Fixed a leak in read-tree
[git.git] / diff.c
diff --git a/diff.c b/diff.c
index 7383489..3230997 100644 (file)
--- a/diff.c
+++ b/diff.c
@@ -3,15 +3,16 @@
  */
 #include <sys/types.h>
 #include <sys/wait.h>
+#include <signal.h>
+#include <limits.h>
 #include "cache.h"
 #include "diff.h"
 
-static char *diff_cmd = "diff -L'k/%s' -L'l/%s'";
-static char *diff_opts = "-pu";
+static const char *diff_opts = "-pu";
 
 static const char *external_diff(void)
 {
-       static char *external_diff_cmd = NULL;
+       static const char *external_diff_cmd = NULL;
        static int done_preparing = 0;
 
        if (done_preparing)
@@ -23,15 +24,13 @@ static const char *external_diff(void)
         * alternative styles you can specify via environment
         * variables are:
         *
-        * GIT_DIFF_CMD="diff -L '%s' -L '%s'"
         * GIT_DIFF_OPTS="-c";
         */
-       if (getenv("GIT_EXTERNAL_DIFF"))
-               external_diff_cmd = getenv("GIT_EXTERNAL_DIFF");
+       if (gitenv("GIT_EXTERNAL_DIFF"))
+               external_diff_cmd = gitenv("GIT_EXTERNAL_DIFF");
 
        /* In case external diff fails... */
-       diff_cmd = getenv("GIT_DIFF_CMD") ? : diff_cmd;
-       diff_opts = getenv("GIT_DIFF_OPTS") ? : diff_opts;
+       diff_opts = gitenv("GIT_DIFF_OPTS") ? : diff_opts;
 
        done_preparing = 1;
        return external_diff_cmd;
@@ -83,42 +82,136 @@ static struct diff_tempfile {
 static void builtin_diff(const char *name,
                         struct diff_tempfile *temp)
 {
-       static char *diff_arg  = "'%s' '%s'";
-       const char *name_1_sq = sq_expand(temp[0].name);
-       const char *name_2_sq = sq_expand(temp[1].name);
+       int i, next_at;
+       const char *diff_cmd = "diff -L'%s%s' -L'%s%s'";
+       const char *diff_arg  = "'%s' '%s'||:"; /* "||:" is to return 0 */
+       const char *input_name_sq[2];
+       const char *path0[2];
+       const char *path1[2];
        const char *name_sq = sq_expand(name);
-
-       /* diff_cmd and diff_arg have 4 %s in total which makes
-        * the sum of these strings 8 bytes larger than required.
+       char *cmd;
+       
+       /* diff_cmd and diff_arg have 6 %s in total which makes
+        * the sum of these strings 12 bytes larger than required.
         * we use 2 spaces around diff-opts, and we need to count
-        * terminating NUL, so we subtract 5 here.
+        * terminating NUL, so we subtract 9 here.
         */
-       int cmd_size = (strlen(diff_cmd) + 
-                       strlen(name_sq) * 2 +
-                       strlen(diff_opts) +
-                       strlen(diff_arg) +
-                       strlen(name_1_sq) + strlen(name_2_sq)
-                       - 5);
-       char *cmd = xmalloc(cmd_size);
-       int next_at = 0;
+       int cmd_size = (strlen(diff_cmd) + strlen(diff_opts) +
+                       strlen(diff_arg) - 9);
+       for (i = 0; i < 2; i++) {
+               input_name_sq[i] = sq_expand(temp[i].name);
+               if (!strcmp(temp[i].name, "/dev/null")) {
+                       path0[i] = "/dev/null";
+                       path1[i] = "";
+               } else {
+                       path0[i] = i ? "b/" : "a/";
+                       path1[i] = name_sq;
+               }
+               cmd_size += (strlen(path0[i]) + strlen(path1[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, name_sq, name_sq);
+                           diff_cmd,
+                           path0[0], path1[0], path0[1], path1[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, name_1_sq, name_2_sq);
+                           diff_arg, input_name_sq[0], input_name_sq[1]);
+
+       if (!path1[0][0])
+               printf("Created: %s (mode:%s)\n", name, temp[1].mode);
+       else if (!path1[1][0])
+               printf("Deleted: %s\n", name);
+       else if (strcmp(temp[0].mode, temp[1].mode)) {
+               printf("Mode changed: %s (%s->%s)\n", name,
+                      temp[0].mode, temp[1].mode);
+               /* Be careful.  We do not want to diff between
+                * symlink and a file.
+                */
+               if (strncmp(temp[0].mode, "120", 3) !=
+                   strncmp(temp[1].mode, "120", 3))
+                       exit(0);
+       }
+       fflush(NULL);
        execlp("/bin/sh","sh", "-c", cmd, NULL);
 }
 
+/*
+ * Given a name and sha1 pair, if the dircache tells us the file in
+ * the work tree has that object contents, return true, so that
+ * prepare_temp_file() does not have to inflate and extract.
+ */
+static int work_tree_matches(const char *name, const unsigned char *sha1)
+{
+       struct cache_entry *ce;
+       struct stat st;
+       int pos, len;
+       
+       /* We do not read the cache ourselves here, because the
+        * benchmark with my previous version that always reads cache
+        * shows that it makes things worse for diff-tree comparing
+        * two linux-2.6 kernel trees in an already checked out work
+        * tree.  This is because most diff-tree comparison deals with
+        * only a small number of files, while reading the cache is
+        * expensive for a large project, and its cost outweighs the
+        * savings we get by not inflating the object to a temporary
+        * file.  Practically, this code only helps when we are used
+        * by diff-cache --cached, which does read the cache before
+        * calling us.
+        */ 
+       if (!active_cache)
+               return 0;
+
+       len = strlen(name);
+       pos = cache_name_pos(name, len);
+       if (pos < 0)
+               return 0;
+       ce = active_cache[pos];
+       if ((lstat(name, &st) < 0) ||
+           !S_ISREG(st.st_mode) ||
+           cache_match_stat(ce, &st) ||
+           memcmp(sha1, ce->sha1, 20))
+               return 0;
+       return 1;
+}
+
+static void prep_temp_blob(struct diff_tempfile *temp,
+                          void *blob,
+                          unsigned long size,
+                          unsigned char *sha1,
+                          int mode)
+{
+       int fd;
+
+       strcpy(temp->tmp_path, ".diff_XXXXXX");
+       fd = mkstemp(temp->tmp_path);
+       if (fd < 0)
+               die("unable to create temp-file");
+       if (write(fd, blob, size) != size)
+               die("unable to write temp-file");
+       close(fd);
+       temp->name = temp->tmp_path;
+       strcpy(temp->hex, sha1_to_hex(sha1));
+       temp->hex[40] = 0;
+       sprintf(temp->mode, "%06o", mode);
+}
+
 static void prepare_temp_file(const char *name,
                              struct diff_tempfile *temp,
                              struct diff_spec *one)
 {
        static unsigned char null_sha1[20] = { 0, };
+       int use_work_tree = 0;
 
        if (!one->file_valid) {
        not_a_valid_file:
+               /* A '-' entry produces this for file-2, and
+                * a '+' entry produces this for file-1.
+                */
                temp->name = "/dev/null";
                strcpy(temp->hex, ".");
                strcpy(temp->mode, ".");
@@ -126,46 +219,53 @@ static void prepare_temp_file(const char *name,
        }
 
        if (one->sha1_valid &&
-           !memcmp(one->u.sha1, null_sha1, sizeof(null_sha1))) {
-               one->sha1_valid = 0;
-               one->u.name = name;
-       }
+           (!memcmp(one->blob_sha1, null_sha1, sizeof(null_sha1)) ||
+            work_tree_matches(name, one->blob_sha1)))
+               use_work_tree = 1;
 
-       if (!one->sha1_valid) {
+       if (!one->sha1_valid || use_work_tree) {
                struct stat st;
-               temp->name = one->u.name;
-               if (stat(temp->name, &st) < 0) {
+               temp->name = name;
+               if (lstat(temp->name, &st) < 0) {
                        if (errno == ENOENT)
                                goto not_a_valid_file;
                        die("stat(%s): %s", temp->name, strerror(errno));
                }
-               strcpy(temp->hex, ".");
-               sprintf(temp->mode, "%06o",
-                       S_IFREG |ce_permissions(st.st_mode));
+               if (S_ISLNK(st.st_mode)) {
+                       int ret;
+                       char *buf, buf_[1024];
+                       buf = ((sizeof(buf_) < st.st_size) ?
+                              xmalloc(st.st_size) : buf_);
+                       ret = readlink(name, buf, st.st_size);
+                       if (ret < 0)
+                               die("readlink(%s)", name);
+                       prep_temp_blob(temp, buf, st.st_size,
+                                      (one->sha1_valid ?
+                                       one->blob_sha1 : null_sha1),
+                                      (one->sha1_valid ?
+                                       one->mode : S_IFLNK));
+               }
+               else {
+                       if (!one->sha1_valid)
+                               strcpy(temp->hex, sha1_to_hex(null_sha1));
+                       else
+                               strcpy(temp->hex, sha1_to_hex(one->blob_sha1));
+                       sprintf(temp->mode, "%06o",
+                               S_IFREG |ce_permissions(st.st_mode));
+               }
+               return;
        }
        else {
-               int fd;
                void *blob;
                char type[20];
                unsigned long size;
 
-               blob = read_sha1_file(one->u.sha1, type, &size);
+               blob = read_sha1_file(one->blob_sha1, type, &size);
                if (!blob || strcmp(type, "blob"))
                        die("unable to read blob object for %s (%s)",
-                           name, sha1_to_hex(one->u.sha1));
-
-               strcpy(temp->tmp_path, ".diff_XXXXXX");
-               fd = mkstemp(temp->tmp_path);
-               if (fd < 0)
-                       die("unable to create temp-file");
-               if (write(fd, blob, size) != size)
-                       die("unable to write temp-file");
-               close(fd);
+                           name, sha1_to_hex(one->blob_sha1));
+               prep_temp_blob(temp, blob, size, one->blob_sha1, one->mode);
                free(blob);
-               temp->name = temp->tmp_path;
-               strcpy(temp->hex, sha1_to_hex(one->u.sha1));
-               temp->hex[40] = 0;
-               sprintf(temp->mode, "%06o", one->mode);
        }
 }
 
@@ -180,6 +280,11 @@ static void remove_tempfile(void)
                }
 }
 
+static void remove_tempfile_on_signal(int signo)
+{
+       remove_tempfile();
+}
+
 /* An external diff command takes:
  *
  * diff-cmd name infile1 infile1-sha1 infile1-mode \
@@ -191,7 +296,8 @@ void run_external_diff(const char *name,
                       struct diff_spec *two)
 {
        struct diff_tempfile *temp = diff_temp;
-       int pid, status;
+       pid_t pid;
+       int status;
        static int atexit_asked = 0;
 
        if (one && two) {
@@ -203,6 +309,7 @@ void run_external_diff(const char *name,
                        atexit_asked = 1;
                        atexit(remove_tempfile);
                }
+               signal(SIGINT, remove_tempfile_on_signal);
        }
 
        fflush(NULL);
@@ -230,9 +337,21 @@ void run_external_diff(const char *name,
                        printf("* Unmerged path %s\n", name);
                exit(0);
        }
-       if (waitpid(pid, &status, 0) < 0 || !WIFEXITED(status))
-               die("diff program failed");
-
+       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();
+               fprintf(stderr, "external diff died, stopping at %s.\n", name);
+               exit(1);
+       }
        remove_tempfile();
 }
 
@@ -243,7 +362,7 @@ void diff_addremove(int addremove, unsigned mode,
        char concatpath[PATH_MAX];
        struct diff_spec spec[2], *one, *two;
 
-       memcpy(spec[0].u.sha1, sha1, 20);
+       memcpy(spec[0].blob_sha1, sha1, 20);
        spec[0].mode = mode;
        spec[0].sha1_valid = spec[0].file_valid = 1;
        spec[1].file_valid = 0;
@@ -256,7 +375,6 @@ void diff_addremove(int addremove, unsigned mode,
        
        if (path) {
                strcpy(concatpath, base);
-               strcat(concatpath, "/");
                strcat(concatpath, path);
        }
        run_external_diff(path ? concatpath : base, one, two);
@@ -269,16 +387,15 @@ void diff_change(unsigned old_mode, unsigned new_mode,
        char concatpath[PATH_MAX];
        struct diff_spec spec[2];
 
-       memcpy(spec[0].u.sha1, old_sha1, 20);
+       memcpy(spec[0].blob_sha1, old_sha1, 20);
        spec[0].mode = old_mode;
-       memcpy(spec[1].u.sha1, new_sha1, 20);
+       memcpy(spec[1].blob_sha1, new_sha1, 20);
        spec[1].mode = new_mode;
        spec[0].sha1_valid = spec[0].file_valid = 1;
        spec[1].sha1_valid = spec[1].file_valid = 1;
 
        if (path) {
                strcpy(concatpath, base);
-               strcat(concatpath, "/");
                strcat(concatpath, path);
        }
        run_external_diff(path ? concatpath : base, &spec[0], &spec[1]);