Initial frame for implementing read_tree using pure python. As git-read-tree can do much more than we can ( and faster assumably ), the .new method is used to create new index instances from up to 3 trees.

Implemented multi-tree traversal to facilitate building a stage list more efficiently ( although I am not sure whether it could be faster to use a dictionary together with some intensive lookup ), including test
Added performance to learn how fast certain operations are, and whether one should be preferred over another

Author: Byron

lib/git/index/base.py

@@ -133,7 +133,6 @@ def _set_cache_(self, attr):
 	def _index_path(self):
 		return join_path_native(self.repo.git_dir, "index")
 
-
 	@property
 	def path(self):
 		""" :return: Path to the index file we are representing """
@@ -240,6 +239,26 @@ def merge_tree(self, rhs, base=None):
 		self.repo.git.read_tree(args)
 		return self
 
+	@classmethod
+	def new(cls, repo, *tree_sha):
+		""" Merge the given treeish revisions into a new index which is returned.
+		This method behaves like git-read-tree --aggressive when doing the merge.
+
+		:param repo: The repository treeish are located in.
+
+		:param *tree_sha:
+			see ``from_tree``
+
+		:return:
+			New IndexFile instance. Its path will be undefined. 
+			If you intend to write such a merged Index, supply an alternate file_path 
+			to its 'write' method."""
+		base_entries = aggressive_tree_merge(repo.odb, tree_sha)
+		
+		inst = cls(self.repo)
+		raise NotImplementedError("convert to entries")
+
+
 	@classmethod
 	def from_tree(cls, repo, *treeish, **kwargs):
 		"""
@@ -275,8 +294,7 @@ def from_tree(cls, repo, *treeish, **kwargs):
 
 			As the underlying git-read-tree command takes into account the current index,
 			it will be temporarily moved out of the way to assure there are no unsuspected
-			interferences.
-		"""
+			interferences."""
 		if len(treeish) == 0 or len(treeish) > 3:
 			raise ValueError("Please specify between 1 and 3 treeish, got %i" % len(treeish))
 

lib/git/index/fun.py

@@ -12,8 +12,10 @@
 						)
 
 from typ import (
+					BaseIndexEntry,
 					IndexEntry,
-					CE_NAMEMASK
+					CE_NAMEMASK, 
+					CE_STAGESHIFT
 				)
 
 from util import 	(
@@ -23,7 +25,6 @@
 
 from gitdb.base import IStream
 from gitdb.typ import str_tree_type
-from binascii import a2b_hex
 
 __all__ = ('write_cache', 'read_cache', 'write_tree_from_cache', 'entry_key' )
 
@@ -150,6 +151,7 @@ def write_tree_from_cache(entries, odb, sl, si=0):
 	:return: tuple(binsha, list(tree_entry, ...)) a tuple of a sha and a list of 
 		tree entries being a tuple of hexsha, mode, name"""
 	tree_items = list()
+	tree_items_append = tree_items.append
 	ci = sl.start
 	end = sl.stop
 	while ci < end:
@@ -161,7 +163,7 @@ def write_tree_from_cache(entries, odb, sl, si=0):
 		rbound = entry.path.find('/', si)
 		if rbound == -1:
 			# its not a tree
-			tree_items.append((entry.binsha, entry.mode, entry.path[si:]))
+			tree_items_append((entry.binsha, entry.mode, entry.path[si:]))
 		else:
 			# find common base range
 			base = entry.path[si:rbound]
@@ -178,7 +180,7 @@ def write_tree_from_cache(entries, odb, sl, si=0):
 			# enter recursion
 			# ci - 1 as we want to count our current item as well
 			sha, tree_entry_list = write_tree_from_cache(entries, odb, slice(ci-1, xi), rbound+1)
-			tree_items.append((sha, S_IFDIR, base))
+			tree_items_append((sha, S_IFDIR, base))
 
 			# skip ahead
 			ci = xi
@@ -193,5 +195,24 @@ def write_tree_from_cache(entries, odb, sl, si=0):
 	istream = odb.store(IStream(str_tree_type, len(sio.getvalue()), sio))
 	return (istream.binsha, tree_items)
 
+def _tree_entry_to_baseindexentry(tree_entry, stage):
+	return BaseIndexEntry(tree_entry[1], tree_entry[0], stage <<CE_STAGESHIFT, tree_entry[2])
 
+def aggressive_tree_merge(odb, tree_shas):
+	"""
+	:return: list of BaseIndexEntries representing the aggressive merge of the given
+		trees. All valid entries are on stage 0, whereas the conflicting ones are left 
+		on stage 1, 2 or 3, whereas stage 1 corresponds to the common ancestor tree, 
+		2 to our tree and 3 to 'their' tree.
+	:param tree_shas: 1, 2 or 3 trees as identified by their shas"""
+	out = list()
+	out_append = out.append
+	if len(tree_shas) == 1:
+		for entry in traverse_tree_recursive(odb, tree_shas[0]):
+			out_append(_tree_entry_to_baseindexentry(entry, 0))
+		# END for each entry
+	else:
+		raise ValueError("Cannot handle %i trees at once" % len(tree_shas))
+	# END handle tree shas
 
+	return out

lib/git/objects/fun.py

@@ -2,6 +2,9 @@
 
 __all__ = ('tree_to_stream', 'tree_entries_from_data')
 
+from stat import S_ISDIR
+				
+
 def tree_to_stream(entries, write):
 	"""Write the give list of entries into a stream using its write method
 	:param entries: **sorted** list of tuples with (binsha, mode, name)
@@ -64,3 +67,118 @@ def tree_entries_from_data(data):
 		out.append((sha, mode, name))
 	# END for each byte in data stream
 	return out
+	
+	
+def _find_by_name(tree_data, name, is_dir, start_at):
+	"""return data entry matching the given name and tree mode
+	or None.
+	Before the item is returned, the respective data item is set 
+	None in the tree_data list to mark it done"""
+	try:
+		item = tree_data[start_at]
+		if item and  item[2] == name and S_ISDIR(item[1]) == is_dir:
+			tree_data[start_at] = None
+			return item
+	except IndexError:
+		pass
+	# END exception handling
+	for index, item in enumerate(tree_data):
+		if item and item[2] == name and S_ISDIR(item[1]) == is_dir:
+			tree_data[index] = None
+			return item
+		# END if item matches
+	# END for each item
+	return None
+
+def _to_full_path(item, path_prefix):
+	"""Rebuild entry with given path prefix"""
+	if not item:
+		return item
+	return (item[0], item[1], path_prefix+item[2])
+	
+def traverse_trees_recursive(odb, tree_shas, path_prefix):
+	"""
+	:return: list with entries according to the given tree-shas. 
+		The result is encoded in a list
+		of n tuple|None per blob/commit, (n == len(tree_shas)), where 
+		* [0] == 20 byte sha
+		* [1] == mode as int
+		* [2] == path relative to working tree root
+		The entry tuple is None if the respective blob/commit did not 
+		exist in the given tree.
+	:param tree_shas: iterable of shas pointing to trees. All trees must 
+		be on the same level. A tree-sha may be None in which case None
+	:param path_prefix: a prefix to be added to the returned paths on this level, 
+		set it '' for the first iteration
+	:note: The ordering of the returned items will be partially lost"""
+	trees_data = list()
+	nt = len(tree_shas)
+	for tree_sha in tree_shas:
+		if tree_sha is None:
+			data = list()
+		else:
+			data = tree_entries_from_data(odb.stream(tree_sha).read())
+		# END handle muted trees
+		trees_data.append(data)
+	# END for each sha to get data for
+	
+	out = list()
+	out_append = out.append
+	
+	# find all matching entries and recursively process them together if the match
+	# is a tree. If the match is a non-tree item, put it into the result.
+	# Processed items will be set None
+	for ti, tree_data in enumerate(trees_data):
+		for ii, item in enumerate(tree_data):
+			if not item:
+				continue
+			# END skip already done items
+			entries = [ None for n in range(nt) ]
+			entries[ti] = item
+			sha, mode, name = item							# its faster to unpack
+			is_dir = S_ISDIR(mode)							# type mode bits
+			
+			# find this item in all other tree data items
+			# wrap around, but stop one before our current index, hence 
+			# ti+nt, not ti+1+nt
+			for tio in range(ti+1, ti+nt):
+				tio = tio % nt
+				entries[tio] = _find_by_name(trees_data[tio], name, is_dir, ii)
+			# END for each other item data
+			
+			# if we are a directory, enter recursion
+			if is_dir:
+				out.extend(traverse_trees_recursive(odb, [ei[0] for ei in entries if ei], path_prefix+name+'/'))
+			else:
+				out_append(tuple(_to_full_path(e, path_prefix) for e in entries))
+			# END handle recursion
+			
+			# finally mark it done
+			tree_data[ii] = None
+		# END for each item
+		
+		# we are done with one tree, set all its data empty
+		del(tree_data[:])
+	# END for each tree_data chunk
+	return out
+	
+def traverse_tree_recursive(odb, tree_sha, path_prefix):
+	"""
+	:return: list of entries of the tree pointed to by tree_sha. An entry
+		has the following format:
+		* [0] 20 byte sha
+		* [1] mode as int
+		* [2] path relative to the repository
+	:param path_prefix: prefix to prepend to the front of all returned paths"""
+	entries = list()
+	data = tree_entries_from_data(odb.stream(tree_sha).read())
+	
+	# unpacking/packing is faster than accessing individual items
+	for sha, mode, name in data:
+		if S_ISDIR(mode):
+			entries.extend(traverse_tree_recursive(odb, sha, path_prefix+name+'/'))
+		else:
+			entries.append((sha, mode, path_prefix+name))
+	# END for each item
+	
+	return entries

lib/git/repo.py

@@ -71,7 +71,7 @@ class Repo(object):
 	# represents the configuration level of a configuration file
 	config_level = ("system", "global", "repository")
 
-	def __init__(self, path=None, odbt = GitCmdObjectDB):
+	def __init__(self, path=None, odbt = GitDB):
 		""" Create a new Repo instance
 
 		:param path: is the path to either the root git directory or the bare git repo::

test/git/performance/test_utils.py

@@ -57,3 +57,97 @@ def __init__(self):
 			na = ni * 3
 			print >> sys.stderr, "Accessed %s[x] %i times in %s s ( %f acc / s)" % (cls.__name__, na, elapsed, na / elapsed)
 		# END for each sequence 
+		
+	def test_instantiation(self):
+		ni = 100000
+		max_num_items = 4
+		for mni in range(max_num_items+1):
+			for cls in (tuple, list):
+				st = time()
+				for i in xrange(ni):
+					if mni == 0:
+						cls()
+					elif mni == 1:
+						cls((1,))
+					elif mni == 2:
+						cls((1,2))
+					elif mni == 3:
+						cls((1,2,3))
+					elif mni == 4:
+						cls((1,2,3,4))
+					else:
+						cls(x for x in xrange(mni))
+					# END handle empty cls
+				# END for each item
+				elapsed = time() - st
+				print >> sys.stderr, "Created %i %ss of size %i in %f s ( %f inst / s)" % (ni, cls.__name__, mni, elapsed, ni / elapsed)
+			# END for each type
+		# END for each item count
+		
+		# tuple and tuple direct
+		st = time()
+		for i in xrange(ni):
+			t = (1,2,3,4)
+		# END for each item
+		elapsed = time() - st
+		print >> sys.stderr, "Created %i tuples (1,2,3,4) in %f s ( %f tuples / s)" % (ni, elapsed, ni / elapsed)
+		
+		st = time()
+		for i in xrange(ni):
+			t = tuple((1,2,3,4))
+		# END for each item
+		elapsed = time() - st
+		print >> sys.stderr, "Created %i tuples tuple((1,2,3,4)) in %f s ( %f tuples / s)" % (ni, elapsed, ni / elapsed)
+		
+	def test_unpacking_vs_indexing(self):
+		ni = 1000000
+		list_items = [1,2,3,4]
+		tuple_items = (1,2,3,4)
+		
+		for sequence in (list_items, tuple_items):
+			st = time()
+			for i in xrange(ni):
+				one, two, three, four = sequence
+			# END for eac iteration
+			elapsed = time() - st
+			print >> sys.stderr, "Unpacked %i %ss of size %i in %f s ( %f acc / s)" % (ni, type(sequence).__name__, len(sequence), elapsed, ni / elapsed)
+			
+			st = time()
+			for i in xrange(ni):
+				one, two, three, four = sequence[0], sequence[1], sequence[2], sequence[3]
+			# END for eac iteration
+			elapsed = time() - st
+			print >> sys.stderr, "Unpacked %i %ss of size %i individually in %f s ( %f acc / s)" % (ni, type(sequence).__name__, len(sequence), elapsed, ni / elapsed)
+			
+			st = time()
+			for i in xrange(ni):
+				one, two = sequence[0], sequence[1]
+			# END for eac iteration
+			elapsed = time() - st
+			print >> sys.stderr, "Unpacked %i %ss of size %i individually (2 of 4) in %f s ( %f acc / s)" % (ni, type(sequence).__name__, len(sequence), elapsed, ni / elapsed)
+		# END for each sequence
+		
+	def test_large_list_vs_iteration(self):
+		# what costs more: alloc/realloc of lists, or the cpu strain of iterators ?
+		def slow_iter(ni):
+			for i in xrange(ni):
+				yield i
+		# END slow iter - be closer to the real world
+		
+		# alloc doesn't play a role here it seems 
+		for ni in (500, 1000, 10000, 20000, 40000):
+			st = time()
+			for i in list(xrange(ni)):
+				i
+			# END for each item
+			elapsed = time() - st
+			print >> sys.stderr, "Iterated %i items from list in %f s ( %f acc / s)" % (ni, elapsed, ni / elapsed)
+			
+			st = time()
+			for i in slow_iter(ni):
+				i
+			# END for each item
+			elapsed = time() - st
+			print >> sys.stderr, "Iterated %i items from iterator in %f s ( %f acc / s)" % (ni, elapsed, ni / elapsed)
+		# END for each number of iterations
+