[PATCH/RFC v7 0/5] git checkout: optimise away lots of lstat() calls

[PATCH/RFC v7 0/5] git checkout: optimise away lots of lstat() calls

[Kjetil Barvik]

Changes since version 6:

a) The patch-split was not correct for patch 3/3.  In v6 it could look
   like the lines:

-	char path[PATH_MAX];
+	char path[PATH_MAX + 1];

   was a random bugfix.  In this version for patch 3/5 we also change
   "&& last_slash < PATH_MAX" into "&& last_slash <= PATH_MAX" (and
   similar for the else-if part).  This could also make the intro-
   duction of the 'FL_FULLPATH' flag more readable for this patch.

b) Cleanup and added 3 comments to 'greatest_match_lstat_cache()'

c) Introduction of the 'invalidate_lstat_cache()' function.  How does
   the interface look?  good?  bad?  Does the function do what people
   expect it to do?

d) Reintroduction of the 'clear_lstat_cache()' function.

Junio, I hope it is possible to use patches 1/5, 2/5 and 3/5 from this
version instead of 1/3, 2/3 and 3/3 from version 6, for the possible
future in origin/pu?  See also a) above.  Thanks in advance!

In general, are we allowed to redesign the patch-series while the
patches is inside origin/pu?


Kjetil Barvik (5):
  lstat_cache(): more cache effective symlink/directory detection
  lstat_cache(): introduce has_symlink_or_noent_leading_path() function
  lstat_cache(): introduce has_dirs_only_path() function
  lstat_cache(): introduce invalidate_lstat_cache() function
  lstat_cache(): introduce clear_lstat_cache() function

 cache.h        |    4 +
 entry.c        |   34 +++-----
 symlinks.c     |  249 ++++++++++++++++++++++++++++++++++++++++++++++----------
 unpack-trees.c |    4 +-
 4 files changed, 222 insertions(+), 69 deletions(-)

[PATCH/RFC v7 1/5] lstat_cache(): more cache effective symlink/directory detection

[Kjetil Barvik]

Make the cache functionality more effective.  Previously when A/B/C/D
was in the cache and A/B/C/E/file.c was called for, there was no match
at all from the cache.  Now we use the fact that the paths "A", "A/B"
and "A/B/C" are already tested, and we only need to do an lstat() call
on "A/B/C/E".

We only cache/store the last path regardless of its type.  Since the
cache functionality is always used with alphabetically sorted names
(at least it seems so for me), there is no need to store both the last
symlink-leading path and the last real-directory path.  Note that if
the cache is not called with (mostly) alphabetically sorted names,
neither the old, nor this new one, would be very effective.

Previously, when symlink A/B/C/S was cached/stored in the symlink-
leading path, and A/B/C/file.c was called for, it was not easy to use
the fact that we already knew that the paths "A", "A/B" and "A/B/C"
are real directories.

Avoid copying the first path components of the name 2 zillion times
when we test new path components.  Since we always cache/store the
last path, we can copy each component as we test those directly into
the cache.  Previously we ended up doing a memcpy() for the full
path/name right before each lstat() call, and when updating the cache
for each time we have tested a new path component.

We also use less memory, that is, PATH_MAX bytes less memory on the
stack and PATH_MAX bytes less memory on the heap.

Thanks to Junio C Hamano, Linus Torvalds and Rene Scharfe for valuable
comments to this patch!

Signed-off-by: Kjetil Barvik <barvik@broadpark.no>
---
 symlinks.c |  157 ++++++++++++++++++++++++++++++++++++++++++++---------------
 1 files changed, 117 insertions(+), 40 deletions(-)

diff --git a/symlinks.c b/symlinks.c
index 5a5e781..b17e4b3 100644
--- a/symlinks.c
+++ b/symlinks.c
@@ -1,64 +1,141 @@
 #include "cache.h"
 
-struct pathname {
-	int len;
+static struct cache_def {
 	char path[PATH_MAX];
-};
+	int len;
+	int flags;
+} cache;
 
-/* Return matching pathname prefix length, or zero if not matching */
-static inline int match_pathname(int len, const char *name, struct pathname *match)
+/* Returns the length (on a path component basis) of the longest
+ * common prefix match of 'name' and the cached path string.
+ */
+static inline int greatest_match_lstat_cache(int len, const char *name)
 {
-	int match_len = match->len;
-	return (len > match_len &&
-		name[match_len] == '/' &&
-		!memcmp(name, match->path, match_len)) ? match_len : 0;
+	int max_len, match_len = 0, i = 0;
+
+	max_len = len < cache.len ? len : cache.len;
+	while (i < max_len && name[i] == cache.path[i]) {
+		if (name[i] == '/')
+			match_len = i;
+		i++;
+	}
+	/* Is the cached path string a substring of 'name'? */
+	if (i == cache.len && len > cache.len && name[cache.len] == '/')
+		match_len = cache.len;
+	/* Is 'name' a substring of the cached path string? */
+	else if ((i == len && len < cache.len && cache.path[len] == '/') ||
+		 (i == len && len == cache.len))
+		match_len = len;
+	return match_len;
 }
 
-static inline void set_pathname(int len, const char *name, struct pathname *match)
+static inline void reset_lstat_cache(void)
 {
-	if (len < PATH_MAX) {
-		match->len = len;
-		memcpy(match->path, name, len);
-		match->path[len] = 0;
-	}
+	cache.path[0] = '\0';
+	cache.len   = 0;
+	cache.flags = 0;
 }
 
-int has_symlink_leading_path(int len, const char *name)
+#define FL_DIR      (1 << 0)
+#define FL_SYMLINK  (1 << 1)
+#define FL_LSTATERR (1 << 2)
+#define FL_ERR      (1 << 3)
+
+/*
+ * Check if name 'name' of length 'len' has a symlink leading
+ * component, or if the directory exists and is real.
+ *
+ * To speed up the check, some information is allowed to be cached.
+ * This is can be indicated by the 'track_flags' argument.
+ */
+static int lstat_cache(int len, const char *name,
+		       int track_flags)
 {
-	static struct pathname link, nonlink;
-	char path[PATH_MAX];
+	int match_len, last_slash, last_slash_dir;
+	int match_flags, ret_flags, save_flags, max_len;
 	struct stat st;
-	char *sp;
-	int known_dir;
 
 	/*
-	 * See if the last known symlink cache matches.
+	 * Check to see if we have a match from the cache for the
+	 * symlink path type.
 	 */
-	if (match_pathname(len, name, &link))
-		return 1;
-
+	match_len = last_slash = greatest_match_lstat_cache(len, name);
+	match_flags = cache.flags & track_flags & FL_SYMLINK;
+	if (match_flags && match_len == cache.len)
+		return match_flags;
 	/*
-	 * Get rid of the last known directory part
+	 * If 'name' is a substring of the cache on a path component
+	 * basis, and a directory is cached, we can return
+	 * immediately.
 	 */
-	known_dir = match_pathname(len, name, &nonlink);
+	match_flags = cache.flags & track_flags & FL_DIR;
+	if (match_flags && match_len == len)
+		return match_flags;
 
-	while ((sp = strchr(name + known_dir + 1, '/')) != NULL) {
-		int thislen = sp - name ;
-		memcpy(path, name, thislen);
-		path[thislen] = 0;
+	/* Okay, no match from the cache so far, so now we have to
+	 * check the rest of the path components.
+	 */
+	ret_flags = FL_DIR;
+	last_slash_dir = last_slash;
+	max_len = len < PATH_MAX ? len : PATH_MAX;
+	while (match_len < max_len) {
+		do {
+			cache.path[match_len] = name[match_len];
+			match_len++;
+		} while (match_len < max_len && name[match_len] != '/');
+		if (match_len >= max_len)
+			break;
+		last_slash = match_len;
+		cache.path[last_slash] = '\0';
 
-		if (lstat(path, &st))
-			return 0;
-		if (S_ISDIR(st.st_mode)) {
-			set_pathname(thislen, path, &nonlink);
-			known_dir = thislen;
+		if (lstat(cache.path, &st)) {
+			ret_flags = FL_LSTATERR;
+		} else if (S_ISDIR(st.st_mode)) {
+			last_slash_dir = last_slash;
 			continue;
-		}
-		if (S_ISLNK(st.st_mode)) {
-			set_pathname(thislen, path, &link);
-			return 1;
+		} else if (S_ISLNK(st.st_mode)) {
+			ret_flags = FL_SYMLINK;
+		} else {
+			ret_flags = FL_ERR;
 		}
 		break;
 	}
-	return 0;
+
+	/* At the end update the cache.  Note that max 2 different
+	 * path types, FL_SYMLINK and FL_DIR, can be cached for the
+	 * moment!
+	 */
+	save_flags = ret_flags & track_flags & FL_SYMLINK;
+	if (save_flags && last_slash > 0 && last_slash < PATH_MAX) {
+		cache.path[last_slash] = '\0';
+		cache.len   = last_slash;
+		cache.flags = save_flags;
+	} else if (track_flags & FL_DIR &&
+		   last_slash_dir > 0 && last_slash_dir < PATH_MAX) {
+		/* We have a separate test for the directory case,
+		 * since it could be that we have found a symlink and
+		 * the track_flags says that we can not cache this
+		 * fact, so the cache would then have been left empty
+		 * in this case.
+		 *
+		 * But, if we is allowed to track real directories, we
+		 * can still cache the path components before the last
+		 * one (the found symlink component).
+		 */
+		cache.path[last_slash_dir] = '\0';
+		cache.len   = last_slash_dir;
+		cache.flags = FL_DIR;
+	} else {
+		reset_lstat_cache();
+	}
+	return ret_flags;
+}
+
+/* Return non-zero if path 'name' has a leading symlink component.
+ */
+int has_symlink_leading_path(int len, const char *name)
+{
+	return lstat_cache(len, name,
+			   FL_SYMLINK|FL_DIR) &
+		FL_SYMLINK;
 }
-- 
1.6.0.2.GIT

[PATCH/RFC v7 2/5] lstat_cache(): introduce has_symlink_or_noent_leading_path() function

[Kjetil Barvik]

In some cases, especially inside the unpack-trees.c file, and inside
the verify_absent() function, we can avoid some unnecessary calls to
lstat(), if the lstat_cache() function can also be told to keep track
of none existing directories.

So we update the lstat_cache() function to handle this new fact,
introduce a new wrapper function, and the result is that we save lots
of lstat() calls for a removed directory which previously contained
lots of files, when we call this new wrapper of lstat_cache() instead
of the old one.

We do similar changes inside the unlink_entry() function, since if we
can already say that the leading directory component of a pathname
does not exist, it is not necessary to try to remove a pathname below
it!

Thanks to Junio C Hamano, Linus Torvalds and Rene Scharfe for valuable
comments to this patch!

Signed-off-by: Kjetil Barvik <barvik@broadpark.no>
---
 cache.h        |    1 +
 symlinks.c     |   87 ++++++++++++++++++++++++++++++++++++--------------------
 unpack-trees.c |    4 +-
 3 files changed, 59 insertions(+), 33 deletions(-)

diff --git a/cache.h b/cache.h
index 231c06d..11181aa 100644
--- a/cache.h
+++ b/cache.h
@@ -720,6 +720,7 @@ struct checkout {
 
 extern int checkout_entry(struct cache_entry *ce, const struct checkout *state, char *topath);
 extern int has_symlink_leading_path(int len, const char *name);
+extern int has_symlink_or_noent_leading_path(int len, const char *name);
 
 extern struct alternate_object_database {
 	struct alternate_object_database *next;
diff --git a/symlinks.c b/symlinks.c
index b17e4b3..52b3b0e 100644
--- a/symlinks.c
+++ b/symlinks.c
@@ -4,6 +4,7 @@ static struct cache_def {
 	char path[PATH_MAX];
 	int len;
 	int flags;
+	int track_flags;
 } cache;
 
 /* Returns the length (on a path component basis) of the longest
@@ -29,21 +30,23 @@ static inline int greatest_match_lstat_cache(int len, const char *name)
 	return match_len;
 }
 
-static inline void reset_lstat_cache(void)
+static inline void reset_lstat_cache(int track_flags)
 {
 	cache.path[0] = '\0';
 	cache.len   = 0;
 	cache.flags = 0;
+	cache.track_flags = track_flags;
 }
 
 #define FL_DIR      (1 << 0)
-#define FL_SYMLINK  (1 << 1)
-#define FL_LSTATERR (1 << 2)
-#define FL_ERR      (1 << 3)
+#define FL_NOENT    (1 << 1)
+#define FL_SYMLINK  (1 << 2)
+#define FL_LSTATERR (1 << 3)
+#define FL_ERR      (1 << 4)
 
 /*
  * Check if name 'name' of length 'len' has a symlink leading
- * component, or if the directory exists and is real.
+ * component, or if the directory exists and is real, or not.
  *
  * To speed up the check, some information is allowed to be cached.
  * This is can be indicated by the 'track_flags' argument.
@@ -55,22 +58,32 @@ static int lstat_cache(int len, const char *name,
 	int match_flags, ret_flags, save_flags, max_len;
 	struct stat st;
 
-	/*
-	 * Check to see if we have a match from the cache for the
-	 * symlink path type.
-	 */
-	match_len = last_slash = greatest_match_lstat_cache(len, name);
-	match_flags = cache.flags & track_flags & FL_SYMLINK;
-	if (match_flags && match_len == cache.len)
-		return match_flags;
-	/*
-	 * If 'name' is a substring of the cache on a path component
-	 * basis, and a directory is cached, we can return
-	 * immediately.
-	 */
-	match_flags = cache.flags & track_flags & FL_DIR;
-	if (match_flags && match_len == len)
-		return match_flags;
+	if (cache.track_flags != track_flags) {
+		/*
+		 * As a safeguard we clear the cache if the value of
+		 * track_flags does not match with the last supplied
+		 * value.
+		 */
+		reset_lstat_cache(track_flags);
+		match_len = last_slash = 0;
+	} else {
+		/*
+		 * Check to see if we have a match from the cache for
+		 * the 2 "excluding" path types.
+		 */
+		match_len = last_slash = greatest_match_lstat_cache(len, name);
+		match_flags = cache.flags & track_flags & (FL_NOENT|FL_SYMLINK);
+		if (match_flags && match_len == cache.len)
+			return match_flags;
+		/*
+		 * If 'name' is a substring of the cache on a path
+		 * component basis, and a directory is cached, we
+		 * can return immediately.
+		 */
+		match_flags = cache.flags & track_flags & FL_DIR;
+		if (match_flags && match_len == len)
+			return match_flags;
+	}
 
 	/* Okay, no match from the cache so far, so now we have to
 	 * check the rest of the path components.
@@ -90,6 +103,8 @@ static int lstat_cache(int len, const char *name,
 
 		if (lstat(cache.path, &st)) {
 			ret_flags = FL_LSTATERR;
+			if (errno == ENOENT)
+				ret_flags |= FL_NOENT;
 		} else if (S_ISDIR(st.st_mode)) {
 			last_slash_dir = last_slash;
 			continue;
@@ -101,11 +116,11 @@ static int lstat_cache(int len, const char *name,
 		break;
 	}
 
-	/* At the end update the cache.  Note that max 2 different
-	 * path types, FL_SYMLINK and FL_DIR, can be cached for the
-	 * moment!
+	/* At the end update the cache.  Note that max 3 different
+	 * path types, FL_NOENT, FL_SYMLINK and FL_DIR, can be cached
+	 * for the moment!
 	 */
-	save_flags = ret_flags & track_flags & FL_SYMLINK;
+	save_flags = ret_flags & track_flags & (FL_NOENT|FL_SYMLINK);
 	if (save_flags && last_slash > 0 && last_slash < PATH_MAX) {
 		cache.path[last_slash] = '\0';
 		cache.len   = last_slash;
@@ -113,20 +128,20 @@ static int lstat_cache(int len, const char *name,
 	} else if (track_flags & FL_DIR &&
 		   last_slash_dir > 0 && last_slash_dir < PATH_MAX) {
 		/* We have a separate test for the directory case,
-		 * since it could be that we have found a symlink and
-		 * the track_flags says that we can not cache this
-		 * fact, so the cache would then have been left empty
-		 * in this case.
+		 * since it could be that we have found a symlink or a
+		 * none existing directory and the track_flags says
+		 * that we can not cache this fact, so the cache would
+		 * then have been left empty in this case.
 		 *
 		 * But, if we is allowed to track real directories, we
 		 * can still cache the path components before the last
-		 * one (the found symlink component).
+		 * one (the found symlink or none existing component).
 		 */
 		cache.path[last_slash_dir] = '\0';
 		cache.len   = last_slash_dir;
 		cache.flags = FL_DIR;
 	} else {
-		reset_lstat_cache();
+		reset_lstat_cache(track_flags);
 	}
 	return ret_flags;
 }
@@ -139,3 +154,13 @@ int has_symlink_leading_path(int len, const char *name)
 			   FL_SYMLINK|FL_DIR) &
 		FL_SYMLINK;
 }
+
+/* Return non-zero if path 'name' has a leading symlink component or
+ * if some leading path component does not exists.
+ */
+int has_symlink_or_noent_leading_path(int len, const char *name)
+{
+	return lstat_cache(len, name,
+			   FL_SYMLINK|FL_NOENT|FL_DIR) &
+		(FL_SYMLINK|FL_NOENT);
+}
diff --git a/unpack-trees.c b/unpack-trees.c
index 54f301d..a3fd383 100644
--- a/unpack-trees.c
+++ b/unpack-trees.c
@@ -61,7 +61,7 @@ static void unlink_entry(struct cache_entry *ce)
 	char *cp, *prev;
 	char *name = ce->name;
 
-	if (has_symlink_leading_path(ce_namelen(ce), ce->name))
+	if (has_symlink_or_noent_leading_path(ce_namelen(ce), ce->name))
 		return;
 	if (unlink(name))
 		return;
@@ -584,7 +584,7 @@ static int verify_absent(struct cache_entry *ce, const char *action,
 	if (o->index_only || o->reset || !o->update)
 		return 0;
 
-	if (has_symlink_leading_path(ce_namelen(ce), ce->name))
+	if (has_symlink_or_noent_leading_path(ce_namelen(ce), ce->name))
 		return 0;
 
 	if (!lstat(ce->name, &st)) {
-- 
1.6.0.2.GIT

[PATCH/RFC v7 3/5] lstat_cache(): introduce has_dirs_only_path() function

[Kjetil Barvik]

Inside the create_directories() function inside entry.c we currently
do an stat() or lstat() call for each path component of the pathname
'path' each and every time called.  For the 'git checkout' command,
this function is called on each file for which we must do an update
(ce->ce_flags & CE_UPDATE), so we get lots and lots of calls...

To fix this, we make a new wrapper to the lstat_cache() function, and
call the wrapper function instead of the calls to the stat() or the
lstat() functions.  Since the paths given to the create_directories()
function, is sorted alphabetically, the new wrapper would be very
cache effective in this situation.

To support it we must update the lstat_cache() function to be able to
say that "pleace test the complete length of 'name'", and also to give
it the length of a prefix, where the cache should use the stat()
function instead of the lstat() function to test each path component.

Thanks to Junio C Hamano, Linus Torvalds and Rene Scharfe for valuable
comments to this patch!

Signed-off-by: Kjetil Barvik <barvik@broadpark.no>
---
 cache.h    |    1 +
 entry.c    |   34 ++++++++++---------------------
 symlinks.c |   63 +++++++++++++++++++++++++++++++++++++++++++----------------
 3 files changed, 58 insertions(+), 40 deletions(-)

diff --git a/cache.h b/cache.h
index 11181aa..7c8c8e4 100644
--- a/cache.h
+++ b/cache.h
@@ -721,6 +721,7 @@ struct checkout {
 extern int checkout_entry(struct cache_entry *ce, const struct checkout *state, char *topath);
 extern int has_symlink_leading_path(int len, const char *name);
 extern int has_symlink_or_noent_leading_path(int len, const char *name);
+extern int has_dirs_only_path(int len, const char *name, int prefix_len);
 
 extern struct alternate_object_database {
 	struct alternate_object_database *next;
diff --git a/entry.c b/entry.c
index aa2ee46..b0295bd 100644
--- a/entry.c
+++ b/entry.c
@@ -8,35 +8,23 @@ static void create_directories(const char *path, const struct checkout *state)
 	const char *slash = path;
 
 	while ((slash = strchr(slash+1, '/')) != NULL) {
-		struct stat st;
-		int stat_status;
-
 		len = slash - path;
 		memcpy(buf, path, len);
 		buf[len] = 0;
 
-		if (len <= state->base_dir_len)
-			/*
-			 * checkout-index --prefix=<dir>; <dir> is
-			 * allowed to be a symlink to an existing
-			 * directory.
-			 */
-			stat_status = stat(buf, &st);
-		else
-			/*
-			 * if there currently is a symlink, we would
-			 * want to replace it with a real directory.
-			 */
-			stat_status = lstat(buf, &st);
-
-		if (!stat_status && S_ISDIR(st.st_mode))
+		/* For 'checkout-index --prefix=<dir>', <dir> is
+		 * allowed to be a symlink to an existing directory,
+		 * and we set 'state->base_dir_len' below, such that
+		 * we test the path components of the prefix with the
+		 * stat() function instead of the lstat() function.
+		 */
+		if (has_dirs_only_path(len, buf, state->base_dir_len))
 			continue; /* ok, it is already a directory. */
 
-		/*
-		 * We know stat_status == 0 means something exists
-		 * there and this mkdir would fail, but that is an
-		 * error codepath; we do not care, as we unlink and
-		 * mkdir again in such a case.
+		/* If this mkdir() would fail, it could be that there
+		 * is already a symlink or something else exists
+		 * there, therefore we then try to unlink it and try
+		 * one more time to create the directory.
 		 */
 		if (mkdir(buf, 0777)) {
 			if (errno == EEXIST && state->force &&
diff --git a/symlinks.c b/symlinks.c
index 52b3b0e..f9d1821 100644
--- a/symlinks.c
+++ b/symlinks.c
@@ -1,10 +1,11 @@
 #include "cache.h"
 
 static struct cache_def {
-	char path[PATH_MAX];
+	char path[PATH_MAX + 1];
 	int len;
 	int flags;
 	int track_flags;
+	int prefix_len_stat_func;
 } cache;
 
 /* Returns the length (on a path component basis) of the longest
@@ -30,12 +31,13 @@ static inline int greatest_match_lstat_cache(int len, const char *name)
 	return match_len;
 }
 
-static inline void reset_lstat_cache(int track_flags)
+static inline void reset_lstat_cache(int track_flags, int prefix_len_stat_func)
 {
 	cache.path[0] = '\0';
 	cache.len   = 0;
 	cache.flags = 0;
 	cache.track_flags = track_flags;
+	cache.prefix_len_stat_func = prefix_len_stat_func;
 }
 
 #define FL_DIR      (1 << 0)
@@ -43,28 +45,35 @@ static inline void reset_lstat_cache(int track_flags)
 #define FL_SYMLINK  (1 << 2)
 #define FL_LSTATERR (1 << 3)
 #define FL_ERR      (1 << 4)
+#define FL_FULLPATH (1 << 5)
 
 /*
  * Check if name 'name' of length 'len' has a symlink leading
  * component, or if the directory exists and is real, or not.
  *
  * To speed up the check, some information is allowed to be cached.
- * This is can be indicated by the 'track_flags' argument.
+ * This is can be indicated by the 'track_flags' argument, which also
+ * can be used to indicate that we should always check the full path.
+ *
+ * The 'prefix_len_stat_func' parameter can be used to set the length
+ * of the prefix, where the cache should use the stat() function
+ * instead of the lstat() function to test each path component.
  */
 static int lstat_cache(int len, const char *name,
-		       int track_flags)
+		       int track_flags, int prefix_len_stat_func)
 {
 	int match_len, last_slash, last_slash_dir;
-	int match_flags, ret_flags, save_flags, max_len;
+	int match_flags, ret_flags, save_flags, max_len, ret;
 	struct stat st;
 
-	if (cache.track_flags != track_flags) {
+	if (cache.track_flags != track_flags ||
+	    cache.prefix_len_stat_func != prefix_len_stat_func) {
 		/*
-		 * As a safeguard we clear the cache if the value of
-		 * track_flags does not match with the last supplied
-		 * value.
+		 * As a safeguard we clear the cache if the values of
+		 * track_flags and/or prefix_len_stat_func does not
+		 * match with the last supplied values.
 		 */
-		reset_lstat_cache(track_flags);
+		reset_lstat_cache(track_flags, prefix_len_stat_func);
 		match_len = last_slash = 0;
 	} else {
 		/*
@@ -96,12 +105,17 @@ static int lstat_cache(int len, const char *name,
 			cache.path[match_len] = name[match_len];
 			match_len++;
 		} while (match_len < max_len && name[match_len] != '/');
-		if (match_len >= max_len)
+		if (match_len >= max_len && !(track_flags & FL_FULLPATH))
 			break;
 		last_slash = match_len;
 		cache.path[last_slash] = '\0';
 
-		if (lstat(cache.path, &st)) {
+		if (last_slash <= prefix_len_stat_func)
+			ret = stat(cache.path, &st);
+		else
+			ret = lstat(cache.path, &st);
+
+		if (ret) {
 			ret_flags = FL_LSTATERR;
 			if (errno == ENOENT)
 				ret_flags |= FL_NOENT;
@@ -121,12 +135,12 @@ static int lstat_cache(int len, const char *name,
 	 * for the moment!
 	 */
 	save_flags = ret_flags & track_flags & (FL_NOENT|FL_SYMLINK);
-	if (save_flags && last_slash > 0 && last_slash < PATH_MAX) {
+	if (save_flags && last_slash > 0 && last_slash <= PATH_MAX) {
 		cache.path[last_slash] = '\0';
 		cache.len   = last_slash;
 		cache.flags = save_flags;
 	} else if (track_flags & FL_DIR &&
-		   last_slash_dir > 0 && last_slash_dir < PATH_MAX) {
+		   last_slash_dir > 0 && last_slash_dir <= PATH_MAX) {
 		/* We have a separate test for the directory case,
 		 * since it could be that we have found a symlink or a
 		 * none existing directory and the track_flags says
@@ -141,17 +155,19 @@ static int lstat_cache(int len, const char *name,
 		cache.len   = last_slash_dir;
 		cache.flags = FL_DIR;
 	} else {
-		reset_lstat_cache(track_flags);
+		reset_lstat_cache(track_flags, prefix_len_stat_func);
 	}
 	return ret_flags;
 }
 
+#define USE_ONLY_LSTAT  0
+
 /* Return non-zero if path 'name' has a leading symlink component.
  */
 int has_symlink_leading_path(int len, const char *name)
 {
 	return lstat_cache(len, name,
-			   FL_SYMLINK|FL_DIR) &
+			   FL_SYMLINK|FL_DIR, USE_ONLY_LSTAT) &
 		FL_SYMLINK;
 }
 
@@ -161,6 +177,19 @@ int has_symlink_leading_path(int len, const char *name)
 int has_symlink_or_noent_leading_path(int len, const char *name)
 {
 	return lstat_cache(len, name,
-			   FL_SYMLINK|FL_NOENT|FL_DIR) &
+			   FL_SYMLINK|FL_NOENT|FL_DIR, USE_ONLY_LSTAT) &
 		(FL_SYMLINK|FL_NOENT);
 }
+
+/* Return non-zero if all path components of 'name' exists as a
+ * directory.  If prefix_len > 0, we will test with the stat()
+ * function instead of the lstat() function for a prefix length of
+ * 'prefix_len', thus we then allow for symlink(s) in the prefix part
+ * as long as those points to real existing directories.
+ */
+int has_dirs_only_path(int len, const char *name, int prefix_len)
+{
+	return lstat_cache(len, name,
+			   FL_DIR|FL_FULLPATH, prefix_len) &
+		FL_DIR;
+}
-- 
1.6.0.2.GIT

[PATCH/RFC v7 4/5] lstat_cache(): introduce invalidate_lstat_cache() function

[Kjetil Barvik]

In some cases it could maybe be necessary to say to the cache that
"Hey, I deleted this directory and if you currently has it inside your
cache, you should deleted it".  This patch introduce a function which
support this.

Signed-off-by: Kjetil Barvik <barvik@broadpark.no>
---
 cache.h    |    1 +
 symlinks.c |   39 +++++++++++++++++++++++++++++++--------
 2 files changed, 32 insertions(+), 8 deletions(-)

diff --git a/cache.h b/cache.h
index 7c8c8e4..f4452a8 100644
--- a/cache.h
+++ b/cache.h
@@ -722,6 +722,7 @@ extern int checkout_entry(struct cache_entry *ce, const struct checkout *state,
 extern int has_symlink_leading_path(int len, const char *name);
 extern int has_symlink_or_noent_leading_path(int len, const char *name);
 extern int has_dirs_only_path(int len, const char *name, int prefix_len);
+extern void invalidate_lstat_cache(int len, const char *name);
 
 extern struct alternate_object_database {
 	struct alternate_object_database *next;
diff --git a/symlinks.c b/symlinks.c
index f9d1821..3b1c3da 100644
--- a/symlinks.c
+++ b/symlinks.c
@@ -11,23 +11,30 @@ static struct cache_def {
 /* Returns the length (on a path component basis) of the longest
  * common prefix match of 'name' and the cached path string.
  */
-static inline int greatest_match_lstat_cache(int len, const char *name)
+static inline int greatest_match_lstat_cache(int len, const char *name,
+					     int *previous_slash)
 {
-	int max_len, match_len = 0, i = 0;
+	int max_len, match_len = 0, match_len_prev = 0, i = 0;
 
 	max_len = len < cache.len ? len : cache.len;
 	while (i < max_len && name[i] == cache.path[i]) {
-		if (name[i] == '/')
+		if (name[i] == '/') {
+			match_len_prev = match_len;
 			match_len = i;
+		}
 		i++;
 	}
 	/* Is the cached path string a substring of 'name'? */
-	if (i == cache.len && len > cache.len && name[cache.len] == '/')
+	if (i == cache.len && len > cache.len && name[cache.len] == '/') {
+		match_len_prev = match_len;
 		match_len = cache.len;
 	/* Is 'name' a substring of the cached path string? */
-	else if ((i == len && len < cache.len && cache.path[len] == '/') ||
-		 (i == len && len == cache.len))
+	} else if ((i == len && len < cache.len && cache.path[len] == '/') ||
+		   (i == len && len == cache.len)) {
+		match_len_prev = match_len;
 		match_len = len;
+	}
+	*previous_slash = match_len_prev;
 	return match_len;
 }
 
@@ -62,7 +69,7 @@ static inline void reset_lstat_cache(int track_flags, int prefix_len_stat_func)
 static int lstat_cache(int len, const char *name,
 		       int track_flags, int prefix_len_stat_func)
 {
-	int match_len, last_slash, last_slash_dir;
+	int match_len, last_slash, last_slash_dir, previous_slash;
 	int match_flags, ret_flags, save_flags, max_len, ret;
 	struct stat st;
 
@@ -80,7 +87,8 @@ static int lstat_cache(int len, const char *name,
 		 * Check to see if we have a match from the cache for
 		 * the 2 "excluding" path types.
 		 */
-		match_len = last_slash = greatest_match_lstat_cache(len, name);
+		match_len = last_slash =
+			greatest_match_lstat_cache(len, name, &previous_slash);
 		match_flags = cache.flags & track_flags & (FL_NOENT|FL_SYMLINK);
 		if (match_flags && match_len == cache.len)
 			return match_flags;
@@ -160,6 +168,21 @@ static int lstat_cache(int len, const char *name,
 	return ret_flags;
 }
 
+/* Invalidate the given 'name' from the cache, if 'name' matches
+ * completely with the cache.
+ */
+void invalidate_lstat_cache(int len, const char *name)
+{
+	int match_len, previous_slash;
+
+	match_len = greatest_match_lstat_cache(len, name, &previous_slash);
+	if (match_len == len) {
+		cache.path[previous_slash] = '\0';
+		cache.len   = previous_slash;
+		cache.flags = previous_slash ? FL_DIR : 0;
+	}
+}
+
 #define USE_ONLY_LSTAT  0
 
 /* Return non-zero if path 'name' has a leading symlink component.
-- 
1.6.0.2.GIT

[PATCH/RFC v7 5/5] lstat_cache(): introduce clear_lstat_cache() function

[Kjetil Barvik]

If you want to completely clear the contents of the lstat_cache(), then
call this new function.

Signed-off-by: Kjetil Barvik <barvik@broadpark.no>
---
 cache.h    |    1 +
 symlinks.c |    7 +++++++
 2 files changed, 8 insertions(+), 0 deletions(-)

diff --git a/cache.h b/cache.h
index f4452a8..468daf6 100644
--- a/cache.h
+++ b/cache.h
@@ -723,6 +723,7 @@ extern int has_symlink_leading_path(int len, const char *name);
 extern int has_symlink_or_noent_leading_path(int len, const char *name);
 extern int has_dirs_only_path(int len, const char *name, int prefix_len);
 extern void invalidate_lstat_cache(int len, const char *name);
+extern void clear_lstat_cache(void);
 
 extern struct alternate_object_database {
 	struct alternate_object_database *next;
diff --git a/symlinks.c b/symlinks.c
index 3b1c3da..7a634cb 100644
--- a/symlinks.c
+++ b/symlinks.c
@@ -183,6 +183,13 @@ void invalidate_lstat_cache(int len, const char *name)
 	}
 }
 
+/* Completely clear the contents of the cache!
+ */
+void clear_lstat_cache(void)
+{
+	reset_lstat_cache(0, 0);
+}
+
 #define USE_ONLY_LSTAT  0
 
 /* Return non-zero if path 'name' has a leading symlink component.
-- 
1.6.0.2.GIT

Re: [PATCH/RFC v7 0/5] git checkout: optimise away lots of lstat() calls

[Junio C Hamano]

Kjetil Barvik <barvik@broadpark.no> writes:

> Junio, I hope it is possible to use patches 1/5, 2/5 and 3/5 from this
> version instead of 1/3, 2/3 and 3/3 from version 6, for the possible
> future in origin/pu?  See also a) above.  Thanks in advance!
>
> In general, are we allowed to redesign the patch-series while the
> patches is inside origin/pu?

You won't be too far off if you thought of 'pu' branch nothing more than
my bookmarks into the list archive.  They are interesting enough to keep
an eye on their evolution but not yet stable enough to go into initial
round of wider testing and incremental development which starts to happen
when they hit 'next'.

So yes, while you are not yet confident that the series does not need
major redesign, please keep sending replacements (not incremental
updates), making it clear that you still have reservations and I'll keep
them in 'pu', until the initial barrage of "Oh, I think this way is
better", "The previous review comments were valuable, and this round
contains all of them" is over and things seem to stabilize into a testable
shape.

Personally I felt your v5 and later ones were already 'next' material, and
the only reason why the series haven't landed on 'next' was because you
kept sending updates before I sat down in the evening and said "Ok, this
looks good---let's merge it to 'next'".  Instead I ended up saying "Hmph,
I thought it looked good already but the author seems not satisfied yet;
there is another round of update to replace the previous one.  Let's queue
it in 'pu', as it may be updated again, and I have to also look at series
from other people.' ;-)