true built-in diff: run everything in-core.
[git.git] / diff.c
diff --git a/diff.c b/diff.c
index 2e0797b..562ca01 100644 (file)
--- a/diff.c
+++ b/diff.c
@@ -8,8 +8,7 @@
 #include "quote.h"
 #include "diff.h"
 #include "diffcore.h"
-
-static const char *diff_opts = "-pu";
+#include "xdiff/xdiff.h"
 
 static int use_size_cache;
 
@@ -69,25 +68,10 @@ static const char *external_diff(void)
 {
        static const char *external_diff_cmd = NULL;
        static int done_preparing = 0;
-       const char *env_diff_opts;
 
        if (done_preparing)
                return external_diff_cmd;
-
-       /*
-        * Default values above are meant to match the
-        * Linux kernel development style.  Examples of
-        * alternative styles you can specify via environment
-        * variables are:
-        *
-        * GIT_DIFF_OPTS="-c";
-        */
        external_diff_cmd = getenv("GIT_EXTERNAL_DIFF");
-
-       /* In case external diff fails... */
-       env_diff_opts = getenv("GIT_DIFF_OPTS");
-       if (env_diff_opts) diff_opts = env_diff_opts;
-
        done_preparing = 1;
        return external_diff_cmd;
 }
@@ -101,13 +85,12 @@ static struct diff_tempfile {
        char tmp_path[TEMPFILE_PATH_LEN];
 } diff_temp[2];
 
-static int count_lines(const char *filename)
+static int count_lines(const char *data, int size)
 {
-       FILE *in;
        int count, ch, completely_empty = 1, nl_just_seen = 0;
-       in = fopen(filename, "r");
        count = 0;
-       while ((ch = fgetc(in)) != EOF)
+       while (0 < size--) {
+               ch = *data++;
                if (ch == '\n') {
                        count++;
                        nl_just_seen = 1;
@@ -117,7 +100,7 @@ static int count_lines(const char *filename)
                        nl_just_seen = 0;
                        completely_empty = 0;
                }
-       fclose(in);
+       }
        if (completely_empty)
                return 0;
        if (!nl_just_seen)
@@ -140,12 +123,11 @@ static void print_line_count(int count)
        }
 }
 
-static void copy_file(int prefix, const char *filename)
+static void copy_file(int prefix, const char *data, int size)
 {
-       FILE *in;
        int ch, nl_just_seen = 1;
-       in = fopen(filename, "r");
-       while ((ch = fgetc(in)) != EOF) {
+       while (0 < size--) {
+               ch = *data++;
                if (nl_just_seen)
                        putchar(prefix);
                putchar(ch);
@@ -154,107 +136,152 @@ static void copy_file(int prefix, const char *filename)
                else
                        nl_just_seen = 0;
        }
-       fclose(in);
        if (!nl_just_seen)
                printf("\n\\ No newline at end of file\n");
 }
 
 static void emit_rewrite_diff(const char *name_a,
                              const char *name_b,
-                             struct diff_tempfile *temp)
+                             struct diff_filespec *one, 
+                             struct diff_filespec *two)
 {
        /* Use temp[i].name as input, name_a and name_b as labels */
        int lc_a, lc_b;
-       lc_a = count_lines(temp[0].name);
-       lc_b = count_lines(temp[1].name);
+       lc_a = count_lines(one->data, one->size);
+       lc_b = count_lines(two->data, two->size);
        printf("--- %s\n+++ %s\n@@ -", name_a, name_b);
        print_line_count(lc_a);
        printf(" +");
        print_line_count(lc_b);
        printf(" @@\n");
        if (lc_a)
-               copy_file('-', temp[0].name);
+               copy_file('-', one->data, one->size);
        if (lc_b)
-               copy_file('+', temp[1].name);
+               copy_file('+', two->data, two->size);
+}
+
+static int fill_mmfile(mmfile_t *mf, struct diff_filespec *one)
+{
+       if (!DIFF_FILE_VALID(one)) {
+               mf->ptr = ""; /* does not matter */
+               mf->size = 0;
+               return 0;
+       }
+       else if (diff_populate_filespec(one, 0))
+               return -1;
+       mf->ptr = one->data;
+       mf->size = one->size;
+       return 0;
+}
+
+struct emit_callback {
+       const char **label_path;
+};
+
+static int fn_out(void *priv, mmbuffer_t *mb, int nbuf)
+{
+       int i;
+       struct emit_callback *ecbdata = priv;
+
+       if (ecbdata->label_path[0]) {
+               printf("--- %s\n", ecbdata->label_path[0]);
+               printf("+++ %s\n", ecbdata->label_path[1]);
+               ecbdata->label_path[0] = ecbdata->label_path[1] = NULL;
+       }
+       for (i = 0; i < nbuf; i++)
+               if (!fwrite(mb[i].ptr, mb[i].size, 1, stdout))
+                       return -1;
+       return 0;
+}
+
+#define FIRST_FEW_BYTES 8000
+static int mmfile_is_binary(mmfile_t *mf)
+{
+       long sz = mf->size;
+       if (FIRST_FEW_BYTES < sz)
+               sz = FIRST_FEW_BYTES;
+       if (memchr(mf->ptr, 0, sz))
+               return 1;
+       return 0;
 }
 
 static void builtin_diff(const char *name_a,
                         const char *name_b,
-                        struct diff_tempfile *temp,
+                        struct diff_filespec *one,
+                        struct diff_filespec *two,
                         const char *xfrm_msg,
                         int complete_rewrite)
 {
-       int i, next_at, cmd_size;
-       const char *const diff_cmd = "diff -L%s -L%s";
-       const char *const diff_arg  = "-- %s %s||:"; /* "||:" is to return 0 */
-       const char *input_name_sq[2];
-       const char *label_path[2];
-       char *cmd;
-
-       /* diff_cmd and diff_arg have 4 %s in total which makes
-        * the sum of these strings 8 bytes larger than required.
-        * we use 2 spaces around diff-opts, and we need to count
-        * terminating NUL; we used to subtract 5 here, but we do not
-        * care about small leaks in this subprocess that is about
-        * to exec "diff" anymore.
-        */
-       cmd_size = (strlen(diff_cmd) + strlen(diff_opts) + strlen(diff_arg)
-                   + 128);
-
-       for (i = 0; i < 2; i++) {
-               input_name_sq[i] = sq_quote(temp[i].name);
-               if (!strcmp(temp[i].name, "/dev/null"))
-                       label_path[i] = "/dev/null";
-               else if (!i)
-                       label_path[i] = sq_quote(quote_two("a/", name_a));
-               else
-                       label_path[i] = sq_quote(quote_two("b/", name_b));
-               cmd_size += (strlen(label_path[i]) + strlen(input_name_sq[i]));
-       }
-
-       cmd = xmalloc(cmd_size);
-
-       next_at = 0;
-       next_at += snprintf(cmd+next_at, cmd_size-next_at,
-                           diff_cmd, label_path[0], label_path[1]);
-       next_at += snprintf(cmd+next_at, cmd_size-next_at,
-                           " %s ", diff_opts);
-       next_at += snprintf(cmd+next_at, cmd_size-next_at,
-                           diff_arg, input_name_sq[0], input_name_sq[1]);
-
-       printf("diff --git %s %s\n",
-              quote_two("a/", name_a), quote_two("b/", name_b));
-       if (label_path[0][0] == '/') {
-               /* dev/null */
-               printf("new file mode %s\n", temp[1].mode);
+       mmfile_t mf1, mf2;
+       const char *lbl[2];
+       char *a_one, *b_two;
+
+       a_one = quote_two("a/", name_a);
+       b_two = quote_two("b/", name_b);
+       lbl[0] = DIFF_FILE_VALID(one) ? a_one : "/dev/null";
+       lbl[1] = DIFF_FILE_VALID(two) ? b_two : "/dev/null";
+       printf("diff --git %s %s\n", a_one, b_two);
+       if (lbl[0][0] == '/') {
+               /* /dev/null */
+               printf("new file mode %06o\n", two->mode);
                if (xfrm_msg && xfrm_msg[0])
                        puts(xfrm_msg);
        }
-       else if (label_path[1][0] == '/') {
-               printf("deleted file mode %s\n", temp[0].mode);
+       else if (lbl[1][0] == '/') {
+               printf("deleted file mode %06o\n", one->mode);
                if (xfrm_msg && xfrm_msg[0])
                        puts(xfrm_msg);
        }
        else {
-               if (strcmp(temp[0].mode, temp[1].mode)) {
-                       printf("old mode %s\n", temp[0].mode);
-                       printf("new mode %s\n", temp[1].mode);
+               if (one->mode != two->mode) {
+                       printf("old mode %06o\n", one->mode);
+                       printf("new mode %06o\n", two->mode);
                }
                if (xfrm_msg && xfrm_msg[0])
                        puts(xfrm_msg);
-               if (strncmp(temp[0].mode, temp[1].mode, 3))
-                       /* we do not run diff between different kind
-                        * of objects.
-                        */
-                       exit(0);
+               /*
+                * we do not run diff between different kind
+                * of objects.
+                */
+               if ((one->mode ^ two->mode) & S_IFMT)
+                       goto free_ab_and_return;
                if (complete_rewrite) {
-                       fflush(NULL);
-                       emit_rewrite_diff(name_a, name_b, temp);
-                       exit(0);
+                       emit_rewrite_diff(name_a, name_b, one, two);
+                       goto free_ab_and_return;
                }
        }
-       fflush(NULL);
-       execlp("/bin/sh","sh", "-c", cmd, NULL);
+
+       if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
+               die("unable to read files to diff");
+
+       if (mmfile_is_binary(&mf1) || mmfile_is_binary(&mf2))
+               printf("Binary files %s and %s differ\n", lbl[0], lbl[1]);
+       else {
+               /* Crazy xdl interfaces.. */
+               const char *diffopts = getenv("GIT_DIFF_OPTS");
+               xpparam_t xpp;
+               xdemitconf_t xecfg;
+               xdemitcb_t ecb;
+               struct emit_callback ecbdata;
+
+               ecbdata.label_path = lbl;
+               xpp.flags = XDF_NEED_MINIMAL;
+               xecfg.ctxlen = 3;
+               if (!diffopts)
+                       ;
+               else if (!strncmp(diffopts, "--unified=", 10))
+                       xecfg.ctxlen = strtoul(diffopts + 10, NULL, 10);
+               else if (!strncmp(diffopts, "-u", 2))
+                       xecfg.ctxlen = strtoul(diffopts + 2, NULL, 10);
+               ecb.outf = fn_out;
+               ecb.priv = &ecbdata;
+               xdl_diff(&mf1, &mf2, &xpp, &xecfg, &ecb);
+       }
+
+ free_ab_and_return:
+       free(a_one);
+       free(b_two);
+       return;
 }
 
 struct diff_filespec *alloc_filespec(const char *path)
@@ -311,7 +338,7 @@ static int work_tree_matches(const char *name, const unsigned char *sha1)
        ce = active_cache[pos];
        if ((lstat(name, &st) < 0) ||
            !S_ISREG(st.st_mode) || /* careful! */
-           ce_match_stat(ce, &st) ||
+           ce_match_stat(ce, &st, 0) ||
            memcmp(sha1, ce->sha1, 20))
                return 0;
        /* we return 1 only when we can stat, it is a regular file,
@@ -504,9 +531,9 @@ static void prepare_temp_file(const char *name,
                }
                if (S_ISLNK(st.st_mode)) {
                        int ret;
-                       char *buf, buf_[1024];
-                       buf = ((sizeof(buf_) < st.st_size) ?
-                              xmalloc(st.st_size) : buf_);
+                       char buf[PATH_MAX + 1]; /* ought to be SYMLINK_MAX */
+                       if (sizeof(buf) <= st.st_size)
+                               die("symlink too long: %s", name);
                        ret = readlink(name, buf, st.st_size);
                        if (ret < 0)
                                die("readlink(%s)", name);
@@ -555,6 +582,42 @@ static void remove_tempfile(void)
 static void remove_tempfile_on_signal(int signo)
 {
        remove_tempfile();
+       signal(SIGINT, SIG_DFL);
+       raise(signo);
+}
+
+static int spawn_prog(const char *pgm, const char **arg)
+{
+       pid_t pid;
+       int status;
+
+       fflush(NULL);
+       pid = fork();
+       if (pid < 0)
+               die("unable to fork");
+       if (!pid) {
+               execvp(pgm, (char *const*) arg);
+               exit(255);
+       }
+
+       while (waitpid(pid, &status, 0) < 0) {
+               if (errno == EINTR)
+                       continue;
+               return -1;
+       }
+
+       /* Earlier we did not check the exit status because
+        * diff exits non-zero if files are different, and
+        * we are not interested in knowing that.  It was a
+        * mistake which made it harder to quit a diff-*
+        * session that uses the git-apply-patch-script as
+        * the GIT_EXTERNAL_DIFF.  A custom GIT_EXTERNAL_DIFF
+        * should also exit non-zero only when it wants to
+        * abort the entire diff-* session.
+        */
+       if (WIFEXITED(status) && !WEXITSTATUS(status))
+               return 0;
+       return -1;
 }
 
 /* An external diff command takes:
@@ -571,11 +634,12 @@ static void run_external_diff(const char *pgm,
                              const char *xfrm_msg,
                              int complete_rewrite)
 {
+       const char *spawn_arg[10];
        struct diff_tempfile *temp = diff_temp;
-       pid_t pid;
-       int status;
+       int retval;
        static int atexit_asked = 0;
        const char *othername;
+       const char **arg = &spawn_arg[0];
 
        othername = (other? other : name);
        if (one && two) {
@@ -590,59 +654,50 @@ static void run_external_diff(const char *pgm,
                signal(SIGINT, remove_tempfile_on_signal);
        }
 
-       fflush(NULL);
-       pid = fork();
-       if (pid < 0)
-               die("unable to fork");
-       if (!pid) {
-               if (pgm) {
-                       if (one && two) {
-                               const char *exec_arg[10];
-                               const char **arg = &exec_arg[0];
-                               *arg++ = pgm;
-                               *arg++ = name;
-                               *arg++ = temp[0].name;
-                               *arg++ = temp[0].hex;
-                               *arg++ = temp[0].mode;
-                               *arg++ = temp[1].name;
-                               *arg++ = temp[1].hex;
-                               *arg++ = temp[1].mode;
-                               if (other) {
-                                       *arg++ = other;
-                                       *arg++ = xfrm_msg;
-                               }
-                               *arg = NULL;
-                               execvp(pgm, (char *const*) exec_arg);
-                       }
-                       else
-                               execlp(pgm, pgm, name, NULL);
+       if (one && two) {
+               *arg++ = pgm;
+               *arg++ = name;
+               *arg++ = temp[0].name;
+               *arg++ = temp[0].hex;
+               *arg++ = temp[0].mode;
+               *arg++ = temp[1].name;
+               *arg++ = temp[1].hex;
+               *arg++ = temp[1].mode;
+               if (other) {
+                       *arg++ = other;
+                       *arg++ = xfrm_msg;
                }
-               /*
-                * otherwise we use the built-in one.
-                */
-               if (one && two)
-                       builtin_diff(name, othername, temp, xfrm_msg,
-                                    complete_rewrite);
-               else
-                       printf("* Unmerged path %s\n", name);
-               exit(0);
+       } else {
+               *arg++ = pgm;
+               *arg++ = name;
        }
-       if (waitpid(pid, &status, 0) < 0 ||
-           !WIFEXITED(status) || WEXITSTATUS(status)) {
-               /* Earlier we did not check the exit status because
-                * diff exits non-zero if files are different, and
-                * we are not interested in knowing that.  It was a
-                * mistake which made it harder to quit a diff-*
-                * session that uses the git-apply-patch-script as
-                * the GIT_EXTERNAL_DIFF.  A custom GIT_EXTERNAL_DIFF
-                * should also exit non-zero only when it wants to
-                * abort the entire diff-* session.
-                */
-               remove_tempfile();
+       *arg = NULL;
+       retval = spawn_prog(pgm, spawn_arg);
+       remove_tempfile();
+       if (retval) {
                fprintf(stderr, "external diff died, stopping at %s.\n", name);
                exit(1);
        }
-       remove_tempfile();
+}
+
+static void run_diff_cmd(const char *pgm,
+                        const char *name,
+                        const char *other,
+                        struct diff_filespec *one,
+                        struct diff_filespec *two,
+                        const char *xfrm_msg,
+                        int complete_rewrite)
+{
+       if (pgm) {
+               run_external_diff(pgm, name, other, one, two, xfrm_msg,
+                                 complete_rewrite);
+               return;
+       }
+       if (one && two)
+               builtin_diff(name, other ? other : name,
+                            one, two, xfrm_msg, complete_rewrite);
+       else
+               printf("* Unmerged path %s\n", name);
 }
 
 static void diff_fill_sha1_info(struct diff_filespec *one)
@@ -650,7 +705,7 @@ static void diff_fill_sha1_info(struct diff_filespec *one)
        if (DIFF_FILE_VALID(one)) {
                if (!one->sha1_valid) {
                        struct stat st;
-                       if (stat(one->path, &st) < 0)
+                       if (lstat(one->path, &st) < 0)
                                die("stat %s", one->path);
                        if (index_path(one->sha1, one->path, &st, 0))
                                die("cannot hash %s\n", one->path);
@@ -674,8 +729,7 @@ static void run_diff(struct diff_filepair *p, struct diff_options *o)
 
        if (DIFF_PAIR_UNMERGED(p)) {
                /* unmerged */
-               run_external_diff(pgm, p->one->path, NULL, NULL, NULL, NULL,
-                                 0);
+               run_diff_cmd(pgm, p->one->path, NULL, NULL, NULL, NULL, 0);
                return;
        }
 
@@ -723,11 +777,13 @@ static void run_diff(struct diff_filepair *p, struct diff_options *o)
 
        if (memcmp(one->sha1, two->sha1, 20)) {
                char one_sha1[41];
-               const char *index_fmt = o->full_index ? "index %s..%s" : "index %.7s..%.7s";
+               int abbrev = o->full_index ? 40 : DEFAULT_ABBREV;
                memcpy(one_sha1, sha1_to_hex(one->sha1), 41);
 
                len += snprintf(msg + len, sizeof(msg) - len,
-                               index_fmt, one_sha1, sha1_to_hex(two->sha1));
+                               "index %.*s..%.*s",
+                               abbrev, one_sha1, abbrev,
+                               sha1_to_hex(two->sha1));
                if (one->mode == two->mode)
                        len += snprintf(msg + len, sizeof(msg) - len,
                                        " %06o", one->mode);
@@ -745,15 +801,15 @@ static void run_diff(struct diff_filepair *p, struct diff_options *o)
                 * needs to be split into deletion and creation.
                 */
                struct diff_filespec *null = alloc_filespec(two->path);
-               run_external_diff(NULL, name, other, one, null, xfrm_msg, 0);
+               run_diff_cmd(NULL, name, other, one, null, xfrm_msg, 0);
                free(null);
                null = alloc_filespec(one->path);
-               run_external_diff(NULL, name, other, null, two, xfrm_msg, 0);
+               run_diff_cmd(NULL, name, other, null, two, xfrm_msg, 0);
                free(null);
        }
        else
-               run_external_diff(pgm, name, other, one, two, xfrm_msg,
-                                 complete_rewrite);
+               run_diff_cmd(pgm, name, other, one, two, xfrm_msg,
+                            complete_rewrite);
 
        free(name_munged);
        free(other_munged);
@@ -785,12 +841,14 @@ int diff_setup_done(struct diff_options *options)
                         * so it is safe for us to do this here.  Also
                         * it does not smudge active_cache or active_nr
                         * when it fails, so we do not have to worry about
-                        * cleaning it up oufselves either.
+                        * cleaning it up ourselves either.
                         */
                        read_cache();
        }
        if (options->setup & DIFF_SETUP_USE_SIZE_CACHE)
                use_size_cache = 1;
+       if (options->abbrev <= 0 || 40 < options->abbrev)
+               options->abbrev = 40; /* full */
 
        return 0;
 }
@@ -841,6 +899,15 @@ int diff_opt_parse(struct diff_options *options, const char **av, int ac)
        }
        else if (!strcmp(arg, "--find-copies-harder"))
                options->find_copies_harder = 1;
+       else if (!strcmp(arg, "--abbrev"))
+               options->abbrev = DEFAULT_ABBREV;
+       else if (!strncmp(arg, "--abbrev=", 9)) {
+               options->abbrev = strtoul(arg + 9, NULL, 10);
+               if (options->abbrev < MINIMUM_ABBREV)
+                       options->abbrev = MINIMUM_ABBREV;
+               else if (40 < options->abbrev)
+                       options->abbrev = 40;
+       }
        else
                return 0;
        return 1;
@@ -947,14 +1014,41 @@ void diff_free_filepair(struct diff_filepair *p)
        free(p);
 }
 
+/* This is different from find_unique_abbrev() in that
+ * it stuffs the result with dots for alignment.
+ */
+const char *diff_unique_abbrev(const unsigned char *sha1, int len)
+{
+       int abblen;
+       const char *abbrev;
+       if (len == 40)
+               return sha1_to_hex(sha1);
+
+       abbrev = find_unique_abbrev(sha1, len);
+       if (!abbrev)
+               return sha1_to_hex(sha1);
+       abblen = strlen(abbrev);
+       if (abblen < 37) {
+               static char hex[41];
+               if (len < abblen && abblen <= len + 2)
+                       sprintf(hex, "%s%.*s", abbrev, len+3-abblen, "..");
+               else
+                       sprintf(hex, "%s...", abbrev);
+               return hex;
+       }
+       return sha1_to_hex(sha1);
+}
+
 static void diff_flush_raw(struct diff_filepair *p,
                           int line_termination,
                           int inter_name_termination,
-                          int output_format)
+                          struct diff_options *options)
 {
        int two_paths;
        char status[10];
+       int abbrev = options->abbrev;
        const char *path_one, *path_two;
+       int output_format = options->output_format;
 
        path_one = p->one->path;
        path_two = p->two->path;
@@ -985,8 +1079,10 @@ static void diff_flush_raw(struct diff_filepair *p,
        }
        if (output_format != DIFF_FORMAT_NAME_STATUS) {
                printf(":%06o %06o %s ",
-                      p->one->mode, p->two->mode, sha1_to_hex(p->one->sha1));
-               printf("%s ", sha1_to_hex(p->two->sha1));
+                      p->one->mode, p->two->mode,
+                      diff_unique_abbrev(p->one->sha1, abbrev));
+               printf("%s ",
+                      diff_unique_abbrev(p->two->sha1, abbrev));
        }
        printf("%s%c%s", status, inter_name_termination, path_one);
        if (two_paths)
@@ -1194,7 +1290,7 @@ void diff_flush(struct diff_options *options)
                case DIFF_FORMAT_NAME_STATUS:
                        diff_flush_raw(p, line_termination,
                                       inter_name_termination,
-                                      diff_output_format);
+                                      options);
                        break;
                case DIFF_FORMAT_NAME:
                        diff_flush_name(p,