1 /* Generic infrastructure to implement various diff algorithms. */
2 /*
3  * Copyright (c) 2020 Neels Hofmeyr <neels@hofmeyr.de>
4  *
5  * Permission to use, copy, modify, and distribute this software for any
6  * purpose with or without fee is hereby granted, provided that the above
7  * copyright notice and this permission notice appear in all copies.
8  *
9  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
10  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
11  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
12  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
13  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
14  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
15  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
16  */
17 
18 struct diff_range {
19 	int start;
20 	int end;
21 };
22 
23 /* List of all possible return codes of a diff invocation. */
24 #define DIFF_RC_USE_DIFF_ALGO_FALLBACK	-1
25 #define DIFF_RC_OK			0
26 /* Any positive return values are errno values from sys/errno.h */
27 
28 struct diff_atom;
29 
30 /* For each file, there is a "root" struct diff_data referencing the entire
31  * file, which the atoms are parsed from. In recursion of diff algorithm, there
32  * may be "child" struct diff_data only referencing a subsection of the file,
33  * re-using the atoms parsing. For "root" structs, atoms_allocated will be
34  * nonzero, indicating that the array of atoms is owned by that struct. For
35  * "child" structs, atoms_allocated == 0, to indicate that the struct is
36  * referencing a subset of atoms. */
37 struct diff_data {
38 	FILE *f;		/* if root diff_data and not memory-mapped */
39 	off_t pos;		/* if not memory-mapped */
40 	const uint8_t *data;	/* if memory-mapped */
41 	off_t len;
42 
43 	ARRAYLIST(struct diff_atom) atoms;
44 	struct diff_data *root;
45 
46 	int diff_flags;
47 
48 	int err;
49 };
50 
51 #define DIFF_FLAG_IGNORE_WHITESPACE	0x00000001
52 
53 void diff_data_free(struct diff_data *diff_data);
54 
55 struct diff_chunk;
56 typedef ARRAYLIST(struct diff_chunk) diff_chunk_arraylist_t;
57 
58 struct diff_result {
59 	int rc;
60 	struct diff_data left;
61 	struct diff_data right;
62 	diff_chunk_arraylist_t chunks;
63 };
64 
65 struct diff_state;
66 
67 /* Signature of a utility function to divide both source files into diff atoms.
68  * It is possible that a (future) algorithm requires both source files to decide
69  * on atom split points, hence this gets both left and right to atomize at the
70  * same time.
71  * An example is diff_atomize_text_by_line() in diff_atomize_text.c.
72  *
73  * func_data: context pointer (free to be used by implementation).
74  * left: struct diff_data with left->data and left->len already set up, and
75  * left->atoms to be created.
76  * right: struct diff_data with right->data and right->len already set up, and
77  * right->atoms to be created.
78  */
79 typedef int (*diff_atomize_func_t)(void *func_data,
80 					    struct diff_data *left,
81 					    struct diff_data *right);
82 
83 extern int diff_atomize_text_by_line(void *func_data,
84 					      struct diff_data *left,
85 					      struct diff_data *right);
86 
87 struct diff_algo_config;
88 typedef int (*diff_algo_impl_t)(
89 	const struct diff_algo_config *algo_config, struct diff_state *state);
90 
91 /* Form a result with all left-side removed and all right-side added, i.e. no
92  * actual diff algorithm involved. */
93 int diff_algo_none(const struct diff_algo_config *algo_config,
94 	struct diff_state *state);
95 
96 /* Myers Diff tracing from the start all the way through to the end, requiring
97  * quadratic amounts of memory. This can fail if the required space surpasses
98  * algo_config->permitted_state_size. */
99 extern int diff_algo_myers(const struct diff_algo_config *algo_config,
100 	struct diff_state *state);
101 
102 /* Myers "Divide et Impera": tracing forwards from the start and backwards from
103  * the end to find a midpoint that divides the problem into smaller chunks.
104  * Requires only linear amounts of memory. */
105 extern int diff_algo_myers_divide(
106 	const struct diff_algo_config *algo_config, struct diff_state *state);
107 
108 /* Patience Diff algorithm, which divides a larger diff into smaller chunks. For
109  * very specific scenarios, it may lead to a complete diff result by itself, but
110  * needs a fallback algo to solve chunks that don't have common-unique atoms. */
111 extern int diff_algo_patience(
112 	const struct diff_algo_config *algo_config, struct diff_state *state);
113 
114 /* Diff algorithms to use, possibly nested. For example:
115  *
116  * struct diff_algo_config myers, patience, myers_divide;
117  *
118  * myers = (struct diff_algo_config){
119  *         .impl = diff_algo_myers,
120  *         .permitted_state_size = 32 * 1024 * 1024,
121  *         // When too large, do diff_algo_patience:
122  *         .fallback_algo = &patience,
123  * };
124  *
125  * const struct diff_algo_config patience = (struct diff_algo_config){
126  * 	.impl = diff_algo_patience,
127  * 	// After subdivision, do Patience again:
128  * 	.inner_algo = &patience,
129  * 	// If subdivision failed, do Myers Divide et Impera:
130  * 	.fallback_algo = &myers_then_myers_divide,
131  * };
132  * 
133  * const struct diff_algo_config myers_divide = (struct diff_algo_config){
134  * 	.impl = diff_algo_myers_divide,
135  * 	// When division succeeded, start from the top:
136  * 	.inner_algo = &myers_then_myers_divide,
137  * 	// (fallback_algo = NULL implies diff_algo_none).
138  * };
139  * struct diff_config config = {
140  *         .algo = &myers,
141  *         ...
142  * };
143  * diff_main(&config, ...);
144  */
145 struct diff_algo_config {
146 	diff_algo_impl_t impl;
147 
148 	/* Fail this algo if it would use more than this amount of memory, and
149 	 * instead use fallback_algo (diff_algo_myers). permitted_state_size ==
150 	 * 0 means no limitation. */
151 	size_t permitted_state_size;
152 
153 	/* For algorithms that divide into smaller chunks, use this algorithm to
154 	 * solve the divided chunks. */
155 	const struct diff_algo_config *inner_algo;
156 
157 	/* If the algorithm fails (e.g. diff_algo_myers_if_small needs too large
158 	 * state, or diff_algo_patience can't find any common-unique atoms),
159 	 * then use this algorithm instead. */
160 	const struct diff_algo_config *fallback_algo;
161 };
162 
163 struct diff_config {
164 	diff_atomize_func_t atomize_func;
165 	void *atomize_func_data;
166 
167 	const struct diff_algo_config *algo;
168 
169 	/* How deep to step into subdivisions of a source file, a paranoia /
170 	 * safety measure to guard against infinite loops through diff
171 	 * algorithms. When the maximum recursion is reached, employ
172 	 * diff_algo_none (i.e. remove all left atoms and add all right atoms).
173 	 */
174 	unsigned int max_recursion_depth;
175 };
176 
177 struct diff_result *diff_main(const struct diff_config *config,
178 			      FILE *left_f, const uint8_t *left_data,
179 			      off_t left_len,
180 			      FILE *right_f, const uint8_t *right_data,
181 			      off_t right_len, int diff_flags);
182 void diff_result_free(struct diff_result *result);