Merge branch 'ml/cvs' into next
[gitweb.git] / builtin-grep.c
index 4be1514a4e5ba386ddd0d9c37bbe6e60000504dd..a762c48a5a3de70c47814df830aae0760549d163 100644 (file)
@@ -26,7 +26,7 @@ static int pathspec_matches(const char **paths, const char *name)
        for (i = 0; paths[i]; i++) {
                const char *match = paths[i];
                int matchlen = strlen(match);
-               const char *slash, *cp;
+               const char *cp, *meta;
 
                if ((matchlen <= namelen) &&
                    !strncmp(name, match, matchlen) &&
@@ -38,54 +38,109 @@ static int pathspec_matches(const char **paths, const char *name)
                if (name[namelen-1] != '/')
                        continue;
 
-               /* We are being asked if the name directory is worth
+               /* We are being asked if the directory ("name") is worth
                 * descending into.
                 *
                 * Find the longest leading directory name that does
                 * not have metacharacter in the pathspec; the name
                 * we are looking at must overlap with that directory.
                 */
-               for (cp = match, slash = NULL; cp - match < matchlen; cp++) {
+               for (cp = match, meta = NULL; cp - match < matchlen; cp++) {
                        char ch = *cp;
-                       if (ch == '/')
-                               slash = cp;
-                       if (ch == '*' || ch == '[')
+                       if (ch == '*' || ch == '[' || ch == '?') {
+                               meta = cp;
                                break;
+                       }
                }
-               if (!slash)
-                       slash = match; /* toplevel */
-               else
-                       slash++;
-               if (namelen <= slash - match) {
+               if (!meta)
+                       meta = cp; /* fully literal */
+
+               if (namelen <= meta - match) {
                        /* Looking at "Documentation/" and
                         * the pattern says "Documentation/howto/", or
-                        * "Documentation/diff*.txt".
+                        * "Documentation/diff*.txt".  The name we
+                        * have should match prefix.
                         */
                        if (!memcmp(match, name, namelen))
                                return 1;
+                       continue;
                }
-               else {
+
+               if (meta - match < namelen) {
                        /* Looking at "Documentation/howto/" and
-                        * the pattern says "Documentation/h*".
+                        * the pattern says "Documentation/h*";
+                        * match up to "Do.../h"; this avoids descending
+                        * into "Documentation/technical/".
                         */
-                       if (!memcmp(match, name, slash - match))
+                       if (!memcmp(match, name, meta - match))
                                return 1;
+                       continue;
                }
        }
        return 0;
 }
 
-struct grep_opt {
+struct grep_pat {
+       struct grep_pat *next;
+       const char *origin;
+       int no;
        const char *pattern;
        regex_t regexp;
+};
+
+struct grep_opt {
+       struct grep_pat *pattern_list;
+       struct grep_pat **pattern_tail;
+       regex_t regexp;
        unsigned linenum:1;
        unsigned invert:1;
        unsigned name_only:1;
+       unsigned unmatch_name_only:1;
+       unsigned count:1;
+       unsigned word_regexp:1;
+#define GREP_BINARY_DEFAULT    0
+#define GREP_BINARY_NOMATCH    1
+#define GREP_BINARY_TEXT       2
+       unsigned binary:2;
        int regflags;
        unsigned pre_context;
        unsigned post_context;
 };
 
+static void add_pattern(struct grep_opt *opt, const char *pat,
+                       const char *origin, int no)
+{
+       struct grep_pat *p = xcalloc(1, sizeof(*p));
+       p->pattern = pat;
+       p->origin = origin;
+       p->no = no;
+       *opt->pattern_tail = p;
+       opt->pattern_tail = &p->next;
+       p->next = NULL;
+}
+
+static void compile_patterns(struct grep_opt *opt)
+{
+       struct grep_pat *p;
+       for (p = opt->pattern_list; p; p = p->next) {
+               int err = regcomp(&p->regexp, p->pattern, opt->regflags);
+               if (err) {
+                       char errbuf[1024];
+                       char where[1024];
+                       if (p->no)
+                               sprintf(where, "In '%s' at %d, ",
+                                       p->origin, p->no);
+                       else if (p->origin)
+                               sprintf(where, "%s, ", p->origin);
+                       else
+                               where[0] = 0;
+                       regerror(err, &p->regexp, errbuf, 1024);
+                       regfree(&p->regexp);
+                       die("%s'%s': %s", where, p->pattern, errbuf);
+               }
+       }
+}
+
 static char *end_of_line(char *cp, unsigned long *left)
 {
        unsigned long l = *left;
@@ -97,13 +152,31 @@ static char *end_of_line(char *cp, unsigned long *left)
        return cp;
 }
 
+static int word_char(char ch)
+{
+       return isalnum(ch) || ch == '_';
+}
+
 static void show_line(struct grep_opt *opt, const char *bol, const char *eol,
                      const char *name, unsigned lno, char sign)
 {
        printf("%s%c", name, sign);
        if (opt->linenum)
                printf("%d%c", lno, sign);
-       printf("%.*s\n", eol-bol, bol);
+       printf("%.*s\n", (int)(eol-bol), bol);
+}
+
+/*
+ * NEEDSWORK: share code with diff.c
+ */
+#define FIRST_FEW_BYTES 8000
+static int buffer_is_binary(const char *ptr, unsigned long size)
+{
+       if (FIRST_FEW_BYTES < size)
+               size = FIRST_FEW_BYTES;
+       if (memchr(ptr, 0, size))
+               return 1;
+       return 0;
 }
 
 static int grep_buffer(struct grep_opt *opt, const char *name,
@@ -118,7 +191,22 @@ static int grep_buffer(struct grep_opt *opt, const char *name,
        } *prev = NULL, *pcl;
        unsigned last_hit = 0;
        unsigned last_shown = 0;
+       int binary_match_only = 0;
        const char *hunk_mark = "";
+       unsigned count = 0;
+
+       if (buffer_is_binary(buf, size)) {
+               switch (opt->binary) {
+               case GREP_BINARY_DEFAULT:
+                       binary_match_only = 1;
+                       break;
+               case GREP_BINARY_NOMATCH:
+                       return 0; /* Assume unmatch */
+                       break;
+               default:
+                       break;
+               }
+       }
 
        if (opt->pre_context)
                prev = xcalloc(opt->pre_context, sizeof(*prev));
@@ -128,23 +216,65 @@ static int grep_buffer(struct grep_opt *opt, const char *name,
        while (left) {
                regmatch_t pmatch[10];
                char *eol, ch;
-               int hit;
+               int hit = 0;
+               struct grep_pat *p;
 
                eol = end_of_line(bol, &left);
                ch = *eol;
                *eol = 0;
 
-               hit = !regexec(&opt->regexp, bol, ARRAY_SIZE(pmatch),
-                              pmatch, 0);
+               for (p = opt->pattern_list; p; p = p->next) {
+                       regex_t *exp = &p->regexp;
+                       hit = !regexec(exp, bol, ARRAY_SIZE(pmatch),
+                                      pmatch, 0);
+
+                       if (hit && opt->word_regexp) {
+                               /* Match beginning must be either
+                                * beginning of the line, or at word
+                                * boundary (i.e. the last char must
+                                * not be alnum or underscore).
+                                */
+                               if ((pmatch[0].rm_so < 0) ||
+                                   (eol - bol) <= pmatch[0].rm_so ||
+                                   (pmatch[0].rm_eo < 0) ||
+                                   (eol - bol) < pmatch[0].rm_eo)
+                                       die("regexp returned nonsense");
+                               if (pmatch[0].rm_so != 0 &&
+                                   word_char(bol[pmatch[0].rm_so-1]))
+                                       continue; /* not a word boundary */
+                               if ((eol-bol) < pmatch[0].rm_eo &&
+                                   word_char(bol[pmatch[0].rm_eo]))
+                                       continue; /* not a word boundary */
+                       }
+                       if (hit)
+                               break;
+               }
+               /* "grep -v -e foo -e bla" should list lines
+                * that do not have either, so inversion should
+                * be done outside.
+                */
                if (opt->invert)
                        hit = !hit;
+               if (opt->unmatch_name_only) {
+                       if (hit)
+                               return 0;
+                       goto next_line;
+               }
                if (hit) {
+                       count++;
+                       if (binary_match_only) {
+                               printf("Binary file %s matches\n", name);
+                               return 1;
+                       }
                        if (opt->name_only) {
                                printf("%s\n", name);
                                return 1;
                        }
                        /* Hit at this line.  If we haven't shown the
                         * pre-context lines, we would need to show them.
+                        * When asked to do "count", this still show
+                        * the context which is nonsense, but the user
+                        * deserves to get that ;-).
                         */
                        if (opt->pre_context) {
                                unsigned from;
@@ -166,7 +296,8 @@ static int grep_buffer(struct grep_opt *opt, const char *name,
                        }
                        if (last_shown && lno != last_shown + 1)
                                printf(hunk_mark);
-                       show_line(opt, bol, eol, name, lno, ':');
+                       if (!opt->count)
+                               show_line(opt, bol, eol, name, lno, ':');
                        last_shown = last_hit = lno;
                }
                else if (last_hit &&
@@ -185,11 +316,29 @@ static int grep_buffer(struct grep_opt *opt, const char *name,
                        prev->bol = bol;
                        prev->eol = eol;
                }
+
+       next_line:
                *eol = ch;
                bol = eol + 1;
+               if (!left)
+                       break;
                left--;
                lno++;
        }
+
+       if (opt->unmatch_name_only) {
+               /* We did not see any hit, so we want to show this */
+               printf("%s\n", name);
+               return 1;
+       }
+
+       /* NEEDSWORK:
+        * The real "grep -c foo *.c" gives many "bar.c:0" lines,
+        * which feels mostly useless but sometimes useful.  Maybe
+        * make it another option?  For now suppress them.
+        */
+       if (opt->count && count)
+               printf("%s:%u\n", name, count);
        return !!last_hit;
 }
 
@@ -344,10 +493,8 @@ static const char builtin_grep_usage[] =
 
 int cmd_grep(int argc, const char **argv, char **envp)
 {
-       int err;
        int hit = 0;
        int no_more_flags = 0;
-       int seen_noncommit = 0;
        int cached = 0;
        struct grep_opt opt;
        struct object_list *list, **tail, *object_list = NULL;
@@ -355,6 +502,7 @@ int cmd_grep(int argc, const char **argv, char **envp)
        const char **paths = NULL;
 
        memset(&opt, 0, sizeof(opt));
+       opt.pattern_tail = &opt.pattern_list;
        opt.regflags = REG_NEWLINE;
 
        /*
@@ -367,11 +515,20 @@ int cmd_grep(int argc, const char **argv, char **envp)
                        cached = 1;
                        continue;
                }
+               if (!strcmp("-a", arg) ||
+                   !strcmp("--text", arg)) {
+                       opt.binary = GREP_BINARY_TEXT;
+                       continue;
+               }
                if (!strcmp("-i", arg) ||
                    !strcmp("--ignore-case", arg)) {
                        opt.regflags |= REG_ICASE;
                        continue;
                }
+               if (!strcmp("-I", arg)) {
+                       opt.binary = GREP_BINARY_NOMATCH;
+                       continue;
+               }
                if (!strcmp("-v", arg) ||
                    !strcmp("--invert-match", arg)) {
                        opt.invert = 1;
@@ -402,18 +559,49 @@ int cmd_grep(int argc, const char **argv, char **envp)
                        opt.name_only = 1;
                        continue;
                }
-               if (!strcmp("-A", arg) ||
-                   !strcmp("-B", arg) ||
-                   !strcmp("-C", arg)) {
+               if (!strcmp("-L", arg) ||
+                   !strcmp("--files-without-match", arg)) {
+                       opt.unmatch_name_only = 1;
+                       continue;
+               }
+               if (!strcmp("-c", arg) ||
+                   !strcmp("--count", arg)) {
+                       opt.count = 1;
+                       continue;
+               }
+               if (!strcmp("-w", arg) ||
+                   !strcmp("--word-regexp", arg)) {
+                       opt.word_regexp = 1;
+                       continue;
+               }
+               if (!strncmp("-A", arg, 2) ||
+                   !strncmp("-B", arg, 2) ||
+                   !strncmp("-C", arg, 2) ||
+                   (arg[0] == '-' && '1' <= arg[1] && arg[1] <= '9')) {
                        unsigned num;
-                       if (argc <= 1 ||
-                           sscanf(*++argv, "%u", &num) != 1)
+                       const char *scan;
+                       switch (arg[1]) {
+                       case 'A': case 'B': case 'C':
+                               if (!arg[2]) {
+                                       if (argc <= 1)
+                                               usage(builtin_grep_usage);
+                                       scan = *++argv;
+                                       argc--;
+                               }
+                               else
+                                       scan = arg + 2;
+                               break;
+                       default:
+                               scan = arg + 1;
+                               break;
+                       }
+                       if (sscanf(scan, "%u", &num) != 1)
                                usage(builtin_grep_usage);
-                       argc--;
                        switch (arg[1]) {
                        case 'A':
                                opt.post_context = num;
                                break;
+                       default:
                        case 'C':
                                opt.post_context = num;
                        case 'B':
@@ -422,14 +610,33 @@ int cmd_grep(int argc, const char **argv, char **envp)
                        }
                        continue;
                }
+               if (!strcmp("-f", arg)) {
+                       FILE *patterns;
+                       int lno = 0;
+                       char buf[1024];
+                       if (argc <= 1)
+                               usage(builtin_grep_usage);
+                       patterns = fopen(argv[1], "r");
+                       if (!patterns)
+                               die("'%s': %s", strerror(errno));
+                       while (fgets(buf, sizeof(buf), patterns)) {
+                               int len = strlen(buf);
+                               if (buf[len-1] == '\n')
+                                       buf[len-1] = 0;
+                               /* ignore empty line like grep does */
+                               if (!buf[0])
+                                       continue;
+                               add_pattern(&opt, strdup(buf), argv[1], ++lno);
+                       }
+                       fclose(patterns);
+                       argv++;
+                       argc--;
+                       continue;
+               }
                if (!strcmp("-e", arg)) {
                        if (1 < argc) {
-                               /* We probably would want to do
-                                * -e pat1 -e pat2 as well later...
-                                */
-                               if (opt.pattern)
-                                       die("more than one pattern?");
-                               opt.pattern = *++argv;
+                               add_pattern(&opt, argv[1], "-e option", 0);
+                               argv++;
                                argc--;
                                continue;
                        }
@@ -442,8 +649,8 @@ int cmd_grep(int argc, const char **argv, char **envp)
                /* Either unrecognized option or a single pattern */
                if (!no_more_flags && *arg == '-')
                        usage(builtin_grep_usage);
-               if (!opt.pattern) {
-                       opt.pattern = arg;
+               if (!opt.pattern_list) {
+                       add_pattern(&opt, arg, "command line", 0);
                        break;
                }
                else {
@@ -455,15 +662,9 @@ int cmd_grep(int argc, const char **argv, char **envp)
                        break;
                }
        }
-       if (!opt.pattern)
+       if (!opt.pattern_list)
                die("no pattern given.");
-       err = regcomp(&opt.regexp, opt.pattern, opt.regflags);
-       if (err) {
-               char errbuf[1024];
-               regerror(err, &opt.regexp, errbuf, 1024);
-               regfree(&opt.regexp);
-               die("'%s': %s", opt.pattern, errbuf);
-       }
+       compile_patterns(&opt);
        tail = &object_list;
        while (1 < argc) {
                struct object *object;
@@ -490,21 +691,9 @@ int cmd_grep(int argc, const char **argv, char **envp)
 
        if (!object_list)
                return !grep_cache(&opt, paths, cached);
-       /*
-        * Do not walk "grep -e foo master next pu -- Documentation/"
-        * but do walk "grep -e foo master..next -- Documentation/".
-        * Ranged request mixed with a blob or tree object, like
-        * "grep -e foo v1.0.0:Documentation/ master..next"
-        * so detect that and complain.
-        */
-       for (list = object_list; list; list = list->next) {
-               struct object *real_obj;
-               real_obj = deref_tag(list->item, NULL, 0);
-               if (strcmp(real_obj->type, commit_type))
-                       seen_noncommit = 1;
-       }
+
        if (cached)
-               die("both --cached and revisions given.");
+               die("both --cached and trees are given.");
 
        for (list = object_list; list; list = list->next) {
                struct object *real_obj;