diff --git a/include/git2/object.h b/include/git2/object.h
index 7fca1c988..4cb6af0ad 100644
--- a/include/git2/object.h
+++ b/include/git2/object.h
@@ -58,6 +58,36 @@ GIT_BEGIN_DECL
  */
 GIT_EXTERN(int) git_object_lookup(git_object **object, git_repository *repo, const git_oid *id, git_otype type);
 
+/**
+ * Lookup a reference to one of the objects in a repostory,
+ * given a prefix of its identifier (short id).
+ *
+ * The object obtained will be so that its identifier
+ * matches the first 'len' hexadecimal characters
+ * (packets of 4 bits) of the given 'id'.
+ * 'len' must be long enough to identify a unique
+ * object matching the prefix; otherwise the method will
+ * fail.
+ *
+ * The generated reference is owned by the repository and
+ * should be closed with the `git_object_close` method
+ * instead of free'd manually.
+ *
+ * The 'type' parameter must match the type of the object
+ * in the odb; the method will fail otherwise.
+ * The special value 'GIT_OBJ_ANY' may be passed to let
+ * the method guess the object's type.
+ *
+ * @param object pointer to the looked-up object
+ * @param repo the repository to look up the object
+ * @param id a short identifier for the object
+ * @param len the length of the short identifier
+ * @param type the type of the object
+ * @return a reference to the object
+ */
+GIT_EXTERN(int) git_object_lookup_short_oid(git_object **object_out, git_repository *repo,
+					const git_oid *id, unsigned int len, git_otype type);
+
 /**
  * Get the id (SHA1) of a repository object
  *
diff --git a/include/git2/odb.h b/include/git2/odb.h
index 1d351beea..483934ad3 100644
--- a/include/git2/odb.h
+++ b/include/git2/odb.h
@@ -109,7 +109,7 @@ GIT_EXTERN(void) git_odb_close(git_odb *db);
 /**
  * Read an object from the database.
  *
- * This method queries all avaiable ODB backends
+ * This method queries all available ODB backends
  * trying to read the given OID.
  *
  * The returned object is reference counted and
@@ -125,6 +125,36 @@ GIT_EXTERN(void) git_odb_close(git_odb *db);
  */
 GIT_EXTERN(int) git_odb_read(git_odb_object **out, git_odb *db, const git_oid *id);
 
+/**
+ * Read an object from the database, given a prefix
+ * of its identifier.
+ *
+ * This method queries all available ODB backends
+ * trying to match the 'len' first hexadecimal
+ * characters of the 'short_id'.
+ * The remaining bits (GIT_OID_HEXSZ-len)*4 bits of
+ * 'short_id' must be 0s.
+ * The prefix must be long enough to identify
+ * a unique object in all the backends; the
+ * method will fail otherwise.
+ *
+ * The returned object is reference counted and
+ * internally cached, so it should be closed
+ * by the user once it's no longer in use.
+ *
+ * @param out_oid the oid of the unique object matching
+ * the short id
+ * @param out pointer where to store the read object
+ * @param db database to search for the object in.
+ * @param short_id a prefix of the id of the object to read.
+ * @param len the length of the prefix
+ * @return
+ * - GIT_SUCCESS if the object was read;
+ * - GIT_ENOTFOUND if the object is not in the database.
+ * - GIT_EAMBIGUOUS if the prefix is ambiguous (several objects match the prefix)
+ */
+GIT_EXTERN(int) git_odb_read_unique_short_oid(git_oid *out_oid, git_odb_object **out, git_odb *db, const git_oid *short_id, unsigned int len);
+
 /**
  * Read the header of an object from the database, without
  * reading its full contents.
diff --git a/src/object.c b/src/object.c
index d2e4da359..e6e976756 100644
--- a/src/object.c
+++ b/src/object.c
@@ -95,24 +95,63 @@ static int create_object(git_object **object_out, git_otype type)
 	return GIT_SUCCESS;
 }
 
-int git_object_lookup(git_object **object_out, git_repository *repo, const git_oid *id, git_otype type)
+int git_object_lookup_short_oid(git_object **object_out, git_repository *repo, const git_oid *id, unsigned int len, git_otype type)
 {
 	git_object *object = NULL;
 	git_odb_object *odb_obj;
 	int error = GIT_SUCCESS;
+	git_oid out_oid;
 
 	assert(repo && object_out && id);
 
-	object = git_cache_get(&repo->objects, id);
-	if (object != NULL) {
-		if (type != GIT_OBJ_ANY && type != object->type)
-			return git__throw(GIT_EINVALIDTYPE, "Failed to lookup object. The given type does not match the type on the ODB");
-
-		*object_out = object;
-		return GIT_SUCCESS;
+	if (len == 0)
+		return git__throw(GIT_EAMBIGUOUSOIDPREFIX, "Failed to lookup object. Prefix length should be not be 0.");
+	if (len > GIT_OID_HEXSZ) {
+		len = GIT_OID_HEXSZ;
+	}
+
+	if (len == GIT_OID_HEXSZ)  {
+		/* We want to match the full id : we can first look up in the cache,
+		 * since there is no need to check for non ambiguousity
+		 */
+		object = git_cache_get(&repo->objects, id);
+		if (object != NULL) {
+			if (type != GIT_OBJ_ANY && type != object->type)
+				return git__throw(GIT_EINVALIDTYPE, "Failed to lookup object. The given type does not match the type on the ODB");
+
+			*object_out = object;
+			return GIT_SUCCESS;
+		}
+
+		/* Object was not found in the cache, let's explore the backends.
+		 * We could just use git_odb_read_unique_short_oid,
+		 * it is the same cost for packed and loose object backends,
+		 * but it may be much more costly for sqlite and hiredis.
+		 */
+		error = git_odb_read(&odb_obj, repo->db, id);
+		git_oid_cpy(&out_oid, id);
+	} else {
+		git_oid short_oid;
+
+		/* We copy the first len*4 bits from id and fill the remaining with 0s */
+		memcpy(short_oid.id, id->id, (len + 1) / 2);
+		if (len % 2)
+			short_oid.id[len / 2] &= 0xF0;
+		memset(short_oid.id + (len + 1) / 2, 0, (GIT_OID_HEXSZ - len) / 2);
+
+		/* If len < GIT_OID_HEXSZ (a strict short oid was given), we have
+		 * 2 options :
+		 * - We always search in the cache first. If we find that short oid is
+		 *   ambiguous, we can stop. But in all the other cases, we must then
+		 *   explore all the backends (to find an object if there was match,
+		 *   or to check that oid is not ambiguous if we have found 1 match in
+		 *   the cache)
+		 * - We never explore the cache, go right to exploring the backends
+		 * We chose the latter : we explore directly the backends.
+		 */
+		error = git_odb_read_unique_short_oid(&out_oid, &odb_obj, repo->db, &short_oid, len);
 	}
 
-	error = git_odb_read(&odb_obj, repo->db, id);
 	if (error < GIT_SUCCESS)
 		return git__rethrow(error, "Failed to lookup object");
 
@@ -127,7 +166,7 @@ int git_object_lookup(git_object **object_out, git_repository *repo, const git_o
 		return git__rethrow(error, "Failed to lookup object");
 
 	/* Initialize parent object */
-	git_oid_cpy(&object->cached.oid, id);
+	git_oid_cpy(&object->cached.oid, &out_oid);
 	object->repo = repo;
 
 	switch (type) {
@@ -162,6 +201,10 @@ int git_object_lookup(git_object **object_out, git_repository *repo, const git_o
 	return GIT_SUCCESS;
 }
 
+int git_object_lookup(git_object **object_out, git_repository *repo, const git_oid *id, git_otype type) {
+	return git_object_lookup_short_oid(object_out, repo, id, GIT_OID_HEXSZ, type);
+}
+
 void git_object__free(void *_obj)
 {
 	git_object *object = (git_object *)_obj;
diff --git a/src/odb.c b/src/odb.c
index c669e7c14..ed600ffab 100644
--- a/src/odb.c
+++ b/src/odb.c
@@ -488,6 +488,48 @@ int git_odb_read(git_odb_object **out, git_odb *db, const git_oid *id)
 	return error;
 }
 
+int git_odb_read_unique_short_oid(git_oid *out_oid, git_odb_object **out, git_odb *db, const git_oid *short_id, unsigned int len)
+{
+	unsigned int i;
+	int error = GIT_ENOTFOUND;
+	git_rawobj raw;
+	int found = 0;
+
+	assert(out && db && id && len > 0);
+
+	if (len > GIT_OID_HEXSZ)
+		len = GIT_OID_HEXSZ;
+
+	if (len == GIT_OID_HEXSZ) {
+		*out = git_cache_get(&db->cache, short_id);
+		if (*out != NULL) {
+			git_oid_cpy(out_oid, short_id);
+			return GIT_SUCCESS;
+		}
+	}
+
+	for (i = 0; i < db->backends.length && found < 2; ++i) {
+		backend_internal *internal = git_vector_get(&db->backends, i);
+		git_odb_backend *b = internal->backend;
+
+		if (b->read != NULL) {
+			error = b->read_unique_short_oid(out_oid, &raw.data, &raw.len, &raw.type, b, short_id, len);
+			if (error == GIT_SUCCESS)
+				found++;
+		}
+	}
+
+	if (found == 1) {
+		*out = git_cache_try_store(&db->cache, new_odb_object(out_oid, &raw));
+	} else if (found > 1) {
+		return git__rethrow(GIT_EAMBIGUOUSOIDPREFIX, "Ambiguous sha1");
+	} else {
+		return git__rethrow(GIT_ENOTFOUND, "Failed to read object");
+	}
+
+	return GIT_SUCCESS;
+}
+
 int git_odb_write(git_oid *oid, git_odb *db, const void *data, size_t len, git_otype type)
 {
 	unsigned int i;
diff --git a/src/odb_pack.c b/src/odb_pack.c
index 3125a8c94..605608314 100644
--- a/src/odb_pack.c
+++ b/src/odb_pack.c
@@ -29,7 +29,9 @@
 #include "fileops.h"
 #include "hash.h"
 #include "odb.h"
+#include "oid.h"
 #include "delta-apply.h"
+#include "sha1_lookup.h"
 
 #include "git2/odb_backend.h"
 
@@ -262,15 +264,37 @@ static int packfile_refresh_all(struct pack_backend *backend);
 
 static off_t nth_packed_object_offset(const struct pack_file *p, uint32_t n);
 
-static int pack_entry_find_offset(off_t *offset_out,
-		struct pack_file *p, const git_oid *oid);
+/* Can find the offset of an object given
+ * a prefix of an identifier.
+ * Throws GIT_EAMBIGUOUSOIDPREFIX if short oid
+ * is ambiguous within the pack.
+ */
+static int pack_entry_find_offset(
+		off_t *offset_out,
+		git_oid *found_oid,
+		struct pack_file *p,
+		const git_oid *short_oid,
+		unsigned int len);
 
-static int pack_entry_find1(struct pack_entry *e,
-		struct pack_file *p, const git_oid *oid);
+static int pack_entry_find1(
+		struct pack_entry *e,
+		struct pack_file *p,
+		const git_oid *short_oid,
+		unsigned int len);
 
 static int pack_entry_find(struct pack_entry *e,
 		struct pack_backend *backend, const git_oid *oid);
 
+/* Can find the offset of an object given
+ * a prefix of an identifier.
+ * Throws GIT_EAMBIGUOUSOIDPREFIX if short oid
+ * is ambiguous.
+ */
+static int pack_entry_find_unique_short_oid(struct pack_entry *e,
+					struct pack_backend *backend,
+					const git_oid *short_oid,
+					unsigned int len);
+
 static off_t get_delta_base(struct pack_backend *backend,
 		struct pack_file *p, struct pack_window **w_curs,
 		off_t *curpos, git_otype type,
@@ -923,12 +947,15 @@ static off_t nth_packed_object_offset(const struct pack_file *p, uint32_t n)
 
 static int pack_entry_find_offset(
 		off_t *offset_out,
+		git_oid *found_oid,
 		struct pack_file *p,
-		const git_oid *oid)
+		const git_oid *short_oid,
+		unsigned int len)
 {
 	const uint32_t *level1_ofs = p->index_map.data;
 	const unsigned char *index = p->index_map.data;
 	unsigned hi, lo, stride;
+	int found = 0;
 
 	*offset_out = 0;
 
@@ -950,8 +977,8 @@ static int pack_entry_find_offset(
 	}
 
 	index += 4 * 256;
-	hi = ntohl(level1_ofs[(int)oid->id[0]]);
-	lo = ((oid->id[0] == 0x0) ? 0 : ntohl(level1_ofs[(int)oid->id[0] - 1]));
+	hi = ntohl(level1_ofs[(int)short_oid->id[0]]);
+	lo = ((short_oid->id[0] == 0x0) ? 0 : ntohl(level1_ofs[(int)short_oid->id[0] - 1]));
 
 	if (p->index_version > 1) {
 		stride = 20;
@@ -962,60 +989,79 @@ static int pack_entry_find_offset(
 
 #ifdef INDEX_DEBUG_LOOKUP
 	printf("%02x%02x%02x... lo %u hi %u nr %d\n",
-		oid->id[0], oid->id[1], oid->id[2], lo, hi, p->num_objects);
+		short_oid->id[0], short_oid->id[1], short_oid->id[2], lo, hi, p->num_objects);
 #endif
 
-#ifdef GIT2_INDEX_LOOKUP /* TODO: use the advanced lookup method from git.git */
+	/* Use git.git lookup code */
+	int pos =  sha1_entry_pos(index, stride, 0, lo, hi, p->num_objects, short_oid->id);
 
-	int pos = sha1_entry_pos(index, stride, 0, lo, hi, p->num_objects, oid);
-	if (pos < 0)
-		return git__throw(GIT_ENOTFOUND, "Failed to find offset for pack entry. Entry not found");
+	const unsigned char *current;
+	if (pos >= 0) {
+		/* An object matching exactly the oid was found */
+		found = 1;
+		current = index + pos * stride;
+	} else {
+		/* No object was found */
+		pos = - 1 - pos;
+		/* pos refers to the object with the "closest" oid to short_oid */
+		if (pos < p->num_objects) {
+			current = index + pos * stride;
 
-	*offset_out = nth_packed_object_offset(p, pos);
-	return GIT_SUCCESS;
-
-#else /* use an old and boring binary search */
-
-	do {
-		unsigned mi = (lo + hi) / 2;
-		int cmp = memcmp(index + mi * stride, oid->id, GIT_OID_RAWSZ);
-
-		if (!cmp) {
-			*offset_out = nth_packed_object_offset(p, mi);
-			return GIT_SUCCESS;
+			if (git_oid_match_raw(len, short_oid->id, current)) {
+				found = 1;
+			}
 		}
+	}
+	if (found && pos + 1 < p->num_objects) {
+		/* Check for ambiguousity */
+		const unsigned char *next = current + stride;
 
-		if (cmp > 0)
-			hi = mi;
-		else
-			lo = mi+1;
+		if (git_oid_match_raw(len, short_oid->id, next)) {
+			found = 2;
+		}
+	}
 
-	} while (lo < hi);
+	if (!found) {
+		return git__throw(GIT_ENOTFOUND, "Failed to find offset for pack entry. Entry not found");
+	} else if (found > 1) {
+		return git__throw(GIT_EAMBIGUOUSOIDPREFIX, "Failed to find offset for pack entry. Ambiguous sha1 prefix within pack");
+	} else {
+		*offset_out = nth_packed_object_offset(p, pos);
+		git_oid_mkraw(found_oid, current);
 
-	return git__throw(GIT_ENOTFOUND, "Failed to find offset for pack entry. Entry not found");
+#ifdef INDEX_DEBUG_LOOKUP
+		unsigned char hex_sha1[GIT_OID_HEXSZ + 1];
+		git_oid_fmt(hex_sha1, found_oid);
+		hex_sha1[GIT_OID_HEXSZ] = '\0';
+		printf("found lo=%d %s\n", lo, hex_sha1);
 #endif
+		return GIT_SUCCESS;
+	}
 }
 
 static int pack_entry_find1(
 		struct pack_entry *e,
 		struct pack_file *p,
-		const git_oid *oid)
+		const git_oid *short_oid,
+		unsigned int len)
 {
 	off_t offset;
 
 	assert(p);
 
-	if (p->num_bad_objects) {
+	if (len == GIT_OID_HEXSZ && p->num_bad_objects) {
 		unsigned i;
 		for (i = 0; i < p->num_bad_objects; i++)
-			if (git_oid_cmp(oid, &p->bad_object_sha1[i]) == 0)
+			if (git_oid_cmp(short_oid, &p->bad_object_sha1[i]) == 0)
 				return git__throw(GIT_ERROR, "Failed to find pack entry. Bad object found");
 	}
 
-	if (pack_entry_find_offset(&offset, p, oid) < GIT_SUCCESS)
-		return git__throw(GIT_ENOTFOUND, "Failed to find pack entry. Couldn't find offset");
-	
-	/* we found an entry in the index;
+	git_oid found_oid;
+	int error = pack_entry_find_offset(&offset, &found_oid, p, short_oid, len);
+	if (error < GIT_SUCCESS)
+		return git__rethrow(error, "Failed to find pack entry. Couldn't find offset");
+
+	/* we found a unique entry in the index;
 	 * make sure the packfile backing the index 
 	 * still exists on disk */
 	if (p->pack_fd == -1 && packfile_open(p) < GIT_SUCCESS)
@@ -1024,7 +1070,7 @@ static int pack_entry_find1(
 	e->offset = offset;
 	e->p = p;
 
-	git_oid_cpy(&e->sha1, oid);
+	git_oid_cpy(&e->sha1, &found_oid);
 	return GIT_SUCCESS;
 }
 
@@ -1037,7 +1083,7 @@ static int pack_entry_find(struct pack_entry *e, struct pack_backend *backend, c
 		return git__rethrow(error, "Failed to find pack entry");
 
 	if (backend->last_found &&
-		pack_entry_find1(e, backend->last_found, oid) == GIT_SUCCESS)
+		pack_entry_find1(e, backend->last_found, oid, GIT_OID_HEXSZ) == GIT_SUCCESS)
 		return GIT_SUCCESS;
 
 	for (i = 0; i < backend->packs.length; ++i) {
@@ -1047,7 +1093,7 @@ static int pack_entry_find(struct pack_entry *e, struct pack_backend *backend, c
 		if (p == backend->last_found)
 			continue;
 
-		if (pack_entry_find1(e, p, oid) == GIT_SUCCESS) {
+		if (pack_entry_find1(e, p, oid, GIT_OID_HEXSZ) == GIT_SUCCESS) {
 			backend->last_found = p;
 			return GIT_SUCCESS;
 		}
@@ -1056,6 +1102,53 @@ static int pack_entry_find(struct pack_entry *e, struct pack_backend *backend, c
 	return git__throw(GIT_ENOTFOUND, "Failed to find pack entry");
 }
 
+static int pack_entry_find_unique_short_oid(struct pack_entry *e, struct pack_backend *backend,
+					const git_oid *short_oid, unsigned int len)
+{
+	int error;
+	size_t i;
+
+	if ((error = packfile_refresh_all(backend)) < GIT_SUCCESS)
+		return git__rethrow(error, "Failed to find pack entry");
+
+	unsigned found = 0;
+	if (backend->last_found) {
+		error = pack_entry_find1(e, backend->last_found, short_oid, len);
+		if (error == GIT_EAMBIGUOUSOIDPREFIX) {
+			return git__rethrow(error, "Failed to find pack entry. Ambiguous sha1 prefix");
+		} else if (error == GIT_SUCCESS) {
+			found = 1;
+		}
+	}
+
+	for (i = 0; i < backend->packs.length; ++i) {
+		struct pack_file *p;
+
+		p = git_vector_get(&backend->packs, i);
+		if (p == backend->last_found)
+			continue;
+
+		error = pack_entry_find1(e, p, short_oid, len);
+		if (error == GIT_EAMBIGUOUSOIDPREFIX) {
+			return git__rethrow(error, "Failed to find pack entry. Ambiguous sha1 prefix");
+		} else if (error == GIT_SUCCESS) {
+			found++;
+			if (found > 1);
+				break;
+			backend->last_found = p;
+		}
+	}
+
+	if (!found) {
+		return git__rethrow(GIT_ENOTFOUND, "Failed to find pack entry");
+	} else if (found > 1) {
+		return git__rethrow(GIT_EAMBIGUOUSOIDPREFIX, "Failed to find pack entry. Ambiguous sha1 prefix");
+	} else {
+		return GIT_SUCCESS;
+	}
+
+}
+
 
 
 
@@ -1190,6 +1283,7 @@ static off_t get_delta_base(
 {
 	unsigned char *base_info = pack_window_open(backend, p, w_curs, *curpos, NULL);
 	off_t base_offset;
+	git_oid unused;
 
 	/* pack_window_open() assured us we have [base_info, base_info + 20)
 	 * as a range that we can look at without walking off the
@@ -1214,7 +1308,7 @@ static off_t get_delta_base(
 		*curpos += used;
 	} else if (type == GIT_OBJ_REF_DELTA) {
 		/* The base entry _must_ be in the same pack */
-		if (pack_entry_find_offset(&base_offset, p, (git_oid *)base_info) < GIT_SUCCESS)
+		if (pack_entry_find_offset(&base_offset, &unused, p, (git_oid *)base_info, GIT_OID_HEXSZ) < GIT_SUCCESS)
 			return git__throw(GIT_EPACKCORRUPTED, "Base entry delta is not in the same pack");
 		*curpos += 20;
 	} else
@@ -1367,15 +1461,23 @@ int pack_backend__read(void **buffer_p, size_t *len_p, git_otype *type_p, git_od
 int pack_backend__read_unique_short_oid(git_oid *out_oid, void **buffer_p, size_t *len_p, git_otype *type_p, git_odb_backend *backend,
 					const git_oid *short_oid, unsigned int len)
 {
-	if (len >= GIT_OID_HEXSZ) {
-		int error = pack_backend__read(buffer_p, len_p, type_p, backend, short_oid);
-		if (error == GIT_SUCCESS)
-			git_oid_cpy(out_oid, short_oid);
+	struct pack_entry e;
+	git_rawobj raw;
+	int error;
 
-		return error;
-	} else if (len < GIT_OID_HEXSZ) {
-		return git__throw(GIT_ENOTIMPLEMENTED, "Pack backend cannot search objects from short oid");
-	}
+	if ((error = pack_entry_find_unique_short_oid(&e, (struct pack_backend *)backend, short_oid, len)) < GIT_SUCCESS)
+		return git__rethrow(error, "Failed to read pack backend");
+
+	if ((error = packfile_unpack(&raw, (struct pack_backend *)backend, e.p, e.offset)) < GIT_SUCCESS)
+		return git__rethrow(error, "Failed to read pack backend");
+
+	*buffer_p = raw.data;
+	*len_p = raw.len;
+	*type_p = raw.type;
+	git_oid_cpy(out_oid, &e.sha1);
+
+
+	return GIT_SUCCESS;
 }
 
 int pack_backend__exists(git_odb_backend *backend, const git_oid *oid)
diff --git a/src/sha1_lookup.c b/src/sha1_lookup.c
new file mode 100644
index 000000000..f4a3c42cc
--- /dev/null
+++ b/src/sha1_lookup.c
@@ -0,0 +1,196 @@
+/*
+ * This file is basically taken from git code.
+ * This file is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License, version 2,
+ * as published by the Free Software Foundation.
+ *
+ * In addition to the permissions in the GNU General Public License,
+ * the authors give you unlimited permission to link the compiled
+ * version of this file into combinations with other programs,
+ * and to distribute those combinations without any restriction
+ * coming from the use of this file.  (The General Public License
+ * restrictions do apply in other respects; for example, they cover
+ * modification of the file, and distribution when not linked into
+ * a combined executable.)
+ *
+ * This file is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; see the file COPYING.  If not, write to
+ * the Free Software Foundation, 51 Franklin Street, Fifth Floor,
+ * Boston, MA 02110-1301, USA.
+ */
+
+#include <stdio.h>
+
+#include "sha1_lookup.h"
+#include "common.h"
+
+/*
+ * Conventional binary search loop looks like this:
+ *
+ *	unsigned lo, hi;
+ *      do {
+ *              unsigned mi = (lo + hi) / 2;
+ *              int cmp = "entry pointed at by mi" minus "target";
+ *              if (!cmp)
+ *                      return (mi is the wanted one)
+ *              if (cmp > 0)
+ *                      hi = mi; "mi is larger than target"
+ *              else
+ *                      lo = mi+1; "mi is smaller than target"
+ *      } while (lo < hi);
+ *
+ * The invariants are:
+ *
+ * - When entering the loop, lo points at a slot that is never
+ *   above the target (it could be at the target), hi points at a
+ *   slot that is guaranteed to be above the target (it can never
+ *   be at the target).
+ *
+ * - We find a point 'mi' between lo and hi (mi could be the same
+ *   as lo, but never can be as same as hi), and check if it hits
+ *   the target.  There are three cases:
+ *
+ *    - if it is a hit, we are happy.
+ *
+ *    - if it is strictly higher than the target, we set it to hi,
+ *      and repeat the search.
+ *
+ *    - if it is strictly lower than the target, we update lo to
+ *      one slot after it, because we allow lo to be at the target.
+ *
+ *   If the loop exits, there is no matching entry.
+ *
+ * When choosing 'mi', we do not have to take the "middle" but
+ * anywhere in between lo and hi, as long as lo <= mi < hi is
+ * satisfied.  When we somehow know that the distance between the
+ * target and lo is much shorter than the target and hi, we could
+ * pick mi that is much closer to lo than the midway.
+ *
+ * Now, we can take advantage of the fact that SHA-1 is a good hash
+ * function, and as long as there are enough entries in the table, we
+ * can expect uniform distribution.  An entry that begins with for
+ * example "deadbeef..." is much likely to appear much later than in
+ * the midway of the table.  It can reasonably be expected to be near
+ * 87% (222/256) from the top of the table.
+ *
+ * However, we do not want to pick "mi" too precisely.  If the entry at
+ * the 87% in the above example turns out to be higher than the target
+ * we are looking for, we would end up narrowing the search space down
+ * only by 13%, instead of 50% we would get if we did a simple binary
+ * search.  So we would want to hedge our bets by being less aggressive.
+ *
+ * The table at "table" holds at least "nr" entries of "elem_size"
+ * bytes each.  Each entry has the SHA-1 key at "key_offset".  The
+ * table is sorted by the SHA-1 key of the entries.  The caller wants
+ * to find the entry with "key", and knows that the entry at "lo" is
+ * not higher than the entry it is looking for, and that the entry at
+ * "hi" is higher than the entry it is looking for.
+ */
+int sha1_entry_pos(const void *table,
+		   size_t elem_size,
+		   size_t key_offset,
+		   unsigned lo, unsigned hi, unsigned nr,
+		   const unsigned char *key)
+{
+	const unsigned char *base = table;
+	const unsigned char *hi_key, *lo_key;
+	unsigned ofs_0;
+
+	if (!nr || lo >= hi)
+		return -1;
+
+	if (nr == hi)
+		hi_key = NULL;
+	else
+		hi_key = base + elem_size * hi + key_offset;
+	lo_key = base + elem_size * lo + key_offset;
+
+	ofs_0 = 0;
+	do {
+		int cmp;
+		unsigned ofs, mi, range;
+		unsigned lov, hiv, kyv;
+		const unsigned char *mi_key;
+
+		range = hi - lo;
+		if (hi_key) {
+			for (ofs = ofs_0; ofs < 20; ofs++)
+				if (lo_key[ofs] != hi_key[ofs])
+					break;
+			ofs_0 = ofs;
+			/*
+			 * byte 0 thru (ofs-1) are the same between
+			 * lo and hi; ofs is the first byte that is
+			 * different.
+			 */
+			hiv = hi_key[ofs_0];
+			if (ofs_0 < 19)
+				hiv = (hiv << 8) | hi_key[ofs_0+1];
+		} else {
+			hiv = 256;
+			if (ofs_0 < 19)
+				hiv <<= 8;
+		}
+		lov = lo_key[ofs_0];
+		kyv = key[ofs_0];
+		if (ofs_0 < 19) {
+			lov = (lov << 8) | lo_key[ofs_0+1];
+			kyv = (kyv << 8) | key[ofs_0+1];
+		}
+		assert(lov < hiv);
+
+		if (kyv < lov)
+			return -1 - lo;
+		if (hiv < kyv)
+			return -1 - hi;
+
+		/*
+		 * Even if we know the target is much closer to 'hi'
+		 * than 'lo', if we pick too precisely and overshoot
+		 * (e.g. when we know 'mi' is closer to 'hi' than to
+		 * 'lo', pick 'mi' that is higher than the target), we
+		 * end up narrowing the search space by a smaller
+		 * amount (i.e. the distance between 'mi' and 'hi')
+		 * than what we would have (i.e. about half of 'lo'
+		 * and 'hi').  Hedge our bets to pick 'mi' less
+		 * aggressively, i.e. make 'mi' a bit closer to the
+		 * middle than we would otherwise pick.
+		 */
+		kyv = (kyv * 6 + lov + hiv) / 8;
+		if (lov < hiv - 1) {
+			if (kyv == lov)
+				kyv++;
+			else if (kyv == hiv)
+				kyv--;
+		}
+		mi = (range - 1) * (kyv - lov) / (hiv - lov) + lo;
+
+#ifdef INDEX_DEBUG_LOOKUP
+		printf("lo %u hi %u rg %u mi %u ", lo, hi, range, mi);
+		printf("ofs %u lov %x, hiv %x, kyv %x\n",
+		       ofs_0, lov, hiv, kyv);
+#endif
+
+		if (!(lo <= mi && mi < hi)) {
+			return git__throw(GIT_ERROR, "Assertion failure. Binary search invariant is false");
+		}
+
+		mi_key = base + elem_size * mi + key_offset;
+		cmp = memcmp(mi_key + ofs_0, key + ofs_0, 20 - ofs_0);
+		if (!cmp)
+			return mi;
+		if (cmp > 0) {
+			hi = mi;
+			hi_key = mi_key;
+		} else {
+			lo = mi + 1;
+			lo_key = mi_key + elem_size;
+		}
+	} while (lo < hi);
+	return -lo-1;
+}
diff --git a/src/sha1_lookup.h b/src/sha1_lookup.h
new file mode 100644
index 000000000..5caa2f5ed
--- /dev/null
+++ b/src/sha1_lookup.h
@@ -0,0 +1,12 @@
+#ifndef INCLUDE_sha1_lookup_h__
+#define INCLUDE_sha1_lookup_h__
+
+#include <stdlib.h>
+
+int sha1_entry_pos(const void *table,
+		   size_t elem_size,
+		   size_t key_offset,
+		   unsigned lo, unsigned hi, unsigned nr,
+		   const unsigned char *key);
+
+#endif