tree-diff.con commit tree-diff: reuse base str(buf) memory on sub-tree recursion (12cd817)
   1/*
   2 * Helper functions for tree diff generation
   3 */
   4#include "cache.h"
   5#include "diff.h"
   6#include "diffcore.h"
   7#include "tree.h"
   8
   9
  10static int ll_diff_tree_sha1(const unsigned char *old, const unsigned char *new,
  11                             struct strbuf *base, struct diff_options *opt);
  12
  13/*
  14 * Compare two tree entries, taking into account only path/S_ISDIR(mode),
  15 * but not their sha1's.
  16 *
  17 * NOTE files and directories *always* compare differently, even when having
  18 *      the same name - thanks to base_name_compare().
  19 *
  20 * NOTE empty (=invalid) descriptor(s) take part in comparison as +infty,
  21 *      so that they sort *after* valid tree entries.
  22 *
  23 *      Due to this convention, if trees are scanned in sorted order, all
  24 *      non-empty descriptors will be processed first.
  25 */
  26static int tree_entry_pathcmp(struct tree_desc *t1, struct tree_desc *t2)
  27{
  28        struct name_entry *e1, *e2;
  29        int cmp;
  30
  31        /* empty descriptors sort after valid tree entries */
  32        if (!t1->size)
  33                return t2->size ? 1 : 0;
  34        else if (!t2->size)
  35                return -1;
  36
  37        e1 = &t1->entry;
  38        e2 = &t2->entry;
  39        cmp = base_name_compare(e1->path, tree_entry_len(e1), e1->mode,
  40                                e2->path, tree_entry_len(e2), e2->mode);
  41        return cmp;
  42}
  43
  44
  45/* convert path, t1/t2 -> opt->diff_*() callbacks */
  46static void emit_diff(struct diff_options *opt, struct strbuf *path,
  47                      struct tree_desc *t1, struct tree_desc *t2)
  48{
  49        unsigned int mode1 = t1 ? t1->entry.mode : 0;
  50        unsigned int mode2 = t2 ? t2->entry.mode : 0;
  51
  52        if (mode1 && mode2) {
  53                opt->change(opt, mode1, mode2, t1->entry.sha1, t2->entry.sha1,
  54                        1, 1, path->buf, 0, 0);
  55        }
  56        else {
  57                const unsigned char *sha1;
  58                unsigned int mode;
  59                int addremove;
  60
  61                if (mode2) {
  62                        addremove = '+';
  63                        sha1 = t2->entry.sha1;
  64                        mode = mode2;
  65                } else {
  66                        addremove = '-';
  67                        sha1 = t1->entry.sha1;
  68                        mode = mode1;
  69                }
  70
  71                opt->add_remove(opt, addremove, mode, sha1, 1, path->buf, 0);
  72        }
  73}
  74
  75
  76/* new path should be added to diff
  77 *
  78 * 3 cases on how/when it should be called and behaves:
  79 *
  80 *      !t1,  t2        -> path added, parent lacks it
  81 *       t1, !t2        -> path removed from parent
  82 *       t1,  t2        -> path modified
  83 */
  84static void show_path(struct strbuf *base, struct diff_options *opt,
  85                      struct tree_desc *t1, struct tree_desc *t2)
  86{
  87        unsigned mode;
  88        const char *path;
  89        int pathlen;
  90        int old_baselen = base->len;
  91        int isdir, recurse = 0, emitthis = 1;
  92
  93        /* at least something has to be valid */
  94        assert(t1 || t2);
  95
  96        if (t2) {
  97                /* path present in resulting tree */
  98                tree_entry_extract(t2, &path, &mode);
  99                pathlen = tree_entry_len(&t2->entry);
 100                isdir = S_ISDIR(mode);
 101        } else {
 102                /*
 103                 * a path was removed - take path from parent. Also take
 104                 * mode from parent, to decide on recursion.
 105                 */
 106                tree_entry_extract(t1, &path, &mode);
 107                pathlen = tree_entry_len(&t1->entry);
 108
 109                isdir = S_ISDIR(mode);
 110                mode = 0;
 111        }
 112
 113        if (DIFF_OPT_TST(opt, RECURSIVE) && isdir) {
 114                recurse = 1;
 115                emitthis = DIFF_OPT_TST(opt, TREE_IN_RECURSIVE);
 116        }
 117
 118        strbuf_add(base, path, pathlen);
 119
 120        if (emitthis)
 121                emit_diff(opt, base, t1, t2);
 122
 123        if (recurse) {
 124                strbuf_addch(base, '/');
 125                ll_diff_tree_sha1(t1 ? t1->entry.sha1 : NULL,
 126                                  t2 ? t2->entry.sha1 : NULL, base, opt);
 127        }
 128
 129        strbuf_setlen(base, old_baselen);
 130}
 131
 132static void skip_uninteresting(struct tree_desc *t, struct strbuf *base,
 133                               struct diff_options *opt)
 134{
 135        enum interesting match;
 136
 137        while (t->size) {
 138                match = tree_entry_interesting(&t->entry, base, 0, &opt->pathspec);
 139                if (match) {
 140                        if (match == all_entries_not_interesting)
 141                                t->size = 0;
 142                        break;
 143                }
 144                update_tree_entry(t);
 145        }
 146}
 147
 148static int ll_diff_tree_sha1(const unsigned char *old, const unsigned char *new,
 149                             struct strbuf *base, struct diff_options *opt)
 150{
 151        struct tree_desc t1, t2;
 152        void *t1tree, *t2tree;
 153
 154        t1tree = fill_tree_descriptor(&t1, old);
 155        t2tree = fill_tree_descriptor(&t2, new);
 156
 157        /* Enable recursion indefinitely */
 158        opt->pathspec.recursive = DIFF_OPT_TST(opt, RECURSIVE);
 159
 160        for (;;) {
 161                int cmp;
 162
 163                if (diff_can_quit_early(opt))
 164                        break;
 165                if (opt->pathspec.nr) {
 166                        skip_uninteresting(&t1, base, opt);
 167                        skip_uninteresting(&t2, base, opt);
 168                }
 169                if (!t1.size && !t2.size)
 170                        break;
 171
 172                cmp = tree_entry_pathcmp(&t1, &t2);
 173
 174                /* t1 = t2 */
 175                if (cmp == 0) {
 176                        if (DIFF_OPT_TST(opt, FIND_COPIES_HARDER) ||
 177                            hashcmp(t1.entry.sha1, t2.entry.sha1) ||
 178                            (t1.entry.mode != t2.entry.mode))
 179                                show_path(base, opt, &t1, &t2);
 180
 181                        update_tree_entry(&t1);
 182                        update_tree_entry(&t2);
 183                }
 184
 185                /* t1 < t2 */
 186                else if (cmp < 0) {
 187                        show_path(base, opt, &t1, /*t2=*/NULL);
 188                        update_tree_entry(&t1);
 189                }
 190
 191                /* t1 > t2 */
 192                else {
 193                        show_path(base, opt, /*t1=*/NULL, &t2);
 194                        update_tree_entry(&t2);
 195                }
 196        }
 197
 198        free(t2tree);
 199        free(t1tree);
 200        return 0;
 201}
 202
 203/*
 204 * Does it look like the resulting diff might be due to a rename?
 205 *  - single entry
 206 *  - not a valid previous file
 207 */
 208static inline int diff_might_be_rename(void)
 209{
 210        return diff_queued_diff.nr == 1 &&
 211                !DIFF_FILE_VALID(diff_queued_diff.queue[0]->one);
 212}
 213
 214static void try_to_follow_renames(const unsigned char *old, const unsigned char *new, struct strbuf *base, struct diff_options *opt)
 215{
 216        struct diff_options diff_opts;
 217        struct diff_queue_struct *q = &diff_queued_diff;
 218        struct diff_filepair *choice;
 219        int i;
 220
 221        /*
 222         * follow-rename code is very specific, we need exactly one
 223         * path. Magic that matches more than one path is not
 224         * supported.
 225         */
 226        GUARD_PATHSPEC(&opt->pathspec, PATHSPEC_FROMTOP | PATHSPEC_LITERAL);
 227#if 0
 228        /*
 229         * We should reject wildcards as well. Unfortunately we
 230         * haven't got a reliable way to detect that 'foo\*bar' in
 231         * fact has no wildcards. nowildcard_len is merely a hint for
 232         * optimization. Let it slip for now until wildmatch is taught
 233         * about dry-run mode and returns wildcard info.
 234         */
 235        if (opt->pathspec.has_wildcard)
 236                die("BUG:%s:%d: wildcards are not supported",
 237                    __FILE__, __LINE__);
 238#endif
 239
 240        /* Remove the file creation entry from the diff queue, and remember it */
 241        choice = q->queue[0];
 242        q->nr = 0;
 243
 244        diff_setup(&diff_opts);
 245        DIFF_OPT_SET(&diff_opts, RECURSIVE);
 246        DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
 247        diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 248        diff_opts.single_follow = opt->pathspec.items[0].match;
 249        diff_opts.break_opt = opt->break_opt;
 250        diff_opts.rename_score = opt->rename_score;
 251        diff_setup_done(&diff_opts);
 252        ll_diff_tree_sha1(old, new, base, &diff_opts);
 253        diffcore_std(&diff_opts);
 254        free_pathspec(&diff_opts.pathspec);
 255
 256        /* Go through the new set of filepairing, and see if we find a more interesting one */
 257        opt->found_follow = 0;
 258        for (i = 0; i < q->nr; i++) {
 259                struct diff_filepair *p = q->queue[i];
 260
 261                /*
 262                 * Found a source? Not only do we use that for the new
 263                 * diff_queued_diff, we will also use that as the path in
 264                 * the future!
 265                 */
 266                if ((p->status == 'R' || p->status == 'C') &&
 267                    !strcmp(p->two->path, opt->pathspec.items[0].match)) {
 268                        const char *path[2];
 269
 270                        /* Switch the file-pairs around */
 271                        q->queue[i] = choice;
 272                        choice = p;
 273
 274                        /* Update the path we use from now on.. */
 275                        path[0] = p->one->path;
 276                        path[1] = NULL;
 277                        free_pathspec(&opt->pathspec);
 278                        parse_pathspec(&opt->pathspec,
 279                                       PATHSPEC_ALL_MAGIC & ~PATHSPEC_LITERAL,
 280                                       PATHSPEC_LITERAL_PATH, "", path);
 281
 282                        /*
 283                         * The caller expects us to return a set of vanilla
 284                         * filepairs to let a later call to diffcore_std()
 285                         * it makes to sort the renames out (among other
 286                         * things), but we already have found renames
 287                         * ourselves; signal diffcore_std() not to muck with
 288                         * rename information.
 289                         */
 290                        opt->found_follow = 1;
 291                        break;
 292                }
 293        }
 294
 295        /*
 296         * Then, discard all the non-relevant file pairs...
 297         */
 298        for (i = 0; i < q->nr; i++) {
 299                struct diff_filepair *p = q->queue[i];
 300                diff_free_filepair(p);
 301        }
 302
 303        /*
 304         * .. and re-instate the one we want (which might be either the
 305         * original one, or the rename/copy we found)
 306         */
 307        q->queue[0] = choice;
 308        q->nr = 1;
 309}
 310
 311int diff_tree_sha1(const unsigned char *old, const unsigned char *new, const char *base_str, struct diff_options *opt)
 312{
 313        struct strbuf base;
 314        int retval;
 315
 316        strbuf_init(&base, PATH_MAX);
 317        strbuf_addstr(&base, base_str);
 318
 319        retval = ll_diff_tree_sha1(old, new, &base, opt);
 320        if (!*base_str && DIFF_OPT_TST(opt, FOLLOW_RENAMES) && diff_might_be_rename())
 321                try_to_follow_renames(old, new, &base, opt);
 322
 323        strbuf_release(&base);
 324
 325        return retval;
 326}
 327
 328int diff_root_tree_sha1(const unsigned char *new, const char *base, struct diff_options *opt)
 329{
 330        return diff_tree_sha1(NULL, new, base, opt);
 331}