This patch remembers backup file names patch has already created in a binary tree, and prevents patch from overwriting backup files it has created before. The bug can be reproduced as follows: $ cat > f < one $ cat > f.patch <--- f.orig 2003-04-09 01:48:01.000000000 +0200 <+++ f 2003-04-09 01:49:17.000000000 +0200 <@@ -2 +2 @@ <-one <+two <--- f.orig 2003-04-09 01:48:01.000000000 +0200 <+++ f 2003-04-09 01:49:17.000000000 +0200 <@@ -2 +2 @@ <-two <+three $ patch -p0 < f.patch > patching file f > Hunk #1 succeeded at 1 (offset -1 lines). > patching file f > Hunk #1 succeeded at 1 (offset -1 lines). $ cat f.orig > one $ rm f f.orig f.patch Here is another test case with hard links between source files. $ cat > f < one $ ln f g $ cat > fg.patch <--- f.orig 2003-04-09 01:48:01.000000000 +0200 <+++ f 2003-04-09 01:49:17.000000000 +0200 <@@ -2 +2 @@ <-one <+two <--- g.orig 2003-04-09 01:48:01.000000000 +0200 <+++ g 2003-04-09 01:49:17.000000000 +0200 <@@ -2 +2 @@ <-one <+two $ patch -p0 < fg.patch > patching file f > Hunk #1 succeeded at 1 (offset -1 lines). > patching file g > Hunk #1 succeeded at 1 (offset -1 lines). $ cat f.orig > one $ cat g.orig > one $ rm f f.orig g g.orig fg.patch Index: patch-2.5.9/Makefile.in =================================================================== --- patch-2.5.9.orig/Makefile.in 2003-05-20 18:15:48.000000000 +0200 +++ patch-2.5.9/Makefile.in 2003-10-07 14:53:07.000000000 +0200 @@ -62,7 +62,7 @@ CONFIG_STATUS = config.status SHELL = /bin/sh LIBSRCS = error.c malloc.c memchr.c mkdir.c \ - realloc.c rmdir.c strcasecmp.c strncasecmp.c + realloc.c rmdir.c strcasecmp.c strncasecmp.c hash.c SRCS = $(LIBSRCS) \ addext.c argmatch.c backupfile.c \ basename.c dirname.c \ @@ -78,12 +78,12 @@ OBJS = $(LIBOBJS) \ maketime.$(OBJEXT) partime.$(OBJEXT) \ patch.$(OBJEXT) pch.$(OBJEXT) \ quote.$(OBJEXT) quotearg.$(OBJEXT) quotesys.$(OBJEXT) \ - util.$(OBJEXT) version.$(OBJEXT) xmalloc.$(OBJEXT) + util.$(OBJEXT) version.$(OBJEXT) xmalloc.$(OBJEXT) hash.$(OBJEXT) HDRS = argmatch.h backupfile.h common.h dirname.h \ error.h getopt.h gettext.h \ inp.h maketime.h partime.h pch.h \ quote.h quotearg.h quotesys.h \ - unlocked-io.h util.h version.h xalloc.h + unlocked-io.h util.h version.h xalloc.h hash.h MISC = AUTHORS COPYING ChangeLog INSTALL Makefile.in NEWS README \ aclocal.m4 \ config.hin configure configure.ac \ Index: patch-2.5.9/configure.ac =================================================================== --- patch-2.5.9.orig/configure.ac 2003-05-19 08:44:57.000000000 +0200 +++ patch-2.5.9/configure.ac 2003-10-07 16:48:30.000000000 +0200 @@ -64,6 +64,9 @@ gl_GETOPT gl_PREREQ_XMALLOC gl_QUOTE gl_QUOTEARG +gl_HASH + +ag_CHECK_NANOSECOND_STAT dnl This should be in gnulib, but isn't for some reason. AC_DEFUN([jm_PREREQ_ADDEXT], Index: patch-2.5.9/m4/hash.m4 =================================================================== --- patch-2.5.9.orig/m4/hash.m4 2003-10-07 14:53:07.000000000 +0200 +++ patch-2.5.9/m4/hash.m4 2003-10-07 14:53:07.000000000 +0200 @@ -0,0 +1,15 @@ +# hash.m4 serial 1 +dnl Copyright (C) 2002 Free Software Foundation, Inc. +dnl This file is free software, distributed under the terms of the GNU +dnl General Public License. As a special exception to the GNU General +dnl Public License, this file may be distributed as part of a program +dnl that contains a configuration script generated by Autoconf, under +dnl the same distribution terms as the rest of that program. + +AC_DEFUN([gl_HASH], +[ + dnl Prerequisites of lib/hash.c. + AC_CHECK_HEADERS_ONCE(stdlib.h) + AC_HEADER_STDBOOL + AC_CHECK_DECLS_ONCE(free malloc) +]) Index: patch-2.5.9/util.c =================================================================== --- patch-2.5.9.orig/util.c 2003-05-20 16:04:53.000000000 +0200 +++ patch-2.5.9/util.c 2003-10-07 17:24:24.000000000 +0200 @@ -45,9 +45,17 @@ # define raise(sig) kill (getpid (), sig) #endif +#if defined(HAVE_STAT_TIMEVAL) +#include +#endif + #include +#include static void makedirs (char *); +static bool fid_search (const char *, const struct stat *, bool); +# define fid_exists(name, pst) fid_search (name, pst, false) +# define insert_fid(name) fid_search (name, NULL, true) /* Move a file FROM (where *FROM_NEEDS_REMOVAL is nonzero if FROM needs removal when cleaning up at the end of execution) @@ -64,7 +72,7 @@ move_file (char const *from, int volatil struct stat to_st; int to_errno = ! backup ? -1 : stat (to, &to_st) == 0 ? 0 : errno; - if (backup) + if (backup && (to_errno || ! fid_exists (to, &to_st))) { int try_makedirs_errno = 0; char *bakname; @@ -165,6 +173,7 @@ move_file (char const *from, int volatil if (! to_dir_known_to_exist) makedirs (to); copy_file (from, to, 0, mode); + insert_fid (to); return; } @@ -173,6 +182,7 @@ move_file (char const *from, int volatil } rename_succeeded: + insert_fid (to); /* Do not clear *FROM_NEEDS_REMOVAL if it's possible that the rename returned zero because FROM and TO are hard links to the same file. */ @@ -1011,3 +1021,105 @@ Fseek (FILE *stream, file_offset offset, if (file_seek (stream, offset, ptrname) != 0) pfatal ("fseek"); } + +typedef struct +{ + dev_t fid_dev; + ino_t fid_ino; + time_t fid_mtime; + unsigned long fid_mtimensec; +} file_id; + +unsigned +file_id_hasher (file_id *entry, unsigned table_size) +{ + return ((unsigned long) entry->fid_ino + + (unsigned long) entry->fid_dev + + (unsigned long) entry->fid_mtime + + (unsigned long) entry->fid_mtimensec) % table_size; +} + +bool +file_id_comparator (file_id *entry1, file_id *entry2) +{ + return (entry1->fid_dev == entry2->fid_dev && + entry1->fid_ino == entry2->fid_ino && + entry1->fid_mtime == entry2->fid_mtime && + entry1->fid_mtimensec == entry2->fid_mtimensec); +} + +void +file_id_freer (file_id *entry) +{ + free (entry); +} + +Hash_table *file_id_hash; + +/* Check if the file identified by FILENAME and PST was already seen. If the + file was already seen, returns TRUE. If the file has not yet been seen + and INSERT is TRUE, it is inserted. PST or FILENAME may be NULL (but not + both of them). */ + +static bool +fid_search (const char *filename, const struct stat *pst, bool insert) +{ + struct stat st; + + if (!file_id_hash) + { + file_id_hash = hash_initialize (0, NULL, (Hash_hasher) file_id_hasher, + (Hash_comparator) file_id_comparator, + (Hash_data_freer) file_id_freer); + if (!file_id_hash) + pfatal ("hash_initialize"); + } + + if (!pst) + { + if (stat (filename, &st) != 0) + pfatal ("%s", quotearg (filename)); + pst = &st; + } + + if (insert) + { + file_id *pfid = xmalloc (sizeof (file_id)), *old_pfid; + pfid->fid_dev = pst->st_dev; + pfid->fid_ino = pst->st_ino; + pfid->fid_mtime = pst->st_mtime; +#if defined(HAVE_STAT_NSEC) + pfid->fid_mtimensec = pst->st_mtimensec; +#elif defined(HAVE_STAT_TIMEVAL) + pfid->fid_mtimensec = pst->st_mtim.tv_nsec; +#else + pfid->fid_mtimensec = 0; +#endif + old_pfid = hash_insert (file_id_hash, pfid); + if (!old_pfid) + pfatal ("hash_insert"); + else if (old_pfid != pfid) + { + free (pfid); + return true; + } + else + return false; + } + else + { + file_id fid; + fid.fid_dev = pst->st_dev; + fid.fid_ino = pst->st_ino; + fid.fid_mtime = pst->st_mtime; +#if defined(HAVE_STAT_NSEC) + fid.fid_mtimensec = pst->st_mtimensec; +#elif defined(HAVE_STAT_TIMEVAL) + fid.fid_mtimensec = pst->st_mtim.tv_nsec; +#else + fid.fid_mtimensec = 0; +#endif + return hash_lookup (file_id_hash, &fid) != 0; + } +} + Index: patch-2.5.9/hash.c =================================================================== --- patch-2.5.9.orig/hash.c 2003-10-07 14:53:07.000000000 +0200 +++ patch-2.5.9/hash.c 2003-10-07 14:53:07.000000000 +0200 @@ -0,0 +1,1051 @@ +/* hash - hashing table processing. + + Copyright (C) 1998, 1999, 2000, 2001, 2002, 2003 Free Software + Foundation, Inc. + + Written by Jim Meyering, 1992. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 2, or (at your option) + any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program; if not, write to the Free Software Foundation, + Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */ + +/* A generic hash table package. */ + +/* Define USE_OBSTACK to 1 if you want the allocator to use obstacks instead + of malloc. If you change USE_OBSTACK, you have to recompile! */ + +#if HAVE_CONFIG_H +# include +#endif +#if HAVE_STDLIB_H +# include +#endif + +#include +#include +#include + +#ifndef HAVE_DECL_FREE +"this configure-time declaration test was not run" +#endif +#if !HAVE_DECL_FREE +void free (); +#endif + +#ifndef HAVE_DECL_MALLOC +"this configure-time declaration test was not run" +#endif +#if !HAVE_DECL_MALLOC +char *malloc (); +#endif + +#if USE_OBSTACK +# include "obstack.h" +# ifndef obstack_chunk_alloc +# define obstack_chunk_alloc malloc +# endif +# ifndef obstack_chunk_free +# define obstack_chunk_free free +# endif +#endif + +#include "hash.h" + +struct hash_table + { + /* The array of buckets starts at BUCKET and extends to BUCKET_LIMIT-1, + for a possibility of N_BUCKETS. Among those, N_BUCKETS_USED buckets + are not empty, there are N_ENTRIES active entries in the table. */ + struct hash_entry *bucket; + struct hash_entry *bucket_limit; + unsigned n_buckets; + unsigned n_buckets_used; + unsigned n_entries; + + /* Tuning arguments, kept in a physicaly separate structure. */ + const Hash_tuning *tuning; + + /* Three functions are given to `hash_initialize', see the documentation + block for this function. In a word, HASHER randomizes a user entry + into a number up from 0 up to some maximum minus 1; COMPARATOR returns + true if two user entries compare equally; and DATA_FREER is the cleanup + function for a user entry. */ + Hash_hasher hasher; + Hash_comparator comparator; + Hash_data_freer data_freer; + + /* A linked list of freed struct hash_entry structs. */ + struct hash_entry *free_entry_list; + +#if USE_OBSTACK + /* Whenever obstacks are used, it is possible to allocate all overflowed + entries into a single stack, so they all can be freed in a single + operation. It is not clear if the speedup is worth the trouble. */ + struct obstack entry_stack; +#endif + }; + +/* A hash table contains many internal entries, each holding a pointer to + some user provided data (also called a user entry). An entry indistinctly + refers to both the internal entry and its associated user entry. A user + entry contents may be hashed by a randomization function (the hashing + function, or just `hasher' for short) into a number (or `slot') between 0 + and the current table size. At each slot position in the hash table, + starts a linked chain of entries for which the user data all hash to this + slot. A bucket is the collection of all entries hashing to the same slot. + + A good `hasher' function will distribute entries rather evenly in buckets. + In the ideal case, the length of each bucket is roughly the number of + entries divided by the table size. Finding the slot for a data is usually + done in constant time by the `hasher', and the later finding of a precise + entry is linear in time with the size of the bucket. Consequently, a + larger hash table size (that is, a larger number of buckets) is prone to + yielding shorter chains, *given* the `hasher' function behaves properly. + + Long buckets slow down the lookup algorithm. One might use big hash table + sizes in hope to reduce the average length of buckets, but this might + become inordinate, as unused slots in the hash table take some space. The + best bet is to make sure you are using a good `hasher' function (beware + that those are not that easy to write! :-), and to use a table size + larger than the actual number of entries. */ + +/* If an insertion makes the ratio of nonempty buckets to table size larger + than the growth threshold (a number between 0.0 and 1.0), then increase + the table size by multiplying by the growth factor (a number greater than + 1.0). The growth threshold defaults to 0.8, and the growth factor + defaults to 1.414, meaning that the table will have doubled its size + every second time 80% of the buckets get used. */ +#define DEFAULT_GROWTH_THRESHOLD 0.8 +#define DEFAULT_GROWTH_FACTOR 1.414 + +/* If a deletion empties a bucket and causes the ratio of used buckets to + table size to become smaller than the shrink threshold (a number between + 0.0 and 1.0), then shrink the table by multiplying by the shrink factor (a + number greater than the shrink threshold but smaller than 1.0). The shrink + threshold and factor default to 0.0 and 1.0, meaning that the table never + shrinks. */ +#define DEFAULT_SHRINK_THRESHOLD 0.0 +#define DEFAULT_SHRINK_FACTOR 1.0 + +/* Use this to initialize or reset a TUNING structure to + some sensible values. */ +static const Hash_tuning default_tuning = + { + DEFAULT_SHRINK_THRESHOLD, + DEFAULT_SHRINK_FACTOR, + DEFAULT_GROWTH_THRESHOLD, + DEFAULT_GROWTH_FACTOR, + false + }; + +/* Information and lookup. */ + +/* The following few functions provide information about the overall hash + table organization: the number of entries, number of buckets and maximum + length of buckets. */ + +/* Return the number of buckets in the hash table. The table size, the total + number of buckets (used plus unused), or the maximum number of slots, are + the same quantity. */ + +unsigned +hash_get_n_buckets (const Hash_table *table) +{ + return table->n_buckets; +} + +/* Return the number of slots in use (non-empty buckets). */ + +unsigned +hash_get_n_buckets_used (const Hash_table *table) +{ + return table->n_buckets_used; +} + +/* Return the number of active entries. */ + +unsigned +hash_get_n_entries (const Hash_table *table) +{ + return table->n_entries; +} + +/* Return the length of the longest chain (bucket). */ + +unsigned +hash_get_max_bucket_length (const Hash_table *table) +{ + struct hash_entry *bucket; + unsigned max_bucket_length = 0; + + for (bucket = table->bucket; bucket < table->bucket_limit; bucket++) + { + if (bucket->data) + { + struct hash_entry *cursor = bucket; + unsigned bucket_length = 1; + + while (cursor = cursor->next, cursor) + bucket_length++; + + if (bucket_length > max_bucket_length) + max_bucket_length = bucket_length; + } + } + + return max_bucket_length; +} + +/* Do a mild validation of a hash table, by traversing it and checking two + statistics. */ + +bool +hash_table_ok (const Hash_table *table) +{ + struct hash_entry *bucket; + unsigned n_buckets_used = 0; + unsigned n_entries = 0; + + for (bucket = table->bucket; bucket < table->bucket_limit; bucket++) + { + if (bucket->data) + { + struct hash_entry *cursor = bucket; + + /* Count bucket head. */ + n_buckets_used++; + n_entries++; + + /* Count bucket overflow. */ + while (cursor = cursor->next, cursor) + n_entries++; + } + } + + if (n_buckets_used == table->n_buckets_used && n_entries == table->n_entries) + return true; + + return false; +} + +void +hash_print_statistics (const Hash_table *table, FILE *stream) +{ + unsigned n_entries = hash_get_n_entries (table); + unsigned n_buckets = hash_get_n_buckets (table); + unsigned n_buckets_used = hash_get_n_buckets_used (table); + unsigned max_bucket_length = hash_get_max_bucket_length (table); + + fprintf (stream, "# entries: %u\n", n_entries); + fprintf (stream, "# buckets: %u\n", n_buckets); + fprintf (stream, "# buckets used: %u (%.2f%%)\n", n_buckets_used, + (100.0 * n_buckets_used) / n_buckets); + fprintf (stream, "max bucket length: %u\n", max_bucket_length); +} + +/* If ENTRY matches an entry already in the hash table, return the + entry from the table. Otherwise, return NULL. */ + +void * +hash_lookup (const Hash_table *table, const void *entry) +{ + struct hash_entry *bucket + = table->bucket + table->hasher (entry, table->n_buckets); + struct hash_entry *cursor; + + if (! (bucket < table->bucket_limit)) + abort (); + + if (bucket->data == NULL) + return NULL; + + for (cursor = bucket; cursor; cursor = cursor->next) + if (table->comparator (entry, cursor->data)) + return cursor->data; + + return NULL; +} + +/* Walking. */ + +/* The functions in this page traverse the hash table and process the + contained entries. For the traversal to work properly, the hash table + should not be resized nor modified while any particular entry is being + processed. In particular, entries should not be added or removed. */ + +/* Return the first data in the table, or NULL if the table is empty. */ + +void * +hash_get_first (const Hash_table *table) +{ + struct hash_entry *bucket; + + if (table->n_entries == 0) + return NULL; + + for (bucket = table->bucket; ; bucket++) + if (! (bucket < table->bucket_limit)) + abort (); + else if (bucket->data) + return bucket->data; +} + +/* Return the user data for the entry following ENTRY, where ENTRY has been + returned by a previous call to either `hash_get_first' or `hash_get_next'. + Return NULL if there are no more entries. */ + +void * +hash_get_next (const Hash_table *table, const void *entry) +{ + struct hash_entry *bucket + = table->bucket + table->hasher (entry, table->n_buckets); + struct hash_entry *cursor; + + if (! (bucket < table->bucket_limit)) + abort (); + + /* Find next entry in the same bucket. */ + for (cursor = bucket; cursor; cursor = cursor->next) + if (cursor->data == entry && cursor->next) + return cursor->next->data; + + /* Find first entry in any subsequent bucket. */ + while (++bucket < table->bucket_limit) + if (bucket->data) + return bucket->data; + + /* None found. */ + return NULL; +} + +/* Fill BUFFER with pointers to active user entries in the hash table, then + return the number of pointers copied. Do not copy more than BUFFER_SIZE + pointers. */ + +unsigned +hash_get_entries (const Hash_table *table, void **buffer, + unsigned buffer_size) +{ + unsigned counter = 0; + struct hash_entry *bucket; + struct hash_entry *cursor; + + for (bucket = table->bucket; bucket < table->bucket_limit; bucket++) + { + if (bucket->data) + { + for (cursor = bucket; cursor; cursor = cursor->next) + { + if (counter >= buffer_size) + return counter; + buffer[counter++] = cursor->data; + } + } + } + + return counter; +} + +/* Call a PROCESSOR function for each entry of a hash table, and return the + number of entries for which the processor function returned success. A + pointer to some PROCESSOR_DATA which will be made available to each call to + the processor function. The PROCESSOR accepts two arguments: the first is + the user entry being walked into, the second is the value of PROCESSOR_DATA + as received. The walking continue for as long as the PROCESSOR function + returns nonzero. When it returns zero, the walking is interrupted. */ + +unsigned +hash_do_for_each (const Hash_table *table, Hash_processor processor, + void *processor_data) +{ + unsigned counter = 0; + struct hash_entry *bucket; + struct hash_entry *cursor; + + for (bucket = table->bucket; bucket < table->bucket_limit; bucket++) + { + if (bucket->data) + { + for (cursor = bucket; cursor; cursor = cursor->next) + { + if (!(*processor) (cursor->data, processor_data)) + return counter; + counter++; + } + } + } + + return counter; +} + +/* Allocation and clean-up. */ + +/* Return a hash index for a NUL-terminated STRING between 0 and N_BUCKETS-1. + This is a convenience routine for constructing other hashing functions. */ + +#if USE_DIFF_HASH + +/* About hashings, Paul Eggert writes to me (FP), on 1994-01-01: "Please see + B. J. McKenzie, R. Harries & T. Bell, Selecting a hashing algorithm, + Software--practice & experience 20, 2 (Feb 1990), 209-224. Good hash + algorithms tend to be domain-specific, so what's good for [diffutils'] io.c + may not be good for your application." */ + +unsigned +hash_string (const char *string, unsigned n_buckets) +{ +# define ROTATE_LEFT(Value, Shift) \ + ((Value) << (Shift) | (Value) >> ((sizeof (unsigned) * CHAR_BIT) - (Shift))) +# define HASH_ONE_CHAR(Value, Byte) \ + ((Byte) + ROTATE_LEFT (Value, 7)) + + unsigned value = 0; + + for (; *string; string++) + value = HASH_ONE_CHAR (value, *(const unsigned char *) string); + return value % n_buckets; + +# undef ROTATE_LEFT +# undef HASH_ONE_CHAR +} + +#else /* not USE_DIFF_HASH */ + +/* This one comes from `recode', and performs a bit better than the above as + per a few experiments. It is inspired from a hashing routine found in the + very old Cyber `snoop', itself written in typical Greg Mansfield style. + (By the way, what happened to this excellent man? Is he still alive?) */ + +unsigned +hash_string (const char *string, unsigned n_buckets) +{ + unsigned value = 0; + + while (*string) + value = ((value * 31 + (int) *(const unsigned char *) string++) + % n_buckets); + return value; +} + +#endif /* not USE_DIFF_HASH */ + +/* Return true if CANDIDATE is a prime number. CANDIDATE should be an odd + number at least equal to 11. */ + +static bool +is_prime (unsigned long candidate) +{ + unsigned long divisor = 3; + unsigned long square = divisor * divisor; + + while (square < candidate && (candidate % divisor)) + { + divisor++; + square += 4 * divisor; + divisor++; + } + + return (candidate % divisor ? true : false); +} + +/* Round a given CANDIDATE number up to the nearest prime, and return that + prime. Primes lower than 10 are merely skipped. */ + +static unsigned long +next_prime (unsigned long candidate) +{ + /* Skip small primes. */ + if (candidate < 10) + candidate = 10; + + /* Make it definitely odd. */ + candidate |= 1; + + while (!is_prime (candidate)) + candidate += 2; + + return candidate; +} + +void +hash_reset_tuning (Hash_tuning *tuning) +{ + *tuning = default_tuning; +} + +/* For the given hash TABLE, check the user supplied tuning structure for + reasonable values, and return true if there is no gross error with it. + Otherwise, definitively reset the TUNING field to some acceptable default + in the hash table (that is, the user loses the right of further modifying + tuning arguments), and return false. */ + +static bool +check_tuning (Hash_table *table) +{ + const Hash_tuning *tuning = table->tuning; + + if (tuning->growth_threshold > 0.0 + && tuning->growth_threshold < 1.0 + && tuning->growth_factor > 1.0 + && tuning->shrink_threshold >= 0.0 + && tuning->shrink_threshold < 1.0 + && tuning->shrink_factor > tuning->shrink_threshold + && tuning->shrink_factor <= 1.0 + && tuning->shrink_threshold < tuning->growth_threshold) + return true; + + table->tuning = &default_tuning; + return false; +} + +/* Allocate and return a new hash table, or NULL upon failure. The initial + number of buckets is automatically selected so as to _guarantee_ that you + may insert at least CANDIDATE different user entries before any growth of + the hash table size occurs. So, if have a reasonably tight a-priori upper + bound on the number of entries you intend to insert in the hash table, you + may save some table memory and insertion time, by specifying it here. If + the IS_N_BUCKETS field of the TUNING structure is true, the CANDIDATE + argument has its meaning changed to the wanted number of buckets. + + TUNING points to a structure of user-supplied values, in case some fine + tuning is wanted over the default behavior of the hasher. If TUNING is + NULL, the default tuning parameters are used instead. + + The user-supplied HASHER function should be provided. It accepts two + arguments ENTRY and TABLE_SIZE. It computes, by hashing ENTRY contents, a + slot number for that entry which should be in the range 0..TABLE_SIZE-1. + This slot number is then returned. + + The user-supplied COMPARATOR function should be provided. It accepts two + arguments pointing to user data, it then returns true for a pair of entries + that compare equal, or false otherwise. This function is internally called + on entries which are already known to hash to the same bucket index. + + The user-supplied DATA_FREER function, when not NULL, may be later called + with the user data as an argument, just before the entry containing the + data gets freed. This happens from within `hash_free' or `hash_clear'. + You should specify this function only if you want these functions to free + all of your `data' data. This is typically the case when your data is + simply an auxiliary struct that you have malloc'd to aggregate several + values. */ + +Hash_table * +hash_initialize (unsigned candidate, const Hash_tuning *tuning, + Hash_hasher hasher, Hash_comparator comparator, + Hash_data_freer data_freer) +{ + Hash_table *table; + struct hash_entry *bucket; + + if (hasher == NULL || comparator == NULL) + return NULL; + + table = (Hash_table *) malloc (sizeof (Hash_table)); + if (table == NULL) + return NULL; + + if (!tuning) + tuning = &default_tuning; + table->tuning = tuning; + if (!check_tuning (table)) + { + /* Fail if the tuning options are invalid. This is the only occasion + when the user gets some feedback about it. Once the table is created, + if the user provides invalid tuning options, we silently revert to + using the defaults, and ignore further request to change the tuning + options. */ + free (table); + return NULL; + } + + table->n_buckets + = next_prime (tuning->is_n_buckets ? candidate + : (unsigned) (candidate / tuning->growth_threshold)); + + table->bucket = (struct hash_entry *) + malloc (table->n_buckets * sizeof (struct hash_entry)); + if (table->bucket == NULL) + { + free (table); + return NULL; + } + table->bucket_limit = table->bucket + table->n_buckets; + + for (bucket = table->bucket; bucket < table->bucket_limit; bucket++) + { + bucket->data = NULL; + bucket->next = NULL; + } + table->n_buckets_used = 0; + table->n_entries = 0; + + table->hasher = hasher; + table->comparator = comparator; + table->data_freer = data_freer; + + table->free_entry_list = NULL; +#if USE_OBSTACK + obstack_init (&table->entry_stack); +#endif + return table; +} + +/* Make all buckets empty, placing any chained entries on the free list. + Apply the user-specified function data_freer (if any) to the datas of any + affected entries. */ + +void +hash_clear (Hash_table *table) +{ + struct hash_entry *bucket; + + for (bucket = table->bucket; bucket < table->bucket_limit; bucket++) + { + if (bucket->data) + { + struct hash_entry *cursor; + struct hash_entry *next; + + /* Free the bucket overflow. */ + for (cursor = bucket->next; cursor; cursor = next) + { + if (table->data_freer) + (*table->data_freer) (cursor->data); + cursor->data = NULL; + + next = cursor->next; + /* Relinking is done one entry at a time, as it is to be expected + that overflows are either rare or short. */ + cursor->next = table->free_entry_list; + table->free_entry_list = cursor; + } + + /* Free the bucket head. */ + if (table->data_freer) + (*table->data_freer) (bucket->data); + bucket->data = NULL; + bucket->next = NULL; + } + } + + table->n_buckets_used = 0; + table->n_entries = 0; +} + +/* Reclaim all storage associated with a hash table. If a data_freer + function has been supplied by the user when the hash table was created, + this function applies it to the data of each entry before freeing that + entry. */ + +void +hash_free (Hash_table *table) +{ + struct hash_entry *bucket; + struct hash_entry *cursor; + struct hash_entry *next; + + /* Call the user data_freer function. */ + if (table->data_freer && table->n_entries) + { + for (bucket = table->bucket; bucket < table->bucket_limit; bucket++) + { + if (bucket->data) + { + for (cursor = bucket; cursor; cursor = cursor->next) + { + (*table->data_freer) (cursor->data); + } + } + } + } + +#if USE_OBSTACK + + obstack_free (&table->entry_stack, NULL); + +#else + + /* Free all bucket overflowed entries. */ + for (bucket = table->bucket; bucket < table->bucket_limit; bucket++) + { + for (cursor = bucket->next; cursor; cursor = next) + { + next = cursor->next; + free (cursor); + } + } + + /* Also reclaim the internal list of previously freed entries. */ + for (cursor = table->free_entry_list; cursor; cursor = next) + { + next = cursor->next; + free (cursor); + } + +#endif + + /* Free the remainder of the hash table structure. */ + free (table->bucket); + free (table); +} + +/* Insertion and deletion. */ + +/* Get a new hash entry for a bucket overflow, possibly by reclying a + previously freed one. If this is not possible, allocate a new one. */ + +static struct hash_entry * +allocate_entry (Hash_table *table) +{ + struct hash_entry *new; + + if (table->free_entry_list) + { + new = table->free_entry_list; + table->free_entry_list = new->next; + } + else + { +#if USE_OBSTACK + new = (struct hash_entry *) + obstack_alloc (&table->entry_stack, sizeof (struct hash_entry)); +#else + new = (struct hash_entry *) malloc (sizeof (struct hash_entry)); +#endif + } + + return new; +} + +/* Free a hash entry which was part of some bucket overflow, + saving it for later recycling. */ + +static void +free_entry (Hash_table *table, struct hash_entry *entry) +{ + entry->data = NULL; + entry->next = table->free_entry_list; + table->free_entry_list = entry; +} + +/* This private function is used to help with insertion and deletion. When + ENTRY matches an entry in the table, return a pointer to the corresponding + user data and set *BUCKET_HEAD to the head of the selected bucket. + Otherwise, return NULL. When DELETE is true and ENTRY matches an entry in + the table, unlink the matching entry. */ + +static void * +hash_find_entry (Hash_table *table, const void *entry, + struct hash_entry **bucket_head, bool delete) +{ + struct hash_entry *bucket + = table->bucket + table->hasher (entry, table->n_buckets); + struct hash_entry *cursor; + + if (! (bucket < table->bucket_limit)) + abort (); + + *bucket_head = bucket; + + /* Test for empty bucket. */ + if (bucket->data == NULL) + return NULL; + + /* See if the entry is the first in the bucket. */ + if ((*table->comparator) (entry, bucket->data)) + { + void *data = bucket->data; + + if (delete) + { + if (bucket->next) + { + struct hash_entry *next = bucket->next; + + /* Bump the first overflow entry into the bucket head, then save + the previous first overflow entry for later recycling. */ + *bucket = *next; + free_entry (table, next); + } + else + { + bucket->data = NULL; + } + } + + return data; + } + + /* Scan the bucket overflow. */ + for (cursor = bucket; cursor->next; cursor = cursor->next) + { + if ((*table->comparator) (entry, cursor->next->data)) + { + void *data = cursor->next->data; + + if (delete) + { + struct hash_entry *next = cursor->next; + + /* Unlink the entry to delete, then save the freed entry for later + recycling. */ + cursor->next = next->next; + free_entry (table, next); + } + + return data; + } + } + + /* No entry found. */ + return NULL; +} + +/* For an already existing hash table, change the number of buckets through + specifying CANDIDATE. The contents of the hash table are preserved. The + new number of buckets is automatically selected so as to _guarantee_ that + the table may receive at least CANDIDATE different user entries, including + those already in the table, before any other growth of the hash table size + occurs. If TUNING->IS_N_BUCKETS is true, then CANDIDATE specifies the + exact number of buckets desired. */ + +bool +hash_rehash (Hash_table *table, unsigned candidate) +{ + Hash_table *new_table; + struct hash_entry *bucket; + struct hash_entry *cursor; + struct hash_entry *next; + + new_table = hash_initialize (candidate, table->tuning, table->hasher, + table->comparator, table->data_freer); + if (new_table == NULL) + return false; + + /* Merely reuse the extra old space into the new table. */ +#if USE_OBSTACK + obstack_free (&new_table->entry_stack, NULL); + new_table->entry_stack = table->entry_stack; +#endif + new_table->free_entry_list = table->free_entry_list; + + for (bucket = table->bucket; bucket < table->bucket_limit; bucket++) + if (bucket->data) + for (cursor = bucket; cursor; cursor = next) + { + void *data = cursor->data; + struct hash_entry *new_bucket + = (new_table->bucket + + new_table->hasher (data, new_table->n_buckets)); + + if (! (new_bucket < new_table->bucket_limit)) + abort (); + + next = cursor->next; + + if (new_bucket->data) + { + if (cursor == bucket) + { + /* Allocate or recycle an entry, when moving from a bucket + header into a bucket overflow. */ + struct hash_entry *new_entry = allocate_entry (new_table); + + if (new_entry == NULL) + return false; + + new_entry->data = data; + new_entry->next = new_bucket->next; + new_bucket->next = new_entry; + } + else + { + /* Merely relink an existing entry, when moving from a + bucket overflow into a bucket overflow. */ + cursor->next = new_bucket->next; + new_bucket->next = cursor; + } + } + else + { + /* Free an existing entry, when moving from a bucket + overflow into a bucket header. Also take care of the + simple case of moving from a bucket header into a bucket + header. */ + new_bucket->data = data; + new_table->n_buckets_used++; + if (cursor != bucket) + free_entry (new_table, cursor); + } + } + + free (table->bucket); + table->bucket = new_table->bucket; + table->bucket_limit = new_table->bucket_limit; + table->n_buckets = new_table->n_buckets; + table->n_buckets_used = new_table->n_buckets_used; + table->free_entry_list = new_table->free_entry_list; + /* table->n_entries already holds its value. */ +#if USE_OBSTACK + table->entry_stack = new_table->entry_stack; +#endif + free (new_table); + + return true; +} + +/* If ENTRY matches an entry already in the hash table, return the pointer + to the entry from the table. Otherwise, insert ENTRY and return ENTRY. + Return NULL if the storage required for insertion cannot be allocated. */ + +void * +hash_insert (Hash_table *table, const void *entry) +{ + void *data; + struct hash_entry *bucket; + + /* The caller cannot insert a NULL entry. */ + if (! entry) + abort (); + + /* If there's a matching entry already in the table, return that. */ + if ((data = hash_find_entry (table, entry, &bucket, false)) != NULL) + return data; + + /* ENTRY is not matched, it should be inserted. */ + + if (bucket->data) + { + struct hash_entry *new_entry = allocate_entry (table); + + if (new_entry == NULL) + return NULL; + + /* Add ENTRY in the overflow of the bucket. */ + + new_entry->data = (void *) entry; + new_entry->next = bucket->next; + bucket->next = new_entry; + table->n_entries++; + return (void *) entry; + } + + /* Add ENTRY right in the bucket head. */ + + bucket->data = (void *) entry; + table->n_entries++; + table->n_buckets_used++; + + /* If the growth threshold of the buckets in use has been reached, increase + the table size and rehash. There's no point in checking the number of + entries: if the hashing function is ill-conditioned, rehashing is not + likely to improve it. */ + + if (table->n_buckets_used + > table->tuning->growth_threshold * table->n_buckets) + { + /* Check more fully, before starting real work. If tuning arguments + became invalid, the second check will rely on proper defaults. */ + check_tuning (table); + if (table->n_buckets_used + > table->tuning->growth_threshold * table->n_buckets) + { + const Hash_tuning *tuning = table->tuning; + unsigned candidate + = (unsigned) (tuning->is_n_buckets + ? (table->n_buckets * tuning->growth_factor) + : (table->n_buckets * tuning->growth_factor + * tuning->growth_threshold)); + + /* If the rehash fails, arrange to return NULL. */ + if (!hash_rehash (table, candidate)) + entry = NULL; + } + } + + return (void *) entry; +} + +/* If ENTRY is already in the table, remove it and return the just-deleted + data (the user may want to deallocate its storage). If ENTRY is not in the + table, don't modify the table and return NULL. */ + +void * +hash_delete (Hash_table *table, const void *entry) +{ + void *data; + struct hash_entry *bucket; + + data = hash_find_entry (table, entry, &bucket, true); + if (!data) + return NULL; + + table->n_entries--; + if (!bucket->data) + { + table->n_buckets_used--; + + /* If the shrink threshold of the buckets in use has been reached, + rehash into a smaller table. */ + + if (table->n_buckets_used + < table->tuning->shrink_threshold * table->n_buckets) + { + /* Check more fully, before starting real work. If tuning arguments + became invalid, the second check will rely on proper defaults. */ + check_tuning (table); + if (table->n_buckets_used + < table->tuning->shrink_threshold * table->n_buckets) + { + const Hash_tuning *tuning = table->tuning; + unsigned candidate + = (unsigned) (tuning->is_n_buckets + ? table->n_buckets * tuning->shrink_factor + : (table->n_buckets * tuning->shrink_factor + * tuning->growth_threshold)); + + hash_rehash (table, candidate); + } + } + } + + return data; +} + +/* Testing. */ + +#if TESTING + +void +hash_print (const Hash_table *table) +{ + struct hash_entry *bucket; + + for (bucket = table->bucket; bucket < table->bucket_limit; bucket++) + { + struct hash_entry *cursor; + + if (bucket) + printf ("%d:\n", bucket - table->bucket); + + for (cursor = bucket; cursor; cursor = cursor->next) + { + char *s = (char *) cursor->data; + /* FIXME */ + if (s) + printf (" %s\n", s); + } + } +} + +#endif /* TESTING */ Index: patch-2.5.9/hash.h =================================================================== --- patch-2.5.9.orig/hash.h 2003-10-07 14:53:07.000000000 +0200 +++ patch-2.5.9/hash.h 2003-10-07 14:53:07.000000000 +0200 @@ -0,0 +1,93 @@ +/* hash - hashing table processing. + Copyright (C) 1998, 1999, 2001 Free Software Foundation, Inc. + Written by Jim Meyering , 1998. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 2, or (at your option) + any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program; if not, write to the Free Software Foundation, + Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */ + +/* A generic hash table package. */ + +/* Make sure USE_OBSTACK is defined to 1 if you want the allocator to use + obstacks instead of malloc, and recompile `hash.c' with same setting. */ + +#ifndef HASH_H_ +# define HASH_H_ + +# ifndef PARAMS +# if PROTOTYPES || __STDC__ +# define PARAMS(Args) Args +# else +# define PARAMS(Args) () +# endif +# endif + +typedef unsigned (*Hash_hasher) PARAMS ((const void *, unsigned)); +typedef bool (*Hash_comparator) PARAMS ((const void *, const void *)); +typedef void (*Hash_data_freer) PARAMS ((void *)); +typedef bool (*Hash_processor) PARAMS ((void *, void *)); + +struct hash_entry + { + void *data; + struct hash_entry *next; + }; + +struct hash_tuning + { + /* This structure is mainly used for `hash_initialize', see the block + documentation of `hash_reset_tuning' for more complete comments. */ + + float shrink_threshold; /* ratio of used buckets to trigger a shrink */ + float shrink_factor; /* ratio of new smaller size to original size */ + float growth_threshold; /* ratio of used buckets to trigger a growth */ + float growth_factor; /* ratio of new bigger size to original size */ + bool is_n_buckets; /* if CANDIDATE really means table size */ + }; + +typedef struct hash_tuning Hash_tuning; + +struct hash_table; + +typedef struct hash_table Hash_table; + +/* Information and lookup. */ +unsigned hash_get_n_buckets PARAMS ((const Hash_table *)); +unsigned hash_get_n_buckets_used PARAMS ((const Hash_table *)); +unsigned hash_get_n_entries PARAMS ((const Hash_table *)); +unsigned hash_get_max_bucket_length PARAMS ((const Hash_table *)); +bool hash_table_ok PARAMS ((const Hash_table *)); +void hash_print_statistics PARAMS ((const Hash_table *, FILE *)); +void *hash_lookup PARAMS ((const Hash_table *, const void *)); + +/* Walking. */ +void *hash_get_first PARAMS ((const Hash_table *)); +void *hash_get_next PARAMS ((const Hash_table *, const void *)); +unsigned hash_get_entries PARAMS ((const Hash_table *, void **, unsigned)); +unsigned hash_do_for_each PARAMS ((const Hash_table *, Hash_processor, void *)); + +/* Allocation and clean-up. */ +unsigned hash_string PARAMS ((const char *, unsigned)); +void hash_reset_tuning PARAMS ((Hash_tuning *)); +Hash_table *hash_initialize PARAMS ((unsigned, const Hash_tuning *, + Hash_hasher, Hash_comparator, + Hash_data_freer)); +void hash_clear PARAMS ((Hash_table *)); +void hash_free PARAMS ((Hash_table *)); + +/* Insertion and deletion. */ +bool hash_rehash PARAMS ((Hash_table *, unsigned)); +void *hash_insert PARAMS ((Hash_table *, const void *)); +void *hash_delete PARAMS ((Hash_table *, const void *)); + +#endif Index: patch-2.5.9/m4/nanosecond_stat.m4 =================================================================== --- patch-2.5.9.orig/m4/nanosecond_stat.m4 2003-10-07 16:47:24.000000000 +0200 +++ patch-2.5.9/m4/nanosecond_stat.m4 2003-10-07 17:14:48.000000000 +0200 @@ -0,0 +1,35 @@ +AC_DEFUN([ag_CHECK_NANOSECOND_STAT], + [AC_CACHE_CHECK([for nanosecond timestamps in struct stat], + [ac_cv_stat_nsec], + [AC_TRY_COMPILE( + [ + #include + #include + #include + struct stat st; + ], + [ return st.st_atimensec + st.st_mtimensec + st.st_ctimensec; ], + [ac_cv_stat_nsec=yes], + [ac_cv_stat_nsec=no]) + ]) + if test $ac_cv_stat_nsec = yes; then + AC_DEFINE(HAVE_STAT_NSEC, 1, [Define to 1 if struct stat has nanosecond timestamps.]) + fi + + AC_CACHE_CHECK([for nanosecond timestamps in struct stat], + [ac_cv_stat_timeval], + [AC_TRY_COMPILE( + [ + #include + #include + #include + #include + struct stat st; + ], + [ return st.st_atim.tv_nsec + st.st_mtim.tv_nsec + st.st_ctim.tv_nsec; ], + [ac_cv_stat_timeval=yes], + [ac_cv_stat_timeval=no]) + ]) + if test $ac_cv_stat_timeval = yes; then + AC_DEFINE(HAVE_STAT_TIMEVAL, 1, [Define to 1 if struct stat comtains struct timeval's.]) + fi])