. Most of the conversions are implement through the native */
300 /* vsprintf. Hence they are usually no faster, and */
301 /* idiosyncracies of the native printf are preserved. However, */
302 /* CORD arguments to CORD_sprintf and CORD_vsprintf are NOT copied; */
303 /* the result shares the original structure. This may make them */
304 /* very efficient in some unusual applications. */
305 /* The format string is copied. */
306 /* All functions return the number of characters generated or -1 on */
307 /* error. This complies with the ANSI standard, but is inconsistent */
308 /* with some older implementations of sprintf. */
310 /* The implementation of these is probably less portable than the rest */
311 /* of this package. */
317 int CORD_sprintf(CORD * out, CORD format, ...);
318 int CORD_vsprintf(CORD * out, CORD format, va_list args);
319 int CORD_fprintf(FILE * f, CORD format, ...);
320 int CORD_vfprintf(FILE * f, CORD format, va_list args);
321 int CORD_printf(CORD format, ...);
322 int CORD_vprintf(CORD format, va_list args);
324 #endif /* CORD_NO_IO */