Merge pull request libgit2#1595 from arrbee/even-more-rename-fixes

Even more rename detection fixes

examples/diff.c

@@ -84,6 +84,10 @@ static int check_uint16_param(const char *arg, const char *pattern, uint16_t *va
 	char *endptr = NULL;
 	if (strncmp(arg, pattern, len))
 		return 0;
+	if (arg[len] == '\0' && pattern[len - 1] != '=')
+		return 1;
+	if (arg[len] == '=')
+		len++;
 	strval = strtoul(arg + len, &endptr, 0);
 	if (endptr == arg)
 		return 0;
@@ -110,13 +114,20 @@ static void usage(const char *message, const char *arg)
 	exit(1);
 }
 
+enum {
+	FORMAT_PATCH = 0,
+	FORMAT_COMPACT = 1,
+	FORMAT_RAW = 2
+};
+
 int main(int argc, char *argv[])
 {
 	git_repository *repo = NULL;
 	git_tree *t1 = NULL, *t2 = NULL;
 	git_diff_options opts = GIT_DIFF_OPTIONS_INIT;
+	git_diff_find_options findopts = GIT_DIFF_FIND_OPTIONS_INIT;
 	git_diff_list *diff;
-	int i, color = -1, compact = 0, cached = 0;
+	int i, color = -1, format = FORMAT_PATCH, cached = 0;
 	char *a, *treeish1 = NULL, *treeish2 = NULL;
 	const char *dir = ".";
 
@@ -137,11 +148,13 @@ int main(int argc, char *argv[])
 		}
 		else if (!strcmp(a, "-p") || !strcmp(a, "-u") ||
 			!strcmp(a, "--patch"))
-			compact = 0;
+			format = FORMAT_PATCH;
 		else if (!strcmp(a, "--cached"))
 			cached = 1;
 		else if (!strcmp(a, "--name-status"))
-			compact = 1;
+			format = FORMAT_COMPACT;
+		else if (!strcmp(a, "--raw"))
+			format = FORMAT_RAW;
 		else if (!strcmp(a, "--color"))
 			color = 0;
 		else if (!strcmp(a, "--no-color"))
@@ -160,6 +173,20 @@ int main(int argc, char *argv[])
 			opts.flags |= GIT_DIFF_INCLUDE_IGNORED;
 		else if (!strcmp(a, "--untracked"))
 			opts.flags |= GIT_DIFF_INCLUDE_UNTRACKED;
+		else if (check_uint16_param(a, "-M", &findopts.rename_threshold) ||
+				 check_uint16_param(a, "--find-renames",
+					&findopts.rename_threshold))
+			findopts.flags |= GIT_DIFF_FIND_RENAMES;
+		else if (check_uint16_param(a, "-C", &findopts.copy_threshold) ||
+				 check_uint16_param(a, "--find-copies",
+					&findopts.copy_threshold))
+			findopts.flags |= GIT_DIFF_FIND_COPIES;
+		else if (!strcmp(a, "--find-copies-harder"))
+			findopts.flags |= GIT_DIFF_FIND_COPIES_FROM_UNMODIFIED;
+		else if (!strncmp(a, "-B", 2) || !strncmp(a, "--break-rewrites", 16)) {
+			/* TODO: parse thresholds */
+			findopts.flags |= GIT_DIFF_FIND_REWRITES;
+		}
 		else if (!check_uint16_param(a, "-U", &opts.context_lines) &&
 			!check_uint16_param(a, "--unified=", &opts.context_lines) &&
 			!check_uint16_param(a, "--inter-hunk-context=",
@@ -204,13 +231,24 @@ int main(int argc, char *argv[])
 	else
 		check(git_diff_index_to_workdir(&diff, repo, NULL, &opts), "Diff");
 
+	if ((findopts.flags & GIT_DIFF_FIND_ALL) != 0)
+		check(git_diff_find_similar(diff, &findopts),
+			"finding renames and copies ");
+
 	if (color >= 0)
 		fputs(colors[0], stdout);
 
-	if (compact)
-		check(git_diff_print_compact(diff, printer, &color), "Displaying diff");
-	else
+	switch (format) {
+	case FORMAT_PATCH:
 		check(git_diff_print_patch(diff, printer, &color), "Displaying diff");
+		break;
+	case FORMAT_COMPACT:
+		check(git_diff_print_compact(diff, printer, &color), "Displaying diff");
+		break;
+	case FORMAT_RAW:
+		check(git_diff_print_raw(diff, printer, &color), "Displaying diff");
+		break;
+	}
 
 	if (color >= 0)
 		fputs(colors[0], stdout);

include/git2/diff.h

@@ -243,6 +243,19 @@ typedef struct {
  * `NOT_BINARY` flag set to avoid examining file contents if you do not pass
  * in hunk and/or line callbacks to the diff foreach iteration function.  It
  * will just use the git attributes for those files.
+ *
+ * The similarity score is zero unless you call `git_diff_find_similar()`
+ * which does a similarity analysis of files in the diff.  Use that
+ * function to do rename and copy detection, and to split heavily modified
+ * files in add/delete pairs.  After that call, deltas with a status of
+ * GIT_DELTA_RENAMED or GIT_DELTA_COPIED will have a similarity score
+ * between 0 and 100 indicating how similar the old and new sides are.
+ *
+ * If you ask `git_diff_find_similar` to find heavily modified files to
+ * break, but to not *actually* break the records, then GIT_DELTA_MODIFIED
+ * records may have a non-zero similarity score if the self-similarity is
+ * below the split threshold.  To display this value like core Git, invert
+ * the score (a la `printf("M%03d", 100 - delta->similarity)`).
  */
 typedef struct {
 	git_diff_file old_file;
@@ -408,18 +421,28 @@ typedef enum {
 	/** consider unmodified as copy sources? (`--find-copies-harder`) */
 	GIT_DIFF_FIND_COPIES_FROM_UNMODIFIED = (1 << 3),
 
-	/** split large rewrites into delete/add pairs (`--break-rewrites=/M`) */
-	GIT_DIFF_FIND_AND_BREAK_REWRITES = (1 << 4),
+	/** mark large rewrites for split (`--break-rewrites=/M`) */
+	GIT_DIFF_FIND_REWRITES = (1 << 4),
+	/** actually split large rewrites into delete/add pairs */
+	GIT_DIFF_BREAK_REWRITES = (1 << 5),
+	/** mark rewrites for split and break into delete/add pairs */
+	GIT_DIFF_FIND_AND_BREAK_REWRITES =
+		(GIT_DIFF_FIND_REWRITES | GIT_DIFF_BREAK_REWRITES),
+
+	/** find renames/copies for untracked items in working directory */
+	GIT_DIFF_FIND_FOR_UNTRACKED = (1 << 6),
 
 	/** turn on all finding features */
-	GIT_DIFF_FIND_ALL = (0x1f),
+	GIT_DIFF_FIND_ALL = (0x0ff),
 
 	/** measure similarity ignoring leading whitespace (default) */
 	GIT_DIFF_FIND_IGNORE_LEADING_WHITESPACE = 0,
 	/** measure similarity ignoring all whitespace */
-	GIT_DIFF_FIND_IGNORE_WHITESPACE = (1 << 6),
+	GIT_DIFF_FIND_IGNORE_WHITESPACE = (1 << 12),
 	/** measure similarity including all data */
-	GIT_DIFF_FIND_DONT_IGNORE_WHITESPACE = (1 << 7),
+	GIT_DIFF_FIND_DONT_IGNORE_WHITESPACE = (1 << 13),
+	/** measure similarity only by comparing SHAs (fast and cheap) */
+	GIT_DIFF_FIND_EXACT_MATCH_ONLY = (1 << 14),
 } git_diff_find_t;
 
 /**
@@ -446,7 +469,10 @@ typedef struct {
  * - `copy_threshold` is the same as the -C option with a value
  * - `rename_from_rewrite_threshold` matches the top of the -B option
  * - `break_rewrite_threshold` matches the bottom of the -B option
- * - `target_limit` matches the -l option
+ * - `rename_limit` is the maximum number of matches to consider for
+ *   a particular file.  This is a little different from the `-l` option
+ *   to regular Git because we will still process up to this many matches
+ *   before abandoning the search.
  *
  * The `metric` option allows you to plug in a custom similarity metric.
  * Set it to NULL for the default internal metric which is based on sampling
@@ -458,21 +484,21 @@ typedef struct {
 	unsigned int version;
 
 	/** Combination of git_diff_find_t values (default FIND_RENAMES) */
-	unsigned int flags;
+	uint32_t flags;
 
 	/** Similarity to consider a file renamed (default 50) */
-	unsigned int rename_threshold;
+	uint16_t rename_threshold;
 	/** Similarity of modified to be eligible rename source (default 50) */
-	unsigned int rename_from_rewrite_threshold;
+	uint16_t rename_from_rewrite_threshold;
 	/** Similarity to consider a file a copy (default 50) */
-	unsigned int copy_threshold;
+	uint16_t copy_threshold;
 	/** Similarity to split modify into delete/add pair (default 60) */
-	unsigned int break_rewrite_threshold;
+	uint16_t break_rewrite_threshold;
 
-	/** Maximum similarity sources to examine (a la diff's `-l` option or
-	 *  the `diff.renameLimit` config) (default 200)
+	/** Maximum similarity sources to examine for a file (somewhat like
+	 *  git-diff's `-l` option or `diff.renameLimit` config) (default 200)
 	 */
-	unsigned int target_limit;
+	size_t rename_limit;
 
 	/** Pluggable similarity metric; pass NULL to use internal metric */
 	git_diff_similarity_metric *metric;
@@ -688,6 +714,22 @@ GIT_EXTERN(int) git_diff_print_compact(
 	git_diff_data_cb print_cb,
 	void *payload);
 
+/**
+ * Iterate over a diff generating text output like "git diff --raw".
+ *
+ * Returning a non-zero value from the callbacks will terminate the
+ * iteration and cause this return `GIT_EUSER`.
+ *
+ * @param diff A git_diff_list generated by one of the above functions.
+ * @param print_cb Callback to make per line of diff text.
+ * @param payload Reference pointer that will be passed to your callback.
+ * @return 0 on success, GIT_EUSER on non-zero callback, or error code
+ */
+GIT_EXTERN(int) git_diff_print_raw(
+	git_diff_list *diff,
+	git_diff_data_cb print_cb,
+	void *payload);
+
 /**
  * Look up the single character abbreviation for a delta status code.
  *

include/git2/oid.h

@@ -89,6 +89,17 @@ GIT_EXTERN(void) git_oid_fromraw(git_oid *out, const unsigned char *raw);
  */
 GIT_EXTERN(void) git_oid_fmt(char *out, const git_oid *id);
 
+/**
+ * Format a git_oid into a partial hex string.
+ *
+ * @param out output hex string; you say how many bytes to write.
+ *		If the number of bytes is > GIT_OID_HEXSZ, extra bytes
+ *		will be zeroed; if not, a '\0' terminator is NOT added.
+ * @param n number of characters to write into out string
+ * @param oid oid structure to format.
+ */
+GIT_EXTERN(void) git_oid_nfmt(char *out, size_t n, const git_oid *id);
+
 /**
  * Format a git_oid into a loose-object path string.
  *
@@ -117,10 +128,12 @@ GIT_EXTERN(char *) git_oid_allocfmt(const git_oid *id);
  * Format a git_oid into a buffer as a hex format c-string.
  *
  * If the buffer is smaller than GIT_OID_HEXSZ+1, then the resulting
- * oid c-string will be truncated to n-1 characters. If there are
- * any input parameter errors (out == NULL, n == 0, oid == NULL),
- * then a pointer to an empty string is returned, so that the return
- * value can always be printed.
+ * oid c-string will be truncated to n-1 characters (but will still be
+ * NUL-byte terminated).
+ *
+ * If there are any input parameter errors (out == NULL, n == 0, oid ==
+ * NULL), then a pointer to an empty string is returned, so that the
+ * return value can always be printed.
  *
  * @param out the buffer into which the oid string is output.
  * @param n the size of the out buffer.

src/checkout.c

@@ -676,33 +676,26 @@ static int buffer_to_file(
 	int file_open_flags,
 	mode_t file_mode)
 {
-	int fd, error;
+	int error;
 
 	if ((error = git_futils_mkpath2file(path, dir_mode)) < 0)
 		return error;
 
-	if ((fd = p_open(path, file_open_flags, file_mode)) < 0) {
-		giterr_set(GITERR_OS, "Could not open '%s' for writing", path);
-		return fd;
-	}
-
-	if ((error = p_write(fd, git_buf_cstr(buffer), git_buf_len(buffer))) < 0) {
-		giterr_set(GITERR_OS, "Could not write to '%s'", path);
-		(void)p_close(fd);
-	} else {
-		if ((error = p_close(fd)) < 0)
-			giterr_set(GITERR_OS, "Error while closing '%s'", path);
+	if ((error = git_futils_writebuffer(
+			buffer, path, file_open_flags, file_mode)) < 0)
+		return error;
 
-		if ((error = p_stat(path, st)) < 0)
-			giterr_set(GITERR_OS, "Error while statting '%s'", path);
+	if (st != NULL && (error = p_stat(path, st)) < 0) {
+		giterr_set(GITERR_OS, "Error while statting '%s'", path);
+		return error;
 	}
 
-	if (!error &&
-		(file_mode & 0100) != 0 &&
-		(error = p_chmod(path, file_mode)) < 0)
+	if ((file_mode & 0100) != 0 && (error = p_chmod(path, file_mode)) < 0) {
 		giterr_set(GITERR_OS, "Failed to set permissions on '%s'", path);
+		return error;
+	}
 
-	return error;
+	return 0;
 }
 
 static int blob_content_to_file(

src/diff.c

@@ -231,10 +231,23 @@ static char *diff_strdup_prefix(git_pool *pool, const char *prefix)
 		return git_pool_strndup(pool, prefix, len + 1);
 }
 
+GIT_INLINE(const char *) diff_delta__path(const git_diff_delta *delta)
+{
+	const char *str = delta->old_file.path;
+
+	if (!str ||
+		delta->status == GIT_DELTA_ADDED ||
+		delta->status == GIT_DELTA_RENAMED ||
+		delta->status == GIT_DELTA_COPIED)
+		str = delta->new_file.path;
+
+	return str;
+}
+
 int git_diff_delta__cmp(const void *a, const void *b)
 {
 	const git_diff_delta *da = a, *db = b;
-	int val = strcmp(da->old_file.path, db->old_file.path);
+	int val = strcmp(diff_delta__path(da), diff_delta__path(db));
 	return val ? val : ((int)da->status - (int)db->status);
 }
 

src/diff.h

@@ -34,10 +34,18 @@ enum {
 	GIT_DIFF_FLAG__FREE_DATA  = (1 << 8),  /* internal file data is allocated */
 	GIT_DIFF_FLAG__UNMAP_DATA = (1 << 9),  /* internal file data is mmap'ed */
 	GIT_DIFF_FLAG__NO_DATA    = (1 << 10), /* file data should not be loaded */
-	GIT_DIFF_FLAG__TO_DELETE  = (1 << 11), /* delete entry during rename det. */
-	GIT_DIFF_FLAG__TO_SPLIT   = (1 << 12), /* split entry during rename det. */
+
+	GIT_DIFF_FLAG__TO_DELETE  = (1 << 16), /* delete entry during rename det. */
+	GIT_DIFF_FLAG__TO_SPLIT   = (1 << 17), /* split entry during rename det. */
+	GIT_DIFF_FLAG__IS_RENAME_TARGET = (1 << 18),
+	GIT_DIFF_FLAG__IS_RENAME_SOURCE = (1 << 19),
+	GIT_DIFF_FLAG__HAS_SELF_SIMILARITY = (1 << 20),
 };
 
+#define GIT_DIFF_FLAG__CLEAR_INTERNAL(F) (F) = ((F) & 0x00FFFF)
+
+#define GIT_DIFF__VERBOSE  (1 << 30)
+
 struct git_diff_list {
 	git_refcount     rc;
 	git_repository   *repo;