2 * UCW Library -- Universal Sorter
4 * (c) 2001--2007 Martin Mares <mj@ucw.cz>
5 * (c) 2004 Robert Spalek <robert@ucw.cz>
7 * This software may be freely distributed and used according to the terms
8 * of the GNU Lesser General Public License.
12 * This is not a normal header file, but a generator of sorting
13 * routines. Each time you include it with parameters set in the
14 * corresponding preprocessor macros, it generates a file sorter
15 * with the parameters given.
17 * FIXME: explain the basics (keys and data, callbacks etc.)
19 * Basic parameters and callbacks:
21 * SORT_PREFIX(x) add a name prefix (used on all global names
22 * defined by the sorter)
24 * SORT_KEY data type capable of storing a single key.
25 * SORT_KEY_REGULAR data type holding a single key both in memory and on disk;
26 * in this case, bread() and bwrite() is used to read/write keys
27 * and it's also assumed that the keys are not very long.
28 * int PREFIX_compare(SORT_KEY *a, SORT_KEY *b)
29 * compares two keys, result like strcmp(). Mandatory.
30 * int PREFIX_read_key(struct fastbuf *f, SORT_KEY *k)
31 * reads a key from a fastbuf, returns nonzero=ok, 0=EOF.
32 * Mandatory unless SORT_KEY_REGULAR is defined.
33 * void PREFIX_write_key(struct fastbuf *f, SORT_KEY *k)
34 * writes a key to a fastbuf. Mandatory unless SORT_KEY_REGULAR.
36 * SORT_KEY_SIZE(key) returns the real size of a key (a SORT_KEY type in memory
37 * can be truncated to this number of bytes without any harm;
38 * used to save memory when the keys have variable sizes).
39 * Default: always store the whole SORT_KEY.
40 * SORT_DATA_SIZE(key) gets a key and returns the amount of data following it.
41 * Default: records consist of keys only.
45 * SORT_INT(key) We are sorting by an integer value. In this mode, PREFIX_compare
46 * is supplied automatically and the sorting function gets an extra
47 * parameter specifying a range of the integers. The better the range
48 * fits, the faster we sort. Sets up SORT_HASH_xxx automatically.
50 * Hashing (optional, but it can speed sorting up):
52 * SORT_HASH_BITS signals that a monotone hashing function returning a given number of
53 * bits is available. Monotone hash is a function f such that f(x) < f(y)
54 * implies x < y and which is approximately uniformly distributed.
55 * uns PREFIX_hash(SORT_KEY *a)
59 * SORT_UNIFY merge items with identical keys, needs the following functions:
60 * void PREFIX_write_merged(struct fastbuf *f, SORT_KEY **keys, void **data, uns n, void *buf)
61 * takes n records in memory with keys which compare equal and writes
62 * a single record to the given fastbuf. `buf' points to a buffer which
63 * is guaranteed to hold all given records.
64 * void PREFIX_copy_merged(SORT_KEY **keys, struct fastbuf **data, uns n, struct fastbuf *dest)
65 * takes n records with keys in memory and data in fastbufs and writes
68 * Input (choose one of these):
70 * SORT_INPUT_FILE file of a given name
71 * SORT_INPUT_FB seekable fastbuf stream
72 * SORT_INPUT_PIPE non-seekable fastbuf stream
73 * SORT_INPUT_PRESORT custom presorter. Calls function
74 * int PREFIX_presort(struct fastbuf *dest, void *buf, size_t bufsize);
75 * to get successive batches of pre-sorted data.
76 * The function is passed a page-aligned presorting buffer.
77 * It returns 1 on success or 0 on EOF.
79 * Output (chose one of these):
81 * SORT_OUTPUT_FILE file of a given name
82 * SORT_OUTPUT_FB temporary fastbuf stream
83 * SORT_OUTPUT_THIS_FB a given fastbuf stream which can already contain a header
87 * SORT_UNIQUE all items have distinct keys (checked in debug mode)
89 * The function generated:
91 * <outfb> PREFIX_SORT(<in>, <out> [,<range>]), where:
92 * <in> = input file name/fastbuf or NULL
93 * <out> = output file name/fastbuf or NULL
94 * <range> = maximum integer value for the SORT_INT mode
95 * <outfb> = output fastbuf (in SORT_OUTPUT_FB mode)
97 * After including this file, all parameter macros are automatically
101 #include "lib/sorter/common.h"
102 #include "lib/fastbuf.h"
106 #define P(x) SORT_PREFIX(x)
108 #ifdef SORT_KEY_REGULAR
109 typedef SORT_KEY_REGULAR P(key);
110 static inline int P(read_key) (struct fastbuf *f, P(key) *k)
112 return breadb(f, k, sizeof(P(key)));
114 static inline void P(write_key) (struct fastbuf *f, P(key) *k)
116 bwrite(f, k, sizeof(P(key)));
118 #elif defined(SORT_KEY)
119 typedef SORT_KEY P(key);
121 #error Missing definition of sorting key.
125 static inline int P(compare) (P(key) *x, P(key) *y)
127 if (SORT_INT(*x) < SORT_INT(*y))
129 if (SORT_INT(*x) > SORT_INT(*y))
134 #ifndef SORT_HASH_BITS
135 static inline int P(hash) (P(key) *x)
137 return SORT_INT((*x));
147 #define SWAP(x,y,z) do { z=x; x=y; y=z; } while(0)
149 #if defined(SORT_UNIQUE) && defined(DEBUG_ASSERTS)
150 #define SORT_ASSERT_UNIQUE
156 #define SORT_KEY_SIZE(key) sizeof(key)
159 #ifdef SORT_DATA_SIZE
160 #define SORT_VAR_DATA
162 #define SORT_DATA_SIZE(key) 0
165 static inline void P(copy_data)(P(key) *key, struct fastbuf *in, struct fastbuf *out)
167 P(write_key)(out, key);
169 bbcopy(in, out, SORT_DATA_SIZE(*key));
175 #if defined(SORT_VAR_KEY) || defined(SORT_VAR_DATA) || defined(SORT_UNIFY)
176 #include "lib/sorter/s-internal.h"
178 #include "lib/sorter/s-fixint.h"
181 #include "lib/sorter/s-twoway.h"
183 #if defined(SORT_HASH_BITS) || defined(SORT_INT)
184 #include "lib/sorter/s-radix.h"
187 static struct fastbuf *P(sort)(
188 #ifdef SORT_INPUT_FILE
193 #ifdef SORT_OUTPUT_FILE
203 struct sort_context ctx;
204 bzero(&ctx, sizeof(ctx));
206 #ifdef SORT_INPUT_FILE
207 ctx.in_fb = bopen(in, O_RDONLY, sorter_stream_bufsize);
208 ctx.in_size = bfilesize(ctx.in_fb);
209 #elif defined(SORT_INPUT_FB)
211 ctx.in_size = bfilesize(in);
212 #elif defined(SORT_INPUT_PIPE)
214 ctx.in_size = ~(u64)0;
215 #elif defined(SORT_INPUT_PRESORT)
217 ctx.custom_presort = P(presort);
218 ctx.in_size = ~(u64)0;
220 #error No input given.
223 #ifdef SORT_OUTPUT_FB
225 #elif defined(SORT_OUTPUT_THIS_FB)
227 #elif defined(SORT_OUTPUT_FILE)
228 /* Just assume fastbuf output and rename the fastbuf later */
230 #error No output given.
233 #ifdef SORT_HASH_BITS
234 ctx.hash_bits = SORT_HASH_BITS;
235 ctx.radix_split = P(radix_split);
236 #elif defined(SORT_INT)
238 while (ctx.hash_bits < 32 && (int_range >> ctx.hash_bits))
240 ctx.radix_split = P(radix_split);
243 ctx.internal_sort = P(internal);
244 ctx.twoway_merge = P(twoway_merge);
248 #ifdef SORT_OUTPUT_FILE
249 if (rename(ctx.out_fb->name, out) < 0)
250 die("Cannot rename %s to %s: %m", ctx.out_fb->name, out);
251 bconfig(ctx.out_fb, BCONFIG_IS_TEMP_FILE, 0);
260 #undef SORT_KEY_REGULAR
262 #undef SORT_DATA_SIZE
266 #undef SORT_HASH_BITS
268 #undef SORT_INPUT_FILE
270 #undef SORT_INPUT_PRESORT
271 #undef SORT_OUTPUT_FILE
272 #undef SORT_OUTPUT_FB
273 #undef SORT_OUTPUT_THIS_FB
275 #undef SORT_ASSERT_UNIQUE
279 /* FIXME: Check that we undef everything we should. */