From 0c46863384e9da3746b90ddf81eef6d25d475e5c Mon Sep 17 00:00:00 2001
From: Russell Belfer <rb@github.com>
Date: Thu, 14 Mar 2013 13:40:15 -0700
Subject: [PATCH 1/3] Improved tree iterator internals

This updates the tree iterator internals to be more efficient.

The tree_iterator_entry objects are now kept as pointers that are
allocated from a git_pool, so that we may use git__tsort_r for
sorting (which is better than qsort, given that the tree is
likely mostly ordered already).

Those tree_iterator_entry objects now keep direct pointers to the
data they refer to instead of keeping indirect index values.  This
simplifies a lot of the data structure traversal code.

This also adds bsearch to find the start item position for range-
limited tree iterators, and is more explicit about using
git_path_cmp instead of reimplementing it.  The git_path_cmp
changed a bit to make it easier for tree_iterators to use it (but
it was barely being used previously, so not a big deal).

This adds a git_pool_free_array function that efficiently frees a
list of pool allocated pointers (which the tree_iterator keeps).
Also, added new tests for the git_pool free list functionality
that was not previously being tested (or used).
---
 src/iterator.c         | 350 ++++++++++++++++++-----------------------
 src/path.c             |  29 +---
 src/path.h             |   8 +-
 src/pool.c             |  24 ++-
 src/pool.h             |   7 +
 src/tree.c             |  23 +--
 tests-clar/core/pool.c |  50 ++++++
 7 files changed, 251 insertions(+), 240 deletions(-)

diff --git a/src/iterator.c b/src/iterator.c
index 1ac6a4919..b15bcedd8 100644
--- a/src/iterator.c
+++ b/src/iterator.c
@@ -148,8 +148,7 @@ int git_iterator_for_nothing(
 	const char *start,
 	const char *end)
 {
-	empty_iterator *i = git__calloc(1, sizeof(empty_iterator));
-	GITERR_CHECK_ALLOC(i);
+	empty_iterator *i;
 
 #define empty_iterator__current empty_iterator__noop
 #define empty_iterator__advance empty_iterator__noop
@@ -165,15 +164,16 @@ int git_iterator_for_nothing(
 }
 
 
-typedef struct {
-	size_t parent_entry_index;  /* index in parent entries array */
-	size_t parent_tree_index;   /* index in parent entry tree */
-	git_tree *tree; /* this tree if this is tree (only valid while current) */
-} tree_iterator_entry;
+typedef struct tree_iterator_entry tree_iterator_entry;
+struct tree_iterator_entry {
+	tree_iterator_entry *parent;
+	const git_tree_entry *te;
+	git_tree *tree;
+};
 
 typedef struct tree_iterator_frame tree_iterator_frame;
 struct tree_iterator_frame {
-	tree_iterator_frame *parent, *child;
+	tree_iterator_frame *up, *down;
 
 	size_t n_entries; /* items in this frame */
 	size_t current;   /* start of currently active range in frame */
@@ -182,13 +182,14 @@ struct tree_iterator_frame {
 	const char *start;
 	size_t startlen;
 
-	tree_iterator_entry entries[GIT_FLEX_ARRAY];
+	tree_iterator_entry *entries[GIT_FLEX_ARRAY];
 };
 
 typedef struct {
 	git_iterator base;
 	git_iterator_callbacks cb;
-	tree_iterator_frame *head, *top;
+	tree_iterator_frame *head, *root;
+	git_pool pool;
 	git_index_entry entry;
 	git_buf path;
 	int path_ambiguities;
@@ -196,30 +197,6 @@ typedef struct {
 	int (*strncomp)(const char *a, const char *b, size_t sz);
 } tree_iterator;
 
-static const git_tree_entry *tree_iterator__tree_entry(
-	tree_iterator_frame *tf, const tree_iterator_entry *entry)
-{
-	git_tree *tree = tf->parent->entries[entry->parent_entry_index].tree;
-	if (!tree)
-		return NULL;
-	return git_tree_entry_byindex(tree, entry->parent_tree_index);
-}
-
-static const git_tree_entry *tree_iterator__tree_entry_by_index(
-	tree_iterator_frame *tf, size_t i)
-{
-	git_tree *tree;
-
-	if (i >= tf->n_entries)
-		return NULL;
-
-	tree = tf->parent->entries[tf->entries[i].parent_entry_index].tree;
-	if (!tree)
-		return NULL;
-
-	return git_tree_entry_byindex(tree, tf->entries[i].parent_tree_index);
-}
-
 static char *tree_iterator__current_filename(
 	tree_iterator *ti, const git_tree_entry *te)
 {
@@ -238,90 +215,76 @@ static char *tree_iterator__current_filename(
 
 static void tree_iterator__rewrite_filename(tree_iterator *ti)
 {
-	tree_iterator_frame *scan = ti->head;
-	size_t current = scan->current;
+	tree_iterator_entry *scan = ti->head->entries[ti->head->current];
 	ssize_t strpos = ti->path.size;
 	const git_tree_entry *te;
 
 	if (strpos && ti->path.ptr[strpos - 1] == '/')
 		strpos--;
 
-	while (scan && scan->parent) {
-		tree_iterator_entry *entry = &scan->entries[current];
-
-		if (!(te = tree_iterator__tree_entry(scan, entry)))
-			break;
-
+	for (; scan && (te = scan->te); scan = scan->parent) {
 		strpos -= te->filename_len;
 		memcpy(&ti->path.ptr[strpos], te->filename, te->filename_len);
 		strpos -= 1; /* separator */
-
-		current = entry->parent_entry_index;
-		scan = scan->parent;
 	}
 }
 
-static int tree_iterator__tree_entry_cmp(
+static int tree_iterator__te_cmp(
 	const git_tree_entry *a,
 	const git_tree_entry *b,
-	int (*strncomp)(const char *, const char *, size_t))
+	int (*compare)(const char *, const char *, size_t))
 {
-	size_t common = min(a->filename_len, b->filename_len);
-	int cmp = strncomp(a->filename, b->filename, common);
+	return git_path_cmp(
+		a->filename, a->filename_len, a->attr == GIT_FILEMODE_TREE,
+		b->filename, b->filename_len, b->attr == GIT_FILEMODE_TREE,
+		compare);
+}
+
+static int tree_iterator__ci_cmp(const void *a, const void *b, void *p)
+{
+	const tree_iterator_entry *ae = a, *be = b;
+	int cmp = tree_iterator__te_cmp(ae->te, be->te, git__strncasecmp);
 
 	if (!cmp) {
-		char a_next = a->filename[common], b_next = b->filename[common];
-
-		if (!a_next && a->attr == GIT_FILEMODE_TREE)
-			a_next = '/';
-		if (!b_next && b->attr == GIT_FILEMODE_TREE)
-			b_next = '/';
-
-		cmp = (int)a_next - (int)b_next;
+		/* stabilize sort order among equivalent names */
+		if (!ae->parent->te || !be->parent->te)
+			cmp = tree_iterator__te_cmp(ae->te, be->te, git__strncmp);
+		else
+			cmp = tree_iterator__ci_cmp(ae->parent, be->parent, p);
 	}
 
 	return cmp;
 }
 
-static int tree_iterator__entry_cmp(const void *a, const void *b, void *p)
+static int tree_iterator__search_cmp(const void *key, const void *val, void *p)
 {
-	const tree_iterator_entry *ae = a, *be = b;
-	const git_tree_entry *ate = tree_iterator__tree_entry(p, ae);
-	const git_tree_entry *bte = tree_iterator__tree_entry(p, be);
-	int cmp = tree_iterator__tree_entry_cmp(ate, bte, git__strncasecmp);
+	const tree_iterator_frame *tf = key;
+	const git_tree_entry *te = ((tree_iterator_entry *)val)->te;
 
-	/* stabilize sort order among equivalent names */
-	if (!cmp) {
-		cmp = (ae->parent_entry_index < be->parent_entry_index) ? -1 :
-			(ae->parent_entry_index > be->parent_entry_index) ? 1 : 0;
-		if (!cmp)
-			cmp = (ae->parent_tree_index < be->parent_tree_index) ? -1 :
-				(ae->parent_tree_index > be->parent_tree_index) ? 1 : 0;
-	}
-
-	return cmp;
+	return git_path_cmp(
+		tf->start, tf->startlen, false,
+		te->filename, te->filename_len, te->attr == GIT_FILEMODE_TREE,
+		((tree_iterator *)p)->strncomp);
 }
 
 static int tree_iterator__set_next(tree_iterator *ti, tree_iterator_frame *tf)
 {
-	/* find next and load trees for current range */
-	int error = 0;
+	int error;
 	const git_tree_entry *te, *last = NULL;
 
 	tf->next = tf->current;
 
-	while (tf->next < tf->n_entries) {
-		if (!(te = tree_iterator__tree_entry_by_index(tf, tf->next)) ||
-			(last && tree_iterator__tree_entry_cmp(last, te, ti->strncomp)))
+	for (; tf->next < tf->n_entries; tf->next++, last = te) {
+		te = tf->entries[tf->next]->te;
+
+		if (last && tree_iterator__te_cmp(last, te, ti->strncomp))
 			break;
 
+		/* load trees for items in [current,next) range */
 		if (git_tree_entry__is_tree(te) &&
 			(error = git_tree_lookup(
-				&tf->entries[tf->next].tree, ti->base.repo, &te->oid)) < 0)
-			break;
-
-		tf->next++;
-		last = te;
+				&tf->entries[tf->next]->tree, ti->base.repo, &te->oid)) < 0)
+			return error;
 	}
 
 	if (tf->next > tf->current + 1)
@@ -330,129 +293,139 @@ static int tree_iterator__set_next(tree_iterator *ti, tree_iterator_frame *tf)
 	if (last && !tree_iterator__current_filename(ti, last))
 		return -1;
 
-	return error;
+	return 0;
 }
 
 GIT_INLINE(bool) tree_iterator__at_tree(tree_iterator *ti)
 {
 	return (ti->head->current < ti->head->n_entries &&
-			ti->head->entries[ti->head->current].tree != NULL);
+			ti->head->entries[ti->head->current]->tree != NULL);
 }
 
 static int tree_iterator__push_frame(tree_iterator *ti)
 {
 	int error = 0;
-	tree_iterator_frame *tf = ti->head, *new_tf = NULL;
-	size_t i, n_entries = 0, sz = sizeof(tree_iterator_frame);
-	const git_tree_entry *te;
+	tree_iterator_frame *head = ti->head, *tf = NULL;
+	size_t i, n_entries = 0;
 
-	/* if current item in head is not a tree, do nothing */
-	if (tf->current >= tf->n_entries || !tf->entries[tf->current].tree)
+	if (head->current >= head->n_entries || !head->entries[head->current]->tree)
 		return 0;
 
-	/* build frame - sum tree entries from parent range */
-	for (i = tf->current; i < tf->next; ++i)
-		n_entries += git_tree_entrycount(tf->entries[i].tree);
-	sz += n_entries * sizeof(tree_iterator_entry);
-	new_tf = git__calloc(sz, sizeof(char));
-	GITERR_CHECK_ALLOC(new_tf);
+	for (i = head->current; i < head->next; ++i)
+		n_entries += git_tree_entrycount(head->entries[i]->tree);
 
-	/* populate frame and entries */
-	new_tf->parent = tf;
-	new_tf->n_entries = n_entries;
+	tf = git__calloc(sizeof(tree_iterator_frame) +
+		n_entries * sizeof(tree_iterator_entry *), 1);
+	GITERR_CHECK_ALLOC(tf);
 
-	for (i = tf->current, n_entries = 0; i < tf->next; ++i) {
-		git_tree *tree = tf->entries[i].tree;
+	tf->n_entries = n_entries;
+
+	tf->up     = head;
+	head->down = tf;
+	ti->head   = tf;
+
+	for (i = head->current, n_entries = 0; i < head->next; ++i) {
+		git_tree *tree = head->entries[i]->tree;
 		size_t j, max_j = git_tree_entrycount(tree);
 
 		for (j = 0; j < max_j; ++j) {
-			new_tf->entries[n_entries].parent_entry_index = i;
-			new_tf->entries[n_entries].parent_tree_index = j;
-			n_entries++;
+			tree_iterator_entry *entry = git_pool_malloc(&ti->pool, 1);
+			GITERR_CHECK_ALLOC(entry);
+
+			entry->parent = head->entries[i];
+			entry->te     = git_tree_entry_byindex(tree, j);
+			entry->tree   = NULL;
+
+			tf->entries[n_entries++] = entry;
 		}
 	}
 
 	/* if ignore_case, sort entries case insensitively */
 	if (iterator__ignore_case(ti))
-		git__qsort_r(
-			new_tf->entries, new_tf->n_entries, sizeof(tree_iterator_entry),
-			tree_iterator__entry_cmp, new_tf);
+		git__tsort_r(
+			(void **)tf->entries, tf->n_entries, tree_iterator__ci_cmp, tf);
 
-	/* pick new_tf->current based on "start" (or start at zero) */
-	if (tf->startlen > 0) {
-		/* find first item >= start */
-		for (i = 0; i < new_tf->n_entries; ++i) {
-			if (!(te = tree_iterator__tree_entry_by_index(new_tf, i)))
-				break;
-			sz = min(tf->startlen, te->filename_len);
-			if (ti->strncomp(tf->start, te->filename, sz) <= 0 &&
-				(tf->startlen <= te->filename_len ||
-				 tf->start[te->filename_len] == '/'))
-				break;
-		}
-		new_tf->current = i;
+	/* pick tf->current based on "start" (or start at zero) */
+	if (head->startlen > 0) {
+		git__bsearch_r((void **)tf->entries, tf->n_entries, head,
+			tree_iterator__search_cmp, ti, &tf->current);
 
-		if ((new_tf->start = strchr(tf->start, '/')) != NULL) {
-			new_tf->start++;
-			new_tf->startlen = strlen(new_tf->start);
+		while (tf->current &&
+			   !tree_iterator__search_cmp(head, tf->entries[tf->current-1], ti))
+			tf->current--;
+
+		if ((tf->start = strchr(head->start, '/')) != NULL) {
+			tf->start++;
+			tf->startlen = strlen(tf->start);
 		}
 	}
 
 	ti->path_has_filename = false;
 
-	/* find next and load trees for current range */
-	if ((error = tree_iterator__set_next(ti, new_tf)) < 0)
+	if ((error = tree_iterator__set_next(ti, tf)) < 0)
 		return error;
 
-	tf->child = new_tf;
-	ti->head  = new_tf;
-
+	/* autoexpand as needed */
 	if (!iterator__include_trees(ti) && tree_iterator__at_tree(ti))
 		return tree_iterator__push_frame(ti);
 
 	return 0;
 }
 
-GIT_INLINE(void) tree_iterator__free_tree(tree_iterator_entry *entry)
-{
-	if (entry->tree) {
-		git_tree_free(entry->tree);
-		entry->tree = NULL;
-	}
-}
-
 static bool tree_iterator__move_to_next(
 	tree_iterator *ti, tree_iterator_frame *tf)
 {
 	if (tf->next > tf->current + 1)
 		ti->path_ambiguities--;
 
+	if (!tf->up) { /* at root */
+		tf->current = tf->next;
+		return false;
+	}
+
 	for (; tf->current < tf->next; tf->current++) {
-		if (tf->parent)
-			tree_iterator__free_tree(&tf->entries[tf->current]);
+		git_tree_free(tf->entries[tf->current]->tree);
+		tf->entries[tf->current]->tree = NULL;
 	}
 
 	return (tf->current < tf->n_entries);
 }
 
-static bool tree_iterator__pop_frame(tree_iterator *ti)
+static bool tree_iterator__pop_frame(tree_iterator *ti, bool final)
 {
 	tree_iterator_frame *tf = ti->head;
 
-	if (!tf->parent)
+	if (!tf->up)
 		return false;
 
+	ti->head = tf->up;
+	ti->head->down = NULL;
+
 	tree_iterator__move_to_next(ti, tf);
 
-	ti->head = tf->parent;
-	ti->head->child = NULL;
+	if (!final) { /* if final, don't bother to clean up */
+		git_pool_free_array(&ti->pool, tf->n_entries, (void **)tf->entries);
+		git_buf_rtruncate_at_char(&ti->path, '/');
+	}
+
 	git__free(tf);
 
-	git_buf_rtruncate_at_char(&ti->path, '/');
-
 	return true;
 }
 
+static int tree_iterator__pop_all(tree_iterator *ti, bool to_end, bool final)
+{
+	while (tree_iterator__pop_frame(ti, final)) /* pop to root */;
+
+	if (!final) {
+		ti->head->current = to_end ? ti->head->n_entries : 0;
+		ti->path_ambiguities = 0;
+		git_buf_clear(&ti->path);
+	}
+
+	return 0;
+}
+
 static int tree_iterator__current(
 	const git_index_entry **entry, git_iterator *self)
 {
@@ -462,24 +435,21 @@ static int tree_iterator__current(
 
 	iterator__clear_entry(entry);
 
-	if (!(te = tree_iterator__tree_entry_by_index(tf, tf->current)))
+	if (tf->current >= tf->n_entries)
 		return 0;
+	te = tf->entries[tf->current]->te;
 
 	ti->entry.mode = te->attr;
 	git_oid_cpy(&ti->entry.oid, &te->oid);
 
 	ti->entry.path = tree_iterator__current_filename(ti, te);
-	if (ti->entry.path == NULL)
-		return -1;
+	GITERR_CHECK_ALLOC(ti->entry.path);
 
 	if (ti->path_ambiguities > 0)
 		tree_iterator__rewrite_filename(ti);
 
-	if (iterator__past_end(ti, ti->entry.path)) {
-		while (tree_iterator__pop_frame(ti)) /* pop to top */;
-		ti->head->current = ti->head->n_entries;
-		return 0;
-	}
+	if (iterator__past_end(ti, ti->entry.path))
+		return tree_iterator__pop_all(ti, true, false);
 
 	if (entry)
 		*entry = &ti->entry;
@@ -524,7 +494,8 @@ static int tree_iterator__advance(
 	}
 
 	/* scan forward and up, advancing in frame or popping frame when done */
-	while (!tree_iterator__move_to_next(ti, tf) && tree_iterator__pop_frame(ti))
+	while (!tree_iterator__move_to_next(ti, tf) &&
+		   tree_iterator__pop_frame(ti, false))
 		tf = ti->head;
 
 	/* find next and load trees */
@@ -549,15 +520,12 @@ static int tree_iterator__reset(
 {
 	tree_iterator *ti = (tree_iterator *)self;
 
-	while (tree_iterator__pop_frame(ti)) /* pop to top */;
-	ti->top->current = 0;
+	tree_iterator__pop_all(ti, false, false);
 
 	if (iterator__reset_range(self, start, end) < 0)
 		return -1;
-	git_buf_clear(&ti->path);
-	ti->path_ambiguities = 0;
 
-	return tree_iterator__push_frame(ti); /* re-expand top tree */
+	return tree_iterator__push_frame(ti); /* re-expand root tree */
 }
 
 static int tree_iterator__at_end(git_iterator *self)
@@ -570,30 +538,29 @@ static void tree_iterator__free(git_iterator *self)
 {
 	tree_iterator *ti = (tree_iterator *)self;
 
-	while (tree_iterator__pop_frame(ti)) /* pop to top */;
-
-	if (ti->head) {
-		tree_iterator__free_tree(&ti->head->entries[0]);
-		git__free(ti->head);
-	}
-	ti->head = ti->top = NULL;
+	tree_iterator__pop_all(ti, true, false);
 
+	git_tree_free(ti->head->entries[0]->tree);
+	git__free(ti->head);
+	git_pool_clear(&ti->pool);
 	git_buf_free(&ti->path);
 }
 
-static int tree_iterator__create_top_frame(tree_iterator *ti, git_tree *tree)
+static int tree_iterator__create_root_frame(tree_iterator *ti, git_tree *tree)
 {
 	size_t sz = sizeof(tree_iterator_frame) + sizeof(tree_iterator_entry);
-	tree_iterator_frame *top = git__calloc(sz, sizeof(char));
-	GITERR_CHECK_ALLOC(top);
+	tree_iterator_frame *root = git__calloc(sz, sizeof(char));
+	GITERR_CHECK_ALLOC(root);
 
-	top->n_entries = 1;
-	top->next = 1;
-	top->start = ti->base.start;
-	top->startlen = top->start ? strlen(top->start) : 0;
-	top->entries[0].tree = tree;
+	root->n_entries  = 1;
+	root->next       = 1;
+	root->start      = ti->base.start;
+	root->startlen   = root->start ? strlen(root->start) : 0;
+	root->entries[0] = git_pool_mallocz(&ti->pool, 1);
+	GITERR_CHECK_ALLOC(root->entries[0]);
+	root->entries[0]->tree = tree;
 
-	ti->head = ti->top = top;
+	ti->head = ti->root = root;
 
 	return 0;
 }
@@ -620,8 +587,9 @@ int git_iterator_for_tree(
 		goto fail;
 	ti->strncomp = iterator__ignore_case(ti) ? git__strncasecmp : git__strncmp;
 
-	if ((error = tree_iterator__create_top_frame(ti, tree)) < 0 ||
-		(error = tree_iterator__push_frame(ti)) < 0) /* expand top right now */
+	if ((error = git_pool_init(&ti->pool, sizeof(tree_iterator_entry),0)) < 0 ||
+		(error = tree_iterator__create_root_frame(ti, tree)) < 0 ||
+		(error = tree_iterator__push_frame(ti)) < 0) /* expand root now */
 		goto fail;
 
 	*iter = (git_iterator *)ti;
@@ -878,7 +846,6 @@ typedef struct {
 	git_iterator base;
 	git_iterator_callbacks cb;
 	workdir_iterator_frame *stack;
-	int (*entrycmp)(const void *pfx, const void *item);
 	git_ignores ignores;
 	git_index_entry entry;
 	git_buf path;
@@ -940,16 +907,11 @@ static void workdir_iterator__free_frame(workdir_iterator_frame *wf)
 
 static int workdir_iterator__update_entry(workdir_iterator *wi);
 
-static int workdir_iterator__entry_cmp_case(const void *pfx, const void *item)
+static int workdir_iterator__entry_cmp(const void *i, const void *item)
 {
+	const workdir_iterator *wi = (const workdir_iterator *)i;
 	const git_path_with_stat *ps = item;
-	return git__prefixcmp((const char *)pfx, ps->path);
-}
-
-static int workdir_iterator__entry_cmp_icase(const void *pfx, const void *item)
-{
-	const git_path_with_stat *ps = item;
-	return git__prefixcmp_icase((const char *)pfx, ps->path);
+	return wi->base.prefixcomp(wi->base.start, ps->path);
 }
 
 static void workdir_iterator__seek_frame_start(
@@ -960,7 +922,7 @@ static void workdir_iterator__seek_frame_start(
 
 	if (wi->base.start)
 		git_vector_bsearch2(
-			&wf->index, &wf->entries, wi->entrycmp, wi->base.start);
+			&wf->index, &wf->entries, workdir_iterator__entry_cmp, wi);
 	else
 		wf->index = 0;
 
@@ -1236,10 +1198,7 @@ int git_iterator_for_workdir(
 		git__free(wi);
 		return -1;
 	}
-
 	wi->root_len = wi->path.size;
-	wi->entrycmp = iterator__ignore_case(wi) ?
-		workdir_iterator__entry_cmp_icase : workdir_iterator__entry_cmp_case;
 
 	if ((error = workdir_iterator__expand_dir(wi)) < 0) {
 		if (error != GIT_ENOTFOUND)
@@ -1306,7 +1265,8 @@ int git_iterator_current_tree_entry(
 		*tree_entry = NULL;
 	else {
 		tree_iterator_frame *tf = ((tree_iterator *)iter)->head;
-		*tree_entry = tree_iterator__tree_entry_by_index(tf, tf->current);
+		*tree_entry = (tf->current < tf->n_entries) ?
+			tf->entries[tf->current]->te : NULL;
 	}
 
 	return 0;
@@ -1327,12 +1287,10 @@ int git_iterator_current_parent_tree(
 	if (iter->type != GIT_ITERATOR_TYPE_TREE)
 		return 0;
 
-	tf = ti->top;
-
-	while (*scan) {
-		/* get entry of this parent that child is currently on */
-		if (!(tf = tf->child) ||
-			!(te = tree_iterator__tree_entry_by_index(tf, tf->current)) ||
+	for (tf = ti->root; *scan; ) {
+		if (!(tf = tf->down) ||
+			tf->current >= tf->n_entries ||
+			!(te = tf->entries[tf->current]->te) ||
 			ti->strncomp(scan, te->filename, te->filename_len) != 0)
 			return 0;
 
@@ -1341,7 +1299,7 @@ int git_iterator_current_parent_tree(
 			scan++;
 	}
 
-	*tree_ptr = tf->entries[tf->current].tree;
+	*tree_ptr = tf->entries[tf->current]->tree;
 	return 0;
 }
 
diff --git a/src/path.c b/src/path.c
index 263cf9e7c..5767faeed 100644
--- a/src/path.c
+++ b/src/path.c
@@ -679,37 +679,14 @@ int git_path_apply_relative(git_buf *target, const char *relpath)
 
 int git_path_cmp(
 	const char *name1, size_t len1, int isdir1,
-	const char *name2, size_t len2, int isdir2)
+	const char *name2, size_t len2, int isdir2,
+	int (*compare)(const char *, const char *, size_t))
 {
 	unsigned char c1, c2;
 	size_t len = len1 < len2 ? len1 : len2;
 	int cmp;
 
-	cmp = memcmp(name1, name2, len);
-	if (cmp)
-		return cmp;
-
-	c1 = name1[len];
-	c2 = name2[len];
-
-	if (c1 == '\0' && isdir1)
-		c1 = '/';
-
-	if (c2 == '\0' && isdir2)
-		c2 = '/';
-
-	return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
-}
-
-int git_path_icmp(
-	const char *name1, size_t len1, int isdir1,
-	const char *name2, size_t len2, int isdir2)
-{
-	unsigned char c1, c2;
-	size_t len = len1 < len2 ? len1 : len2;
-	int cmp;
-
-	cmp = strncasecmp(name1, name2, len);
+	cmp = compare(name1, name2, len);
 	if (cmp)
 		return cmp;
 
diff --git a/src/path.h b/src/path.h
index feefd65d1..ead4fa338 100644
--- a/src/path.h
+++ b/src/path.h
@@ -265,12 +265,8 @@ extern int git_path_direach(
  */
 extern int git_path_cmp(
 	const char *name1, size_t len1, int isdir1,
-	const char *name2, size_t len2, int isdir2);
-
-/** Path sort function that is case insensitive */
-extern int git_path_icmp(
-	const char *name1, size_t len1, int isdir1,
-	const char *name2, size_t len2, int isdir2);
+	const char *name2, size_t len2, int isdir2,
+	int (*compare)(const char *, const char *, size_t));
 
 /**
  * Invoke callback up path directory by directory until the ceiling is
diff --git a/src/pool.c b/src/pool.c
index 64b5c6b00..6b78a0b74 100644
--- a/src/pool.c
+++ b/src/pool.c
@@ -235,10 +235,28 @@ char *git_pool_strcat(git_pool *pool, const char *a, const char *b)
 
 void git_pool_free(git_pool *pool, void *ptr)
 {
-	assert(pool && ptr && pool->item_size >= sizeof(void*));
+	assert(pool && pool->item_size >= sizeof(void*));
 
-	*((void **)ptr) = pool->free_list;
-	pool->free_list = ptr;
+	if (ptr) {
+		*((void **)ptr) = pool->free_list;
+		pool->free_list = ptr;
+	}
+}
+
+void git_pool_free_array(git_pool *pool, size_t count, void **ptrs)
+{
+	size_t i;
+
+	assert(pool && ptrs && pool->item_size >= sizeof(void*));
+
+	if (!count)
+		return;
+
+	for (i = count - 1; i > 0; --i)
+		*((void **)ptrs[i]) = ptrs[i - 1];
+
+	*((void **)ptrs[0]) = pool->free_list;
+	pool->free_list = ptrs[count - 1];
 }
 
 uint32_t git_pool__open_pages(git_pool *pool)
diff --git a/src/pool.h b/src/pool.h
index 2b262a588..5ac9b764f 100644
--- a/src/pool.h
+++ b/src/pool.h
@@ -126,6 +126,13 @@ extern char *git_pool_strcat(git_pool *pool, const char *a, const char *b);
  */
 extern void git_pool_free(git_pool *pool, void *ptr);
 
+/**
+ * Push an array of pool allocated blocks efficiently onto the free list.
+ *
+ * This has the same constraints as `git_pool_free()` above.
+ */
+extern void git_pool_free_array(git_pool *pool, size_t count, void **ptrs);
+
 /*
  * Misc utilities
  */
diff --git a/src/tree.c b/src/tree.c
index 11123a18a..17b3c378d 100644
--- a/src/tree.c
+++ b/src/tree.c
@@ -55,23 +55,28 @@ static int valid_entry_name(const char *filename)
 		  strcmp(filename, DOT_GIT) != 0));
 }
 
-int git_tree_entry_cmp(const git_tree_entry *e1, const git_tree_entry *e2)
+static int entry_sort_cmp(const void *a, const void *b)
 {
+	const git_tree_entry *e1 = (const git_tree_entry *)a;
+	const git_tree_entry *e2 = (const git_tree_entry *)b;
+
 	return git_path_cmp(
 		e1->filename, e1->filename_len, git_tree_entry__is_tree(e1),
-		e2->filename, e2->filename_len, git_tree_entry__is_tree(e2));
+		e2->filename, e2->filename_len, git_tree_entry__is_tree(e2),
+		git__strncmp);
+}
+
+int git_tree_entry_cmp(const git_tree_entry *e1, const git_tree_entry *e2)
+{
+	return entry_sort_cmp(e1, e2);
 }
 
 int git_tree_entry_icmp(const git_tree_entry *e1, const git_tree_entry *e2)
 {
-	return git_path_icmp(
+	return git_path_cmp(
 		e1->filename, e1->filename_len, git_tree_entry__is_tree(e1),
-		e2->filename, e2->filename_len, git_tree_entry__is_tree(e2));
-}
-
-static int entry_sort_cmp(const void *a, const void *b)
-{
-	return git_tree_entry_cmp((const git_tree_entry *)a, (const git_tree_entry *)b);
+		e2->filename, e2->filename_len, git_tree_entry__is_tree(e2),
+		git__strncasecmp);
 }
 
 static git_tree_entry *alloc_entry(const char *filename)
diff --git a/tests-clar/core/pool.c b/tests-clar/core/pool.c
index 5ed97366f..c42bb6da0 100644
--- a/tests-clar/core/pool.c
+++ b/tests-clar/core/pool.c
@@ -83,3 +83,53 @@ void test_core_pool__2(void)
 
 	git_pool_clear(&p);
 }
+
+void test_core_pool__free_list(void)
+{
+	int i;
+	git_pool p;
+	void *ptr, *ptrs[50];
+
+	cl_git_pass(git_pool_init(&p, 100, 100));
+
+	for (i = 0; i < 10; ++i) {
+		ptr = git_pool_malloc(&p, 1);
+		cl_assert(ptr != NULL);
+	}
+	cl_assert_equal_i(10, (int)p.items);
+
+	for (i = 0; i < 50; ++i) {
+		ptrs[i] = git_pool_malloc(&p, 1);
+		cl_assert(ptrs[i] != NULL);
+	}
+	cl_assert_equal_i(60, (int)p.items);
+
+	git_pool_free(&p, ptr);
+	cl_assert_equal_i(60, (int)p.items);
+
+	git_pool_free_array(&p, 50, ptrs);
+	cl_assert_equal_i(60, (int)p.items);
+
+	for (i = 0; i < 50; ++i) {
+		ptrs[i] = git_pool_malloc(&p, 1);
+		cl_assert(ptrs[i] != NULL);
+	}
+	cl_assert_equal_i(60, (int)p.items);
+
+	for (i = 0; i < 111; ++i) {
+		ptr = git_pool_malloc(&p, 1);
+		cl_assert(ptr != NULL);
+	}
+	cl_assert_equal_i(170, (int)p.items);
+
+	git_pool_free_array(&p, 50, ptrs);
+	cl_assert_equal_i(170, (int)p.items);
+
+	for (i = 0; i < 50; ++i) {
+		ptrs[i] = git_pool_malloc(&p, 1);
+		cl_assert(ptrs[i] != NULL);
+	}
+	cl_assert_equal_i(170, (int)p.items);
+
+	git_pool_clear(&p);
+}

From d85296ab9b9c4a01adb35d4d2438b72177aeabc4 Mon Sep 17 00:00:00 2001
From: Russell Belfer <rb@github.com>
Date: Thu, 14 Mar 2013 13:50:54 -0700
Subject: [PATCH 2/3] Fix valgrind issues (and mmap fallback for diff)

This fixes a number of issues identified by valgrind - mostly
missed free calls.  Inside valgrind, mmap() may fail which causes
some of the diff tests to fail.  This adds a fallback code path
to diff_output.c:get_workdir_content() where is the mmap() fails
the code will now try to read the file data directly into allocated
memory (which is what it would do if the data needed to be filtered
anyhow).
---
 src/checkout.c             |  2 +-
 src/diff_output.c          | 53 ++++++++++++++++++++++++--------------
 src/errors.c               |  1 +
 tests-clar/repo/iterator.c |  2 ++
 4 files changed, 37 insertions(+), 21 deletions(-)

diff --git a/src/checkout.c b/src/checkout.c
index 68ebbe31d..e52649aec 100644
--- a/src/checkout.c
+++ b/src/checkout.c
@@ -610,7 +610,7 @@ static int checkout_get_actions(
 		if (act & CHECKOUT_ACTION__CONFLICT)
 			counts[CHECKOUT_ACTION__CONFLICT]++;
 	}
-	
+
 	error = checkout_remaining_wd_items(data, workdir, wditem, &pathspec);
 	if (error < 0)
 		goto fail;
diff --git a/src/diff_output.c b/src/diff_output.c
index 43262b1ae..c0aff6826 100644
--- a/src/diff_output.c
+++ b/src/diff_output.c
@@ -330,6 +330,33 @@ static int get_workdir_sm_content(
 	return 0;
 }
 
+static int get_filtered(
+	git_map *map, git_file fd, git_diff_file *file, git_vector *filters)
+{
+	int error;
+	git_buf raw = GIT_BUF_INIT, filtered = GIT_BUF_INIT;
+
+	if ((error = git_futils_readbuffer_fd(&raw, fd, (size_t)file->size)) < 0)
+		return error;
+
+	if (!filters->length)
+		git_buf_swap(&filtered, &raw);
+	else
+		error = git_filters_apply(&filtered, &raw, filters);
+
+	if (!error) {
+		map->len  = git_buf_len(&filtered);
+		map->data = git_buf_detach(&filtered);
+
+		file->flags |= GIT_DIFF_FLAG__FREE_DATA;
+	}
+
+	git_buf_free(&raw);
+	git_buf_free(&filtered);
+
+	return error;
+}
+
 static int get_workdir_content(
 	diff_context *ctxt,
 	git_diff_delta *delta,
@@ -381,8 +408,8 @@ static int get_workdir_content(
 			goto cleanup;
 		}
 
-		if (!file->size)
-			file->size = git_futils_filesize(fd);
+		if (!file->size && !(file->size = git_futils_filesize(fd)))
+			goto close_and_cleanup;
 
 		if ((error = diff_delta_is_binary_by_size(ctxt, delta, file)) < 0 ||
 			(delta->flags & GIT_DIFF_FLAG_BINARY) != 0)
@@ -394,26 +421,12 @@ static int get_workdir_content(
 			goto close_and_cleanup;
 
 		if (error == 0) { /* note: git_filters_load returns filter count */
-			if (!file->size)
-				goto close_and_cleanup;
-
 			error = git_futils_mmap_ro(map, fd, 0, (size_t)file->size);
-			file->flags |= GIT_DIFF_FLAG__UNMAP_DATA;
-		} else {
-			git_buf raw = GIT_BUF_INIT, filtered = GIT_BUF_INIT;
-
-			if (!(error = git_futils_readbuffer_fd(&raw, fd, (size_t)file->size)) &&
-				!(error = git_filters_apply(&filtered, &raw, &filters)))
-			{
-				map->len  = git_buf_len(&filtered);
-				map->data = git_buf_detach(&filtered);
-
-				file->flags |= GIT_DIFF_FLAG__FREE_DATA;
-			}
-
-			git_buf_free(&raw);
-			git_buf_free(&filtered);
+			if (!error)
+				file->flags |= GIT_DIFF_FLAG__UNMAP_DATA;
 		}
+		if (error != 0)
+			error = get_filtered(map, fd, file, &filters);
 
 close_and_cleanup:
 		git_filters_free(&filters);
diff --git a/src/errors.c b/src/errors.c
index c5f0b3b59..e2629f69e 100644
--- a/src/errors.c
+++ b/src/errors.c
@@ -103,6 +103,7 @@ int giterr_set_regex(const regex_t *regex, int error_code)
 
 void giterr_clear(void)
 {
+	set_error(0, NULL);
 	GIT_GLOBAL->last_error = NULL;
 
 	errno = 0;
diff --git a/tests-clar/repo/iterator.c b/tests-clar/repo/iterator.c
index 9e1f09881..00c46d6b1 100644
--- a/tests-clar/repo/iterator.c
+++ b/tests-clar/repo/iterator.c
@@ -337,6 +337,8 @@ void test_repo_iterator__tree_icase(void)
 		&i, head, flag | GIT_ITERATOR_DONT_AUTOEXPAND, "k", "k/Z"));
 	expect_iterator_items(i, 1, NULL, 6, NULL);
 	git_iterator_free(i);
+
+	git_tree_free(head);
 }
 
 void test_repo_iterator__tree_more(void)

From 14bedad90776b750cf69ed6faadf2d6c3a4e0a86 Mon Sep 17 00:00:00 2001
From: Russell Belfer <rb@github.com>
Date: Thu, 14 Mar 2013 15:08:04 -0700
Subject: [PATCH 3/3] Added pool freelist struct for readability

This adds a git_pool_freelist_item struct that makes it a little
easier to follow what's going on with the pool free list block
management code.  It is functionally neutral.
---
 src/pool.c | 22 +++++++++++++++-------
 1 file changed, 15 insertions(+), 7 deletions(-)

diff --git a/src/pool.c b/src/pool.c
index 6b78a0b74..33ea74fc7 100644
--- a/src/pool.c
+++ b/src/pool.c
@@ -10,6 +10,11 @@ struct git_pool_page {
 	char data[GIT_FLEX_ARRAY];
 };
 
+typedef struct git_pool_freelist_item git_pool_freelist_item;
+struct git_pool_freelist_item {
+	git_pool_freelist_item *next;
+};
+
 #define GIT_POOL_MIN_USABLE	4
 #define GIT_POOL_MIN_PAGESZ	2 * sizeof(void*)
 
@@ -150,7 +155,7 @@ void *git_pool_malloc(git_pool *pool, uint32_t items)
 		pool->has_multi_item_alloc = 1;
 	else if (pool->free_list != NULL) {
 		ptr = pool->free_list;
-		pool->free_list = *((void **)pool->free_list);
+		pool->free_list = ((git_pool_freelist_item *)pool->free_list)->next;
 		return ptr;
 	}
 
@@ -235,16 +240,19 @@ char *git_pool_strcat(git_pool *pool, const char *a, const char *b)
 
 void git_pool_free(git_pool *pool, void *ptr)
 {
+	git_pool_freelist_item *item = ptr;
+
 	assert(pool && pool->item_size >= sizeof(void*));
 
-	if (ptr) {
-		*((void **)ptr) = pool->free_list;
-		pool->free_list = ptr;
+	if (item) {
+		item->next = pool->free_list;
+		pool->free_list = item;
 	}
 }
 
 void git_pool_free_array(git_pool *pool, size_t count, void **ptrs)
 {
+	git_pool_freelist_item **items = (git_pool_freelist_item **)ptrs;
 	size_t i;
 
 	assert(pool && ptrs && pool->item_size >= sizeof(void*));
@@ -253,10 +261,10 @@ void git_pool_free_array(git_pool *pool, size_t count, void **ptrs)
 		return;
 
 	for (i = count - 1; i > 0; --i)
-		*((void **)ptrs[i]) = ptrs[i - 1];
+		items[i]->next = items[i - 1];
 
-	*((void **)ptrs[0]) = pool->free_list;
-	pool->free_list = ptrs[count - 1];
+	items[i]->next  = pool->free_list;
+	pool->free_list = items[count - 1];
 }
 
 uint32_t git_pool__open_pages(git_pool *pool)