API Reference


class git.objects.base.Object(repo, binsha)

Implements an Object which may be Blobs, Trees, Commits and Tags

NULL_BIN_SHA = '\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00'
NULL_HEX_SHA = '0000000000000000000000000000000000000000'
TYPES = ('blob', 'tree', 'commit', 'tag')
Returns:File Object compatible stream to the uncompressed raw data of the object
Note:returned streams must be read in order
Returns:40 byte hex version of our 20 byte binary sha
classmethod new(repo, id)
Returns:New Object instance of a type appropriate to the object type behind id. The id of the newly created object will be a binsha even though the input id may have been a Reference or Rev-Spec
Parameters:id – reference, rev-spec, or hexsha
Note:This cannot be a __new__ method as it would always call __init__ with the input id which is not necessarily a binsha.
classmethod new_from_sha(repo, sha1)
Returns:new object instance of a type appropriate to represent the given binary sha1
Parameters:sha1 – 20 byte binary sha1

Writes our data directly to the given output stream :param ostream: File object compatible stream object. :return: self

type = None
class git.objects.base.IndexObject(repo, binsha, mode=None, path=None)

Base for all objects that can be part of the index file , namely Tree, Blob and SubModule objects

Returns:Absolute path to this index object in the file system ( as opposed to the .path field which is a path relative to the git repository ).

The returned path will be native to the system and contains ‘’ on windows.

Returns:Name portion of the path, effectively being the basename


class git.objects.blob.Blob(repo, binsha, mode=None, path=None)

A Blob encapsulates a git blob object

DEFAULT_MIME_TYPE = 'text/plain'
executable_mode = 33261
file_mode = 33188
Returns:String describing the mime type of this file (based on the filename)
Note:Defaults to ‘text/plain’ in case the actual file type is unknown.
type = 'blob'


class git.objects.commit.Commit(repo, binsha, tree=None, author=None, authored_date=None, author_tz_offset=None, committer=None, committed_date=None, committer_tz_offset=None, message=None, parents=None, encoding=None, gpgsig=None)

Wraps a git Commit object.

This class will act lazily on some of its attributes and will query the value on demand only if it involves calling the git binary.

conf_encoding = 'i18n.commitencoding'
count(paths='', **kwargs)

Count the number of commits reachable from this commit

  • paths – is an optinal path or a list of paths restricting the return value to commits actually containing the paths
  • kwargs – Additional options to be passed to git-rev-list. They must not alter the ouput style of the command, or parsing will yield incorrect results

int defining the number of reachable commits

classmethod create_from_tree(repo, tree, message, parent_commits=None, head=False, author=None, committer=None)

Commit the given tree, creating a commit object.

  • repo – Repo object the commit should be part of
  • tree – Tree object or hex or bin sha the tree of the new commit
  • message – Commit message. It may be an empty string if no message is provided. It will be converted to a string in any case.
  • parent_commits – Optional Commit objects to use as parents for the new commit. If empty list, the commit will have no parents at all and become a root commit. If None , the current head commit will be the parent of the new commit object
  • head – If True, the HEAD will be advanced to the new commit automatically. Else the HEAD will remain pointing on the previous commit. This could lead to undesired results when diffing files.
  • author – The name of the author, optional. If unset, the repository configuration is used to obtain this value.
  • committer – The name of the committer, optional. If unset, the repository configuration is used to obtain this value.

Commit object representing the new commit


Additional information about the committer and Author are taken from the environment or from the git configuration, see git-commit-tree for more information

default_encoding = 'UTF-8'
env_author_date = 'GIT_AUTHOR_DATE'
env_committer_date = 'GIT_COMMITTER_DATE'
classmethod iter_items(repo, rev, paths='', **kwargs)

Find all commits matching the given criteria.

  • repo – is the Repo
  • rev – revision specifier, see git-rev-parse for viable options
  • paths – is an optinal path or list of paths, if set only Commits that include the path or paths will be considered
  • kwargs – optional keyword arguments to git rev-list where max_count is the maximum number of commits to fetch skip is the number of commits to skip since all commits since i.e. ‘1970-01-01’

iterator yielding Commit items

iter_parents(paths='', **kwargs)

Iterate _all_ parents of this commit.

  • paths – Optional path or list of paths limiting the Commits to those that contain at least one of the paths
  • kwargs – All arguments allowed by git-rev-list

Iterator yielding Commit objects which are parents of self

Returns:String describing the commits hex sha based on the closest Reference. Mostly useful for UI purposes

Create a git stat from changes between this commit and its first parent or from all changes done if this is the very first commit.

Returns:First line of the commit message
type = 'commit'


Module containing all object based types.

class git.objects.tag.TagObject(repo, binsha, object=None, tag=None, tagger=None, tagged_date=None, tagger_tz_offset=None, message=None)

Non-Lightweight tag carrying additional information about an object we are pointing to.

type = 'tag'


class git.objects.tree.TreeModifier(cache)

A utility class providing methods to alter the underlying cache in a list-like fashion.

Once all adjustments are complete, the _cache, which really is a refernce to the cache of a tree, will be sorted. Assuring it will be in a serializable state

add(sha, mode, name, force=False)

Add the given item to the tree. If an item with the given name already exists, nothing will be done, but a ValueError will be raised if the sha and mode of the existing item do not match the one you add, unless force is True

  • sha – The 20 or 40 byte sha of the item to add
  • mode – int representing the stat compatible mode of the item
  • force – If True, an item with your name and information will overwrite any existing item with the same name, no matter which information it has


add_unchecked(binsha, mode, name)

Add the given item to the tree, its correctness is assumed, which puts the caller into responsibility to assure the input is correct. For more information on the parameters, see add :param binsha: 20 byte binary sha


Call this method once you are done modifying the tree information. It may be called several times, but be aware that each call will cause a sort operation :return self:

class git.objects.tree.Tree(repo, binsha, mode=16384, path=None)

Tree objects represent an ordered list of Blobs and other Trees.

Tree as a list:

Access a specific blob using the
tree['filename'] notation.

You may as well access by index
blob = tree[0]
blob_id = 8
Returns:list(Blob, ...) list of blobs directly below this tree
Returns:An object allowing to modify the internal cache. This can be used to change the tree’s contents. When done, make sure you call set_done on the tree modifier, or serialization behaviour will be incorrect. See the TreeModifier for more information on how to alter the cache
commit_id = 14

Find the named object in this tree’s contents :return: git.Blob or git.Tree or git.Submodule

Raises KeyError:
 if given file or tree does not exist in tree
traverse(predicate=<function <lambda> at 0x7f7a9bf20398>, prune=<function <lambda> at 0x7f7a9bf20410>, depth=-1, branch_first=True, visit_once=False, ignore_self=1)

For documentation, see util.Traversable.traverse Trees are set to visit_once = False to gain more performance in the traversal

tree_id = 4
Returns:list(Tree, ...) list of trees directly below this tree
type = 'tree'


Module with functions which are supposed to be as fast as possible

git.objects.fun.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) :param write: write method which takes a data string


Reads the binary representation of a tree and returns tuples of Tree items :param data: data block with tree data (as bytes) :return: list(tuple(binsha, mode, tree_relative_path), ...)

git.objects.fun.traverse_trees_recursive(odb, tree_shas, path_prefix)

list with entries according to the given binary 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.

  • 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
  • path_prefix – a prefix to be added to the returned paths on this level, set it ‘’ for the first iteration

The ordering of the returned items will be partially lost

git.objects.fun.traverse_tree_recursive(odb, tree_sha, path_prefix)
Returns:list of entries of the tree pointed to by the binary tree_sha. An entry has the following format: * [0] 20 byte sha * [1] mode as int * [2] path relative to the repository
Parameters:path_prefix – prefix to prepend to the front of all returned paths


class git.objects.submodule.base.Submodule(repo, binsha, mode=None, path=None, name=None, parent_commit=None, url=None, branch_path=None)

Implements access to a git submodule. They are special in that their sha represents a commit in the submodule’s repository which is to be checked out at the path of this instance. The submodule type does not have a string type associated with it, as it exists solely as a marker in the tree and index.

All methods work in bare and non-bare repositories.

classmethod add(repo, name, path, url=None, branch=None, no_checkout=False)

Add a new submodule to the given repository. This will alter the index as well as the .gitmodules file, but will not create a new commit. If the submodule already exists, no matter if the configuration differs from the one provided, the existing submodule will be returned.

  • repo – Repository instance which should receive the submodule
  • name – The name/identifier for the submodule
  • path – repository-relative or absolute path at which the submodule should be located It will be created as required during the repository initialization.
  • url – git-clone compatible URL, see git-clone reference for more information If None, the repository is assumed to exist, and the url of the first remote is taken instead. This is useful if you want to make an existing repository a submodule of anotherone.
  • branch – name of branch at which the submodule should (later) be checked out. The given branch must exist in the remote repository, and will be checked out locally as a tracking branch. It will only be written into the configuration if it not None, which is when the checked out branch will be the one the remote HEAD pointed to. The result you get in these situation is somewhat fuzzy, and it is recommended to specify at least ‘master’ here. Examples are ‘master’ or ‘feature/new’
  • no_checkout – if True, and if the repository has to be cloned manually, no checkout will be performed

The newly created submodule instance


works atomically, such that no change will be done if the repository update fails for instance

Returns:The branch instance that we are to checkout
Raises InvalidGitRepositoryError:
 if our module is not yet checked out
Returns:the name of the branch, which is the shortest possible branch name
Returns:full (relative) path as string to the branch we would checkout from the remote and track
Returns:IterableList(Submodule, ...) an iterable list of submodules instances which are children of this submodule or 0 if the submodule is not checked out
Returns:ConfigReader instance which allows you to qurey the configuration values of this submodule, as provided by the .gitmodules file
Note:The config reader will actually read the data directly from the repository and thus does not need nor care about your working tree.
Note:Should be cached by the caller and only kept as long as needed
Raises IOError:If the .gitmodules file/blob could not be read
config_writer(*args, **kwargs)
Returns:True if the submodule exists, False otherwise. Please note that a submodule may exist (in the .gitmodules file) even though its module doesn’t exist on disk
classmethod iter_items(repo, parent_commit='HEAD')
Returns:iterator yielding Submodule instances available in the given repository
k_default_mode = 57344
k_head_default = 'master'
k_head_option = 'branch'
k_modules_file = '.gitmodules'
module(*args, **kwargs)
Returns:True if our module exists and is a valid git repository. See module() method
move(*args, **kwargs)
Returns:The name of this submodule. It is used to identify it within the .gitmodules file.
Note:by default, the name is the path at which to find the submodule, but in git-python it should be a unique identifier similar to the identifiers used for remotes, which allows to change the path of the submodule easily
Returns:Commit instance with the tree containing the .gitmodules file
Note:will always point to the current head’s commit if it was not set explicitly
remove(*args, **kwargs)
set_parent_commit(commit, check=True)

Set this instance to use the given commit whose tree is supposed to contain the .gitmodules blob.

  • commit – Commit’ish reference pointing at the root_tree
  • check – if True, relatively expensive checks will be performed to verify validity of the submodule.
  • ValueError – if the commit’s tree didn’t contain the .gitmodules blob.
  • ValueError – if the parent commit didn’t store this submodule under the current path


type = 'submodule'
update(recursive=False, init=True, to_latest_revision=False, progress=None, dry_run=False)

Update the repository of this submodule to point to the checkout we point at with the binsha of this instance.

  • recursive – if True, we will operate recursively and update child- modules as well.
  • init – if True, the module repository will be cloned into place if necessary
  • to_latest_revision – if True, the submodule’s sha will be ignored during checkout. Instead, the remote will be fetched, and the local tracking branch updated. This only works if we have a local tracking branch, which is the case if the remote repository had a master branch, or of the ‘branch’ option was specified for this submodule and the branch existed remotely
  • progress – UpdateProgress instance or None of no progress should be shown
  • dry_run – if True, the operation will only be simulated, but not performed. All performed operations are read-only

does nothing in bare repositories


method is definitely not atomic if recurisve is True



Returns:The url to the repository which our module-repository refers to
class git.objects.submodule.base.UpdateProgress

Class providing detailed progress information to the caller who should derive from it and implement the update(...) message

CLONE = 128
FETCH = 256
x = 9


class git.objects.submodule.root.RootModule(repo)

A (virtual) Root of all submodules in the given repository. It can be used to more easily traverse all submodules of the master repository

k_root_name = '__ROOT__'
Returns:the actual repository containing the submodules
update(previous_commit=None, recursive=True, force_remove=False, init=True, to_latest_revision=False, progress=None, dry_run=False)

Update the submodules of this repository to the current HEAD commit. This method behaves smartly by determining changes of the path of a submodules repository, next to changes to the to-be-checked-out commit or the branch to be checked out. This works if the submodules ID does not change. Additionally it will detect addition and removal of submodules, which will be handled gracefully.

  • previous_commit – If set to a commit’ish, the commit we should use as the previous commit the HEAD pointed to before it was set to the commit it points to now. If None, it defaults to HEAD@{1} otherwise
  • recursive – if True, the children of submodules will be updated as well using the same technique
  • force_remove – If submodules have been deleted, they will be forcibly removed. Otherwise the update may fail if a submodule’s repository cannot be deleted as changes have been made to it (see Submodule.update() for more information)
  • init – If we encounter a new module which would need to be initialized, then do it.
  • to_latest_revision – If True, instead of checking out the revision pointed to by this submodule’s sha, the checked out tracking branch will be merged with the newest remote branch fetched from the repository’s origin
  • progress – RootUpdateProgress instance or None if no progress should be sent
  • dry_run – if True, operations will not actually be performed. Progress messages will change accordingly to indicate the WOULD DO state of the operation.
class git.objects.submodule.root.RootUpdateProgress

Utility class which adds more opcodes to the UpdateProgress

REMOVE = 1024
x = 13


Returns:section title used in .gitmodules configuration file
Returns:name of the submodule as parsed from the section name
git.objects.submodule.util.mkhead(repo, path)
Returns:New branch/head instance

Methods with this decorator raise InvalidGitRepositoryError if they encounter a bare repository

git.objects.submodule.util.find_first_remote_branch(remotes, branch_name)

Find the remote branch matching the name of the given branch or raise InvalidGitRepositoryError

class git.objects.submodule.util.SubmoduleConfigParser(*args, **kwargs)

Catches calls to _write, and updates the .gitmodules blob in the index with the new data, if we have written into a stream. Otherwise it will add the local file to the index to make it correspond with the working tree. Additionally, the cache must be cleared

Please note that no mutating method will work in bare mode


Flush changes in our configuration file to the index


Set this instance’s submodule. It must be called before the first write operation begins



Module for general utility functions

Returns:type suitable to handle the given object type name. Use the type to create new instances.
Parameters:object_type_name – Member of TYPES
Raises ValueError:
 In case object_type_name is unknown

Parse the given date as one of the following

  • Git internal format: timestamp offset

  • RFC 2822: Thu, 07 Apr 2005 22:13:13 +0200.

  • ISO 8601 2005-04-07T22:13:13

    The T can be a space as well

Returns:Tuple(int(timestamp_UTC), int(offset)), both in seconds since epoch
Raises ValueError:
 If the format could not be understood
Note:Date can also be YYYY.MM.DD, MM/DD/YYYY and DD.MM.YYYY.

Parse out the actor (author or committer) info from a line like:

author Tom Preston-Werner <tom@mojombo.com> 1191999972 -0700
Returns:[Actor, int_seconds_since_epoch, int_timezone_offset]
class git.objects.util.ProcessStreamAdapter(process, stream_name)

Class wireing all calls to the contained Process instance.

Use this type to hide the underlying process to provide access only to a specified stream. The process is usually wrapped into an AutoInterrupt class to kill it if the instance goes out of scope.

class git.objects.util.Traversable

Simple interface to perforam depth-first or breadth-first traversals into one direction. Subclasses only need to implement one function. Instances of the Subclass must be hashable

list_traverse(*args, **kwargs)
Returns:IterableList with the results of the traversal as produced by traverse()
traverse(predicate=<function <lambda> at 0x7f7a9bef7b18>, prune=<function <lambda> at 0x7f7a9bef7b90>, depth=-1, branch_first=True, visit_once=True, ignore_self=1, as_edge=False)

iterator yieling of items found when traversing self

  • predicate – f(i,d) returns False if item i at depth d should not be included in the result
  • prune – f(i,d) return True if the search should stop at item i at depth d. Item i will not be returned.
  • depth – define at which level the iteration should not go deeper if -1, there is no limit if 0, you would effectively only get self, the root of the iteration i.e. if 1, you would only get the first level of predessessors/successors
  • branch_first – if True, items will be returned branch first, otherwise depth first
  • visit_once – if True, items will only be returned once, although they might be encountered several times. Loops are prevented that way.
  • ignore_self – if True, self will be ignored and automatically pruned from the result. Otherwise it will be the first item to be returned. If as_edge is True, the source of the first edge is None
  • as_edge – if True, return a pair of items, first being the source, second the destinatination, i.e. tuple(src, dest) with the edge spanning from source to destination

As above, but inverses the operation, returning a string that can be used in commit objects


we convert utctz to the timezone in seconds, it is the format time.altzone returns. Git stores it as UTC timezone which has the opposite sign as well, which explains the -1 * ( that was made explicit here ) :param utctz: git utc timezone string, i.e. +0200

Raises ValueError:
 if offset is incorrect
class git.objects.util.Actor(name, email)

Actors hold information about a person acting on the repository. They can be committers and authors or anything with a name and an email as mentioned in the git log entries.

classmethod author(config_reader=None)

Same as committer(), but defines the main author. It may be specified in the environment, but defaults to the committer

classmethod committer(config_reader=None)
Returns:Actor instance corresponding to the configured committer. It behaves similar to the git implementation, such that the environment will override configuration values of config_reader. If no value is set at all, it will be generated
Parameters:config_reader – ConfigReader to use to retrieve the values from in case they are not set in the environment
conf_email = 'email'
conf_name = 'name'
env_author_email = 'GIT_AUTHOR_EMAIL'
env_author_name = 'GIT_AUTHOR_NAME'
env_committer_email = 'GIT_COMMITTER_EMAIL'
env_committer_name = 'GIT_COMMITTER_NAME'
name_email_regex = <_sre.SRE_Pattern object at 0x7f7a9c78c468>
name_only_regex = <_sre.SRE_Pattern object at 0x7f7a9c400e90>


class git.index.base.IndexFile(repo, file_path=None)

Implements an Index that can be manipulated using a native implementation in order to save git command function calls wherever possible.

It provides custom merging facilities allowing to merge without actually changing your index or your working tree. This way you can perform own test-merges based on the index only without having to deal with the working copy. This is useful in case of partial working trees.


The index contains an entries dict whose keys are tuples of type IndexEntry to facilitate access.

You may read the entries dict or manipulate it using IndexEntry instance, i.e.:

index.entries[index.entry_key(index_entry_instance)] = index_entry_instance

Make sure you use index.write() once you are done manipulating the index directly before operating on it using the git command

add(items, force=True, fprogress=<function <lambda> at 0x7f7a9be61b90>, path_rewriter=None, write=True)

Add files from the working tree, specific blobs or BaseIndexEntries to the index.

  • items

    Multiple types of items are supported, types can be mixed within one call. Different types imply a different handling. File paths may generally be relative or absolute.

    • path string
      strings denote a relative or absolute path into the repository pointing to an existing file, i.e. CHANGES, lib/myfile.ext, ‘/home/gitrepo/lib/myfile.ext’.

      Paths provided like this must exist. When added, they will be written into the object database.

      PathStrings may contain globs, such as ‘lib/__init__*’ or can be directories like ‘lib’, the latter ones will add all the files within the dirctory and subdirectories.

      This equals a straight git-add.

      They are added at stage 0

    • Blob or Submodule object
      Blobs are added as they are assuming a valid mode is set. The file they refer to may or may not exist in the file system, but must be a path relative to our repository.

      If their sha is null ( 40*0 ), their path must exist in the file system relative to the git repository as an object will be created from the data at the path. The handling now very much equals the way string paths are processed, except that the mode you have set will be kept. This allows you to create symlinks by settings the mode respectively and writing the target of the symlink directly into the file. This equals a default Linux-Symlink which is not dereferenced automatically, except that it can be created on filesystems not supporting it as well.

      Please note that globs or directories are not allowed in Blob objects.

      They are added at stage 0

    • BaseIndexEntry or type
      Handling equals the one of Blob objects, but the stage may be explicitly set. Please note that Index Entries require binary sha’s.
  • forceCURRENTLY INEFFECTIVE If True, otherwise ignored or excluded files will be added anyway. As opposed to the git-add command, we enable this flag by default as the API user usually wants the item to be added even though they might be excluded.
  • fprogress – Function with signature f(path, done=False, item=item) called for each path to be added, one time once it is about to be added where done==False and once after it was added where done=True. item is set to the actual item we handle, either a Path or a BaseIndexEntry Please note that the processed path is not guaranteed to be present in the index already as the index is currently being processed.
  • path_rewriter – Function with signature (string) func(BaseIndexEntry) function returning a path for each passed entry which is the path to be actually recorded for the object created from entry.path. This allows you to write an index which is not identical to the layout of the actual files on your hard-disk. If not None and items contain plain paths, these paths will be converted to Entries beforehand and passed to the path_rewriter. Please note that entry.path is relative to the git repository.
  • write – If True, the index will be written once it was altered. Otherwise the changes only exist in memory and are not available to git commands.

List(BaseIndexEntries) representing the entries just actually added.

Raises OSError:

if a supplied Path did not exist. Please note that BaseIndexEntry Objects that do not have a null sha will be added even if their paths do not exist.

checkout(*args, **kwargs)
commit(message, parent_commits=None, head=True, author=None, committer=None)

Commit the current default index file, creating a commit object. For more information on the arguments, see tree.commit.

Note:If you have manually altered the .entries member of this instance, don’t forget to write() your changes to disk beforehand.
Returns:Commit object representing the new commit
diff(*args, **kwargs)
classmethod entry_key(*entry)
classmethod from_tree(repo, *treeish, **kwargs)

Merge the given treeish revisions into a new index which is returned. The original index will remain unaltered

  • repo – The repository treeish are located in.
  • treeish – One, two or three Tree Objects, Commits or 40 byte hexshas. The result changes according to the amount of trees. If 1 Tree is given, it will just be read into a new index If 2 Trees are given, they will be merged into a new index using a two way merge algorithm. Tree 1 is the ‘current’ tree, tree 2 is the ‘other’ one. It behaves like a fast-forward. If 3 Trees are given, a 3-way merge will be performed with the first tree being the common ancestor of tree 2 and tree 3. Tree 2 is the ‘current’ tree, tree 3 is the ‘other’ one
  • kwargs – Additional arguments passed to git-read-tree

New IndexFile instance. It will point to a temporary index location which does not exist anymore. If you intend to write such a merged Index, supply an alternate file_path to its ‘write’ method.


In the three-way merge case, –aggressive will be specified to automatically resolve more cases in a commonly correct manner. Specify trivial=True as kwarg to override that.

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.

iter_blobs(predicate=<function <lambda> at 0x7f7a9be61578>)
Returns:Iterator yielding tuples of Blob objects and stages, tuple(stage, Blob)
Parameters:predicate – Function(t) returning True if tuple(stage, Blob) should be yielded by the iterator. A default filter, the BlobFilter, allows you to yield blobs only if they match a given list of paths.
merge_tree(*args, **kwargs)
move(*args, **kwargs)
classmethod new(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.

  • repo – The repository treeish are located in.
  • tree_sha – 20 byte or 40 byte tree sha or tree objects

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.

Returns:Path to the index file we are representing
remove(*args, **kwargs)
reset(*args, **kwargs)

Resolve the blobs given in blob iterator. This will effectively remove the index entries of the respective path at all non-null stages and add the given blob as new stage null blob.

For each path there may only be one blob, otherwise a ValueError will be raised claiming the path is already at stage 0.

Raises ValueError:
 if one of the blobs already existed at stage 0
Note:You will have to write the index manually once you are done, i.e. index.resolve_blobs(blobs).write()
Returns:Iterator yielding dict(path : list( tuple( stage, Blob, ...))), being a dictionary associating a path in the index with a list containing sorted stage/blob pairs
Note:Blobs that have been removed in one side simply do not exist in the given stage. I.e. a file removed on the ‘other’ branch whose entries are at stage 3 will not have a stage 3 entry.

Reread the contents of our index file, discarding all cached information we might have.

Note:This is a possibly dangerious operations as it will discard your changes to index.entries
write(file_path=None, ignore_tree_extension_data=False)

Write the current state to our file path or to the given one

  • file_path – If None, we will write to our stored file path from which we have been initialized. Otherwise we write to the given file path. Please note that this will change the file_path of this index to the one you gave.
  • ignore_tree_extension_data – If True, the TREE type extension data read in the index will not be written to disk. Use this if you have altered the index and would like to use git-write-tree afterwards to create a tree representing your written changes. If this data is present in the written index, git-write-tree will instead write the stored/cached tree. Alternatively, use IndexFile.write_tree() to handle this case automatically



Writes this index to a corresponding Tree object into the repository’s object database and return it.


Tree object representing this index


The tree will be written even if one or more objects the tree refers to does not yet exist in the object database. This could happen if you added Entries to the index directly.

  • ValueError – if there are no entries in the cache
  • UnmergedEntriesError
exception git.index.base.CheckoutError(message, failed_files, valid_files, failed_reasons)

Thrown if a file could not be checked out from the index as it contained changes.

The .failed_files attribute contains a list of relative paths that failed to be checked out as they contained changes that did not exist in the index.

The .failed_reasons attribute contains a string informing about the actual cause of the issue.

The .valid_files attribute contains a list of relative paths to files that were checked out successfully and hence match the version stored in the index


git.index.fun.write_cache(entries, stream, extension_data=None, ShaStreamCls=<class 'git.util.IndexFileSHA1Writer'>)

Write the cache represented by entries to a stream

  • entriessorted list of entries
  • stream – stream to wrap into the AdapterStreamCls - it is used for final output.
  • ShaStreamCls – Type to use when writing to the stream. It produces a sha while writing to it, before the data is passed on to the wrapped stream
  • extension_data – any kind of data to write as a trailer, it must begin a 4 byte identifier, followed by its size ( 4 bytes )

Read a cache file from the given stream :return: tuple(version, entries_dict, extension_data, content_sha)

  • version is the integer version number

  • entries dict is a dictionary which maps IndexEntry instances to a path

    at a stage

  • extension_data is ‘’ or 4 bytes of type + 4 bytes of size + size bytes

  • content_sha is a 20 byte sha on all cache file contents

git.index.fun.write_tree_from_cache(entries, odb, sl, si=0)

Create a tree from the given sorted list of entries and put the respective trees into the given object database

  • entriessorted list of IndexEntries
  • odb – object database to store the trees in
  • si – start index at which we should start creating subtrees
  • sl – slice indicating the range we should process on the entries list

tuple(binsha, list(tree_entry, ...)) a tuple of a sha and a list of tree entries being a tuple of hexsha, mode, name

Returns:Key suitable to be used for the index.entries dictionary
Parameters:entry – One instance of type BaseIndexEntry or the path and the stage

Convert the given mode from a stat call to the corresponding index mode and return it

git.index.fun.run_commit_hook(name, index)

Run the commit hook of the given name. Silently ignores hooks that do not exist. :param name: name of hook, like ‘pre-commit’ :param index: IndexFile instance :raises HookExecutionError:

git.index.fun.hook_path(name, git_dir)
Returns:path to the given named hook in the given git repository directory


Module with additional types used by the index

class git.index.typ.BlobFilter(paths)

Predicate to be used by iter_blobs allowing to filter only return blobs which match the given list of directories or files.

The given paths are given relative to the repository.

class git.index.typ.BaseIndexEntry

Small Brother of an index entry which can be created to describe changes done to the index in which case plenty of additional information is not requried.

As the first 4 data members match exactly to the IndexEntry type, methods expecting a BaseIndexEntry can also handle full IndexEntries even if they use numeric indices for performance reasons.


binary sha of the blob

Returns:flags stored with this entry
classmethod from_blob(blob, stage=0)
Returns:Fully equipped BaseIndexEntry at the given stage

hex version of our sha


File Mode, compatible to stat module constants

Returns:our path relative to the repository working tree root

Stage of the entry, either:

  • 0 = default stage
  • 1 = stage before a merge or common ancestor entry in case of a 3 way merge
  • 2 = stage of entries from the ‘left’ side of the merge
  • 3 = stage of entries from the right side of the merge
Note:For more information, see http://www.kernel.org/pub/software/scm/git/docs/git-read-tree.html
Returns:Blob using the information of this index entry
class git.index.typ.IndexEntry

Allows convenient access to IndexEntry data without completely unpacking it.

Attributes usully accessed often are cached in the tuple whereas others are unpacked on demand.

See the properties for a mapping between names and tuple indices.

Returns:Tuple(int_time_seconds_since_epoch, int_nano_seconds) of the file’s creation time

Device ID

classmethod from_base(base)
Returns:Minimal entry as created from the given BaseIndexEntry instance. Missing values will be set to null-like values
Parameters:base – Instance of type BaseIndexEntry
classmethod from_blob(blob, stage=0)
Returns:Minimal entry resembling the given blob object

Group ID


Inode ID


See ctime property, but returns modification time

Returns:Uncompressed size of the blob

User ID


Module containing index utilities

class git.index.util.TemporaryFileSwap(file_path)

Utility class moving a file to a temporary location within the same directory and moving it back on to where on object deletion.


Decorator for functions that alter the index using the git command. This would invalidate our possibly existing entries dictionary which is why it must be deleted to allow it to be lazily reread later.

Note:This decorator will not be required once all functions are implemented natively which in fact is possible, but probably not feasible performance wise.

Decorator assuring the wrapped method may only run if we are the default repository index. This is as we rely on git commands that operate on that index only.


Decorator which changes the current working dir to the one of the git repository in order to assure relative paths are handled correctly


class git.cmd.Git(working_dir=None)

The Git class manages communication with the Git binary.

It provides a convenient interface to calling the Git binary, such as in:

g = Git( git_dir )
g.init()                   # calls 'git init' program
rval = g.ls_files()        # calls 'git ls-files' program
Set the GIT_PYTHON_TRACE environment variable print each invocation of the command to stdout. Set its value to ‘full’ to see details about the returned values.
class AutoInterrupt(proc, args)

Kill/Interrupt the stored process instance once this instance goes out of scope. It is used to prevent processes piling up in case iterators stop reading. Besides all attributes are wired through to the contained process object.

The wait method was overridden to perform automatic status code checking and possibly raise.


Wait for the process and return its status code.

Raises GitCommandError:
 if the return status is not 0
class Git.CatFileContentStream(size, stream)

Object representing a sized read-only stream returning the contents of an object. It behaves like a stream, but counts the data read and simulates an empty stream once our sized content region is empty. If not all data is read to the end of the objects’s lifetime, we read the rest to assure the underlying stream continues to work


Clear all kinds of internal caches to release resources.

Currently persistent commands will be interrupted.

Git.execute(command, istream=None, with_keep_cwd=False, with_extended_output=False, with_exceptions=True, as_process=False, output_stream=None, stdout_as_string=True, **subprocess_kwargs)

Handles executing the command on the shell and consumes and returns the returned information (stdout)

  • command – The command argument list to execute. It should be a string, or a sequence of program arguments. The program to execute is the first item in the args sequence or string.
  • istream – Standard input filehandle passed to subprocess.Popen.
  • with_keep_cwd – Whether to use the current working directory from os.getcwd(). The cmd otherwise uses its own working_dir that it has been initialized with if possible.
  • with_extended_output – Whether to return a (status, stdout, stderr) tuple.
  • with_exceptions – Whether to raise an exception when git returns a non-zero status.
  • as_process – Whether to return the created process instance directly from which streams can be read on demand. This will render with_extended_output and with_exceptions ineffective - the caller will have to deal with the details himself. It is important to note that the process will be placed into an AutoInterrupt wrapper that will interrupt the process once it goes out of scope. If you use the command in iterators, you should pass the whole process instance instead of a single stream.
  • output_stream – If set to a file-like object, data produced by the git command will be output to the given stream directly. This feature only has any effect if as_process is False. Processes will always be created with a pipe due to issues with subprocess. This merely is a workaround as data will be copied from the output pipe to the given output stream directly. Judging from the implementation, you shouldn’t use this flag !
  • stdout_as_string – if False, the commands standard output will be bytes. Otherwise, it will be decoded into a string using the default encoding (usually utf-8). The latter can fail, if the output contains binary data.
  • subprocess_kwargs – Keyword arguments to be passed to subprocess.Popen. Please note that some of the valid kwargs are already set by this method, the ones you specify may not be the same ones.

  • str(output) if extended_output = False (Default)
  • tuple(int(status), str(stdout), str(stderr)) if extended_output = True

if ouput_stream is True, the stdout value will be your output stream: * output_stream if extended_output = False * tuple(int(status), output_stream, str(stderr)) if extended_output = True

Note git is executed with LC_MESSAGES=”C” to ensure consitent output regardless of system language.

Raises GitCommandError:


If you add additional keyword arguments to the signature of this method, you must update the execute_kwargs tuple housed in this module.


As get_object_header, but returns object data as well :return: (hexsha, type_string, size_as_int,data_string) :note: not threadsafe


Use this method to quickly examine the type and size of the object behind the given ref.

Note:The method will only suffer from the costs of command invocation once and reuses the command in subsequent calls.
Returns:(hexsha, type_string, size_as_int)
Git.git_exec_name = 'git'
Git.git_exec_name_win = 'git.cmd'
Git.max_chunk_size = 65536

As get_object_header, but returns the data as a stream :return: (hexsha, type_string, size_as_int, stream) :note: This method is not threadsafe, you need one independent Command instance

per thread to be safe !
Git.transform_kwargs(split_single_char_options=False, **kwargs)

Transforms Python style kwargs into git command line options.

Returns:tuple(int, int, int, int) tuple with integers representing the major, minor and additional version numbers as parsed from git version. This value is generated on demand and is cached
Returns:Git directory we are working on


Module containing module parser implementation able to properly read and write configuration files


alias of write

class git.config.SectionConstraint(config, section)

Constrains a ConfigParser to only option commands which are constrained to always use the section we have been initialized with.

It supports all ConfigParser methods that operate on an option


return: Configparser instance we constrain


Equivalent to GitConfigParser.release(), which is called on our underlying parser instance


class git.diff.Diffable

Common interface for all object that can be diffed against another object of compatible type.

Note:Subclasses require a repo member as it is the case for Object instances, for practical reasons we do not derive from Object.
class Index
Diffable.diff(other=<class 'git.diff.Index'>, paths=None, create_patch=False, **kwargs)

Creates diffs between two items being trees, trees and index or an index and the working tree.

  • other – Is the item to compare us with. If None, we will be compared to the working tree. If Treeish, it will be compared against the respective tree If Index ( type ), it will be compared against the index. It defaults to Index to assure the method will not by-default fail on bare repositories.
  • paths – is a list of paths or a single path to limit the diff to. It will only include at least one of the givne path or paths.
  • create_patch – If True, the returned Diff contains a detailed patch that if applied makes the self to other. Patches are somwhat costly as blobs have to be read and diffed.
  • kwargs – Additional arguments passed to git-diff, such as R=True to swap both sides of the diff.



Rename detection will only work if create_patch is True.

On a bare repository, ‘other’ needs to be provided as Index or as as Tree/Commit, or a git command error will occour

class git.diff.DiffIndex

Implements an Index for diffs, allowing a list of Diffs to be queried by the diff properties.

The class improves the diff handling convenience

change_type = ('A', 'D', 'R', 'M')
Returns:iterator yieling Diff instances that match the given change_type

Member of DiffIndex.change_type, namely:

  • ‘A’ for added paths
  • ‘D’ for deleted paths
  • ‘R’ for renamed paths
  • ‘M’ for paths with modified data
class git.diff.Diff(repo, a_path, b_path, a_blob_id, b_blob_id, a_mode, b_mode, new_file, deleted_file, rename_from, rename_to, diff)

A Diff contains diff information between two Trees.

It contains two sides a and b of the diff, members are prefixed with “a” and “b” respectively to inidcate that.

Diffs keep information about the changed blob objects, the file mode, renames, deletions and new files.

There are a few cases where None has to be expected as member variable value:

New File:

a_mode is None
a_blob is None

Deleted File:

b_mode is None
b_blob is None

Working Tree Blobs

When comparing to working trees, the working tree blob will have a null hexsha as a corresponding object does not yet exist. The mode will be null as well. But the path will be available though. If it is listed in a diff the working tree version of the file must be different to the version in the index or tree, and hence has been modified.
NULL_BIN_SHA = '\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00'
NULL_HEX_SHA = '0000000000000000000000000000000000000000'
re_header = <_sre.SRE_Pattern object at 0x2da8080>
Returns:True if the blob of our diff has been renamed


Module containing all exceptions thrown througout the git package,

exception git.exc.CacheError

Base for all errors related to the git index, which is called cache internally

exception git.exc.CheckoutError(message, failed_files, valid_files, failed_reasons)

Thrown if a file could not be checked out from the index as it contained changes.

The .failed_files attribute contains a list of relative paths that failed to be checked out as they contained changes that did not exist in the index.

The .failed_reasons attribute contains a string informing about the actual cause of the issue.

The .valid_files attribute contains a list of relative paths to files that were checked out successfully and hence match the version stored in the index

exception git.exc.GitCommandError(command, status, stderr=None, stdout=None)

Thrown if execution of the git command fails with non-zero status code.

exception git.exc.HookExecutionError(command, status, stdout, stderr)

Thrown if a hook exits with a non-zero exit code. It provides access to the exit code and the string returned via standard output

exception git.exc.InvalidGitRepositoryError

Thrown if the given repository appears to have an invalid format.

exception git.exc.NoSuchPathError

Thrown if a path could not be access by the system.

exception git.exc.UnmergedEntriesError

Thrown if an operation cannot proceed as there are still unmerged entries in the cache


class git.refs.symbolic.SymbolicReference(repo, path)

Represents a special case of a reference such that this reference is symbolic. It does not point to a specific commit, but to another Head, which itself specifies a commit.

A typical example for a symbolic reference is HEAD.


Query or set commits directly

classmethod create(repo, path, reference='HEAD', force=False, logmsg=None)

Create a new symbolic reference, hence a reference pointing to another reference.

  • repo – Repository to create the reference in
  • path – full path at which the new symbolic reference is supposed to be created at, i.e. “NEW_HEAD” or “symrefs/my_new_symref”
  • reference – The reference to which the new symbolic reference should point to. If it is a commit’ish, the symbolic ref will be detached.
  • force – if True, force creation even if a symbolic reference with that name already exists. Raise OSError otherwise
  • logmsg – If not None, the message to append to the reflog. Otherwise no reflog entry is written.

Newly created symbolic Reference

Raises OSError:

If a (Symbolic)Reference with the same name but different contents already exists.


This does not alter the current HEAD, index or Working Tree

classmethod delete(repo, path)

Delete the reference at the given path

  • repo – Repository to delete the reference from
  • path – Short or full path pointing to the reference, i.e. refs/myreference or just “myreference”, hence ‘refs/’ is implied. Alternatively the symbolic reference to be deleted
classmethod dereference_recursive(repo, ref_path)
Returns:hexsha stored in the reference at the given ref_path, recursively dereferencing all intermediate references as required
Parameters:repo – the repository containing the reference at ref_path
classmethod from_path(repo, path)
Parameters:path – full .git-directory-relative path name to the Reference to instantiate
Note:use to_full_path() if you only have a partial path of a known Reference Type
Returns:Instance of type Reference, Head, or Tag depending on the given path
Returns:True if we are a detached reference, hence we point to a specific commit instead to another reference
Returns:True if this symbolic reference points to a remote branch
Returns:True if the reference is valid, hence it can be read and points to a valid object or reference.
classmethod iter_items(repo, common_path=None)

Find all refs in the repository

  • repo – is the Repo
  • common_path – Optional keyword argument to the path which is to be shared by all returned Ref objects. Defaults to class specific portion if None assuring that only refs suitable for the actual class are returned.

git.SymbolicReference[], each of them is guaranteed to be a symbolic ref which is not detached and pointing to a valid ref

List is lexigraphically sorted The returned objects represent actual subclasses, such as Head or TagReference

Returns:RefLog for this reference. Its last entry reflects the latest change applied to this reference


As the log is parsed every time, its recommended to cache it for use instead of calling this method repeatedly. It should be considered read-only.

log_append(oldbinsha, message, newbinsha=None)

Append a logentry to the logfile of this ref

  • oldbinsha – binary sha this ref used to point to
  • message – A message describing the change
  • newbinsha – The sha the ref points to now. If None, our current commit sha will be used

added RefLogEntry instance

Returns:RefLogEntry at the given index
Parameters:index – python list compatible positive or negative index


This method must read part of the reflog during execution, hence it should be used sparringly, or only if you need just one index. In that case, it will be faster than the log() method

Returns:In case of symbolic references, the shortest assumable name is the path itself.

Return the object our ref currently refers to


Returns the Reference we point to


Returns the Reference we point to

rename(new_path, force=False)

Rename self to a new path

  • new_path – Either a simple name or a full path, i.e. new_name or features/new_name. The prefix refs/ is implied for references and will be set as needed. In case this is a symbolic ref, there is no implied prefix
  • force – If True, the rename will succeed even if a head with the target name already exists. It will be overwritten in that case


Raises OSError:

In case a file at path but a different contents already exists

set_commit(commit, logmsg=None)

As set_object, but restricts the type of object to be a Commit

Raises ValueError:
 If commit is not a Commit object or doesn’t point to a commit
set_object(object, logmsg=None)

Set the object we point to, possibly dereference our symbolic reference first. If the reference does not exist, it will be created

  • object – a refspec, a SymbolicReference or an Object instance. SymbolicReferences will be dereferenced beforehand to obtain the object they point to
  • logmsg – If not None, the message will be used in the reflog entry to be written. Otherwise the reflog is not altered

plain SymbolicReferences may not actually point to objects by convention



set_reference(ref, logmsg=None)

Set ourselves to the given ref. It will stay a symbol if the ref is a Reference. Otherwise an Object, given as Object instance or refspec, is assumed and if valid, will be set which effectively detaches the refererence if it was a purely symbolic one.

  • ref – SymbolicReference instance, Object instance or refspec string Only if the ref is a SymbolicRef instance, we will point to it. Everthiny else is dereferenced to obtain the actual object.
  • logmsg

    If set to a string, the message will be used in the reflog. Otherwise, a reflog entry is not written for the changed reference. The previous commit of the entry will be the commit we point to now.

    See also: log_append()




This symbolic reference will not be dereferenced. For that, see set_object(...)

classmethod to_full_path(path)
Returns:string with a full repository-relative path which can be used to initialize a Reference instance, for instance by using Reference.from_path


class git.refs.reference.Reference(repo, path, check_path=True)

Represents a named reference to any object. Subclasses may apply restrictions though, i.e. Heads can only point to commits.

classmethod iter_items(repo, common_path=None)

Equivalent to SymbolicReference.iter_items, but will return non-detached references as well.

Returns:(shortest) Name of this reference - it may contain path components
set_object(object, logmsg=None)

Special version which checks if the head-log needs an update as well


class git.refs.head.HEAD(repo, path='HEAD')

Special case of a Symbolic Reference as it represents the repository’s HEAD reference.

Returns:SymbolicReference pointing at the ORIG_HEAD, which is maintained to contain the previous value of HEAD
reset(commit='HEAD', index=True, working_tree=False, paths=None, **kwargs)

Reset our HEAD to the given commit optionally synchronizing the index and working tree. The reference we refer to will be set to commit as well.

  • commit – Commit object, Reference Object or string identifying a revision we should reset HEAD to.
  • index – If True, the index will be set to match the given commit. Otherwise it will not be touched.
  • working_tree – If True, the working tree will be forcefully adjusted to match the given commit, possibly overwriting uncommitted changes without warning. If working_tree is True, index must be true as well
  • paths – Single path or list of paths relative to the git root directory that are to be reset. This allows to partially reset individual files.
  • kwargs – Additional arguments passed to git-reset.


class git.refs.head.Head(repo, path, check_path=True)

A Head is a named reference to a Commit. Every Head instance contains a name and a Commit object.


>>> repo = Repo("/path/to/repo")
>>> head = repo.heads[0]

>>> head.name

>>> head.commit
<git.Commit "1c09f116cbc2cb4100fb6935bb162daa4723f455">

>>> head.commit.hexsha
checkout(force=False, **kwargs)

Checkout this head by setting the HEAD to this reference, by updating the index to reflect the tree we point to and by updating the working tree to reflect the latest index.

The command will fail if changed working tree files would be overwritten.

  • force – If True, changes to the index and the working tree will be discarded. If False, GitCommandError will be raised in that situation.
  • kwargs – Additional keyword arguments to be passed to git checkout, i.e. b=’new_branch’ to create a new branch at the given spot.

The active branch after the checkout operation, usually self unless a new branch has been created.


By default it is only allowed to checkout heads - everything else will leave the HEAD detached which is allowed and possible, but remains a special state that some tools might not be able to handle.

Returns:A configuration parser instance constrained to only read this instance’s values
Returns:A configuration writer instance with read-and write acccess to options of this head
classmethod delete(repo, *heads, **kwargs)

Delete the given heads :param force:

If True, the heads will be deleted even if they are not yet merged into the main development stream. Default False
k_config_remote = 'remote'
k_config_remote_ref = 'merge'
rename(new_path, force=False)

Rename self to a new path

  • new_path – Either a simple name or a path, i.e. new_name or features/new_name. The prefix refs/heads is implied
  • force – If True, the rename will succeed even if a head with the target name already exists.



respects the ref log as git commands are used

Configure this branch to track the given remote reference. This will alter
this branch’s configuration accordingly.
Parameters:remote_reference – The remote reference to track or None to untrack any references
Returns:The remote_reference we are tracking, or None if we are not a tracking branch


class git.refs.tag.TagReference(repo, path, check_path=True)

Class representing a lightweight tag reference which either points to a commit ,a tag object or any other object. In the latter case additional information, like the signature or the tag-creator, is available.

This tag object will always point to a commit object, but may carray additional information in a tag object:

tagref = TagReference.list_items(repo)[0]
if tagref.tag is not None:
Returns:Commit object the tag ref points to
classmethod create(repo, path, ref='HEAD', message=None, force=False, **kwargs)

Create a new tag reference.

  • path – The name of the tag, i.e. 1.0 or releases/1.0. The prefix refs/tags is implied
  • ref – A reference to the object you want to tag. It can be a commit, tree or blob.
  • message

    If not None, the message will be used in your tag object. This will also create an additional tag object that allows to obtain that information, i.e.:

  • force – If True, to force creation of a tag even though that tag already exists.
  • kwargs – Additional keyword arguments to be passed to git-tag

A new TagReference

classmethod delete(repo, *tags)

Delete the given existing tag or tags

Returns:The object our ref currently refers to. Refs can be cached, they will always point to the actual object as it gets re-created on each query
Returns:Tag object this tag ref points to or None in case we are a light weight tag

alias of TagReference


class git.refs.remote.RemoteReference(repo, path, check_path=True)

Represents a reference pointing to a remote head.

classmethod create(*args, **kwargs)

Used to disable this method

classmethod delete(repo, *refs, **kwargs)

Delete the given remote references. :note:

kwargs are given for compatability with the base class method as we should not narrow the signature.
classmethod iter_items(repo, common_path=None, remote=None)

Iterate remote references, and if given, constrain them to the given remote


class git.refs.log.RefLog(filepath=None)

A reflog contains reflog entries, each of which defines a certain state of the head in question. Custom query methods allow to retrieve log entries by date or by other criteria.

Reflog entries are orded, the first added entry is first in the list, the last entry, i.e. the last change of the head or reference, is last in the list.

classmethod append_entry(config_reader, filepath, oldbinsha, newbinsha, message)

Append a new log entry to the revlog at filepath.

  • config_reader – configuration reader of the repository - used to obtain user information. May also be an Actor instance identifying the committer directly. May also be None
  • filepath – full path to the log file
  • oldbinsha – binary sha of the previous commit
  • newbinsha – binary sha of the current commit
  • message – message describing the change to the reference
  • write – If True, the changes will be written right away. Otherwise the change will not be written

RefLogEntry objects which was appended to the log


As we are append-only, concurrent access is not a problem as we do not interfere with readers.

classmethod entry_at(filepath, index)

RefLogEntry at the given index

  • filepath – full path to the index file from which to read the entry
  • index – python list compatible index, i.e. it may be negative to specifiy an entry counted from the end of the list
Raises IndexError:

If the entry didn’t exist


This method is faster as it only parses the entry at index, skipping all other lines. Nonetheless, the whole file has to be read if the index is negative

classmethod from_file(filepath)
Returns:a new RefLog instance containing all entries from the reflog at the given filepath
Parameters:filepath – path to reflog
Raises ValueError:
 If the file could not be read or was corrupted in some way
classmethod iter_entries(stream)
Returns:Iterator yielding RefLogEntry instances, one for each line read sfrom the given stream.
Parameters:stream – file-like object containing the revlog in its native format or basestring instance pointing to a file to read
classmethod path(ref)
Returns:string to absolute path at which the reflog of the given ref instance would be found. The path is not guaranteed to point to a valid file though.
Parameters:ref – SymbolicReference instance

Write the contents of the reflog instance to a file at the given filepath. :param filepath: path to file, parent directories are assumed to exist


Write this instance’s data to the file we are originating from :return: self

class git.refs.log.RefLogEntry

Named tuple allowing easy access to the revlog data fields


Actor instance, providing access

Returns:a string suitable to be placed in a reflog file
classmethod from_line(line)
Returns:New RefLogEntry instance from the given revlog line.
Parameters:line – line bytes without trailing newline
Raises ValueError:
 If line could not be parsed

Message describing the operation that acted on the reference

classmethod new(oldhexsha, newhexsha, actor, time, tz_offset, message)
Returns:New instance of a RefLogEntry

The hexsha to the commit the ref now points to, after the change


The hexsha to the commit the ref pointed to before the change


time as tuple:

  • [0] = int(time)
  • [1] = int(timezone_offset) in time.altzone format


class git.remote.RemoteProgress

Handler providing an interface to parse progress information emitted by git-push and git-fetch and to dispatch callbacks allowing subclasses to react to the progress.

END = 2
OP_MASK = -4

Called whenever a line could not be understood and was therefore dropped.

Returns:a progress handler suitable for handle_process_output(), passing lines on to this Progress

handler in a suitable format

re_op_absolute = <_sre.SRE_Pattern object at 0x7f7a9c3ccca8>
re_op_relative = <_sre.SRE_Pattern object at 0x2d16820>
update(op_code, cur_count, max_count=None, message='')

Called whenever the progress changes

  • op_code

    Integer allowing to be compared against Operation IDs and stage IDs.

    Stage IDs are BEGIN and END. BEGIN will only be set once for each Operation ID as well as END. It may be that BEGIN and END are set at once in case only one progress message was emitted due to the speed of the operation. Between BEGIN and END, none of these flags will be set

    Operation IDs are all held within the OP_MASK. Only one Operation ID will be active per call.

  • cur_count – Current absolute count of items
  • max_count – The maximum count of items we expect. It may be None in case there is no maximum number of items or if it is (yet) unknown.
  • message – In case of the ‘WRITING’ operation, it contains the amount of bytes transferred. It may possibly be used for other purposes as well.

You may read the contents of the current line in self._cur_line

x = 6
class git.remote.PushInfo(flags, local_ref, remote_ref_string, remote, old_commit=None, summary='')

Carries information about the result of a push operation of a single head:

info = remote.push()[0]
info.flags          # bitflags providing more information about the result
info.local_ref      # Reference pointing to the local reference that was pushed
                    # It is None if the ref was deleted.
info.remote_ref_string # path to the remote reference located on the remote side
info.remote_ref # Remote Reference on the local side corresponding to
                # the remote_ref_string. It can be a TagReference as well.
info.old_commit # commit at which the remote_ref was standing before we pushed
                # it to local_ref.commit. Will be None if an error was indicated
info.summary    # summary line providing human readable english text about the push
ERROR = 1024
UP_TO_DATE = 512
Returns:Remote Reference or TagReference in the local repository corresponding to the remote_ref_string kept in this instance.
x = 10
class git.remote.FetchInfo(ref, flags, note='', old_commit=None)

Carries information about the results of a fetch operation of a single head:

info = remote.fetch()[0]
info.ref           # Symbolic Reference or RemoteReference to the changed
                   # remote head or FETCH_HEAD
info.flags         # additional flags to be & with enumeration members,
                   # i.e. info.flags & info.REJECTED
                   # is 0 if ref is SymbolicReference
info.note          # additional notes given by git-fetch intended for the user
info.old_commit    # if info.flags & info.FORCED_UPDATE|info.FAST_FORWARD,
                   # field is set to the previous location of ref, otherwise None
ERROR = 128
Returns:Commit of our remote ref
Returns:Name of our remote ref
re_fetch_result = <_sre.SRE_Pattern object at 0x2dbd0f0>
x = 7
class git.remote.Remote(repo, name)

Provides easy read and write access to a git remote.

Everything not part of this interface is considered an option for the current remote, allowing constructs like remote.pushurl to query the pushurl.

NOTE: When querying configuration, the configuration accessor will be cached to speed up subsequent accesses.

classmethod add(repo, name, url, **kwargs)

Create a new remote to the given repository :param repo: Repository instance that is to receive the new remote :param name: Desired name of the remote :param url: URL which corresponds to the remote’s name :param kwargs:

Additional arguments to be passed to the git-remote add command
Returns:New Remote instance
Raises GitCommandError:
 in case an origin with that name already exists
Returns:GitConfigParser compatible object able to read options for only our remote. Hence you may simple type config.get(“pushurl”) to obtain the information

GitConfigParser compatible object able to write options for this remote.


You can only own one writer at a time - delete it to release the configuration file and make it useable by others.

To assure consistent results, you should only query options through the writer. Once you are done writing, you are free to use the config reader once again.

classmethod create(repo, name, url, **kwargs)

Create a new remote to the given repository :param repo: Repository instance that is to receive the new remote :param name: Desired name of the remote :param url: URL which corresponds to the remote’s name :param kwargs:

Additional arguments to be passed to the git-remote add command
Returns:New Remote instance
Raises GitCommandError:
 in case an origin with that name already exists
Returns:True if this is a valid, existing remote.

Valid remotes have an entry in the repository’s configuration

fetch(refspec=None, progress=None, **kwargs)

Fetch the latest changes for this remote

  • refspec

    A “refspec” is used by fetch and push to describe the mapping between remote ref and local ref. They are combined with a colon in the format <src>:<dst>, preceded by an optional plus sign, +. For example: git fetch $URL refs/heads/master:refs/heads/origin means “grab the master branch head from the $URL and store it as my origin branch head”. And git push $URL refs/heads/master:refs/heads/to-upstream means “publish my master branch head as to-upstream branch at $URL”. See also git-push(1).

    Taken from the git manual

    Fetch supports multiple refspecs (as the underlying git-fetch does) - supplying a list rather than a string for ‘refspec’ will make use of this facility.

  • progress – See ‘push’ method
  • kwargs – Additional arguments to be passed to git-fetch

IterableList(FetchInfo, ...) list of FetchInfo instances providing detailed information about the fetch results


As fetch does not provide progress information to non-ttys, we cannot make it available here unfortunately as in the ‘push’ method.

classmethod iter_items(repo)
Returns:Iterator yielding Remote objects of the given repository
pull(refspec=None, progress=None, **kwargs)

Pull changes from the given branch, being the same as a fetch followed by a merge of branch with your local branch.

  • refspec – see ‘fetch’ method
  • progress – see ‘push’ method
  • kwargs – Additional arguments to be passed to git-pull

Please see ‘fetch’ method

push(refspec=None, progress=None, **kwargs)

Push changes from source branch in refspec to target branch in refspec.

  • refspec – see ‘fetch’ method
  • progress – Instance of type RemoteProgress allowing the caller to receive progress information until the method returns. If None, progress information will be discarded
  • kwargs – Additional arguments to be passed to git-push

IterableList(PushInfo, ...) iterable list of PushInfo instances, each one informing about an individual head which had been updated on the remote side. If the push contains rejected heads, these will have the PushInfo.ERROR bit set in their flags. If the operation fails completely, the length of the returned IterableList will be null.

Returns:IterableList of RemoteReference objects. It is prefixed, allowing you to omit the remote path portion, i.e.:
remote.refs.master # yields RemoteReference('/refs/remotes/origin/master')
classmethod remove(repo, name)

Remove the remote with the given name


Rename self to the given new_name :return: self

classmethod rm(repo, name)

Remove the remote with the given name

Returns:IterableList RemoteReference objects that do not have a corresponding head in the remote reference anymore as they have been deleted on the remote side, but are still available locally.

The IterableList is prefixed, hence the ‘origin’ must be omitted. See ‘refs’ property for an example.


Fetch all changes for this remote, including new branches which will be forced in ( in case your local remote branch is not part the new remote branches ancestry anymore ).

Parameters:kwargs – Additional arguments passed to git-remote update


class git.repo.base.Repo(path=None, odbt=<class 'gitdb.db.git.GitDB'>, search_parent_directories=False)

Represents a git repository and allows you to query references, gather commit information, generate diffs, create and clone repositories query the log.

The following attributes are worth using:

‘working_dir’ is the working directory of the git command, which is the working tree directory if available or the .git directory in case of bare repositories

‘working_tree_dir’ is the working tree directory, but will raise AssertionError if we are a bare repository.

‘git_dir’ is the .git repository directory, which is always set.

DAEMON_EXPORT_FILE = 'git-daemon-export-ok'

The name of the currently active branch.

Returns:Head to the active branch

Retrieve a list of alternates paths or set a list paths to be used as alternates

archive(ostream, treeish=None, prefix=None, **kwargs)

Archive the tree at the given revision. :parm ostream: file compatible stream object to which the archive will be written as bytes :parm treeish: is the treeish name/id, defaults to active branch :parm prefix: is the optional prefix to prepend to each filename in the archive :parm kwargs:

Additional arguments passed to git-archive NOTE: Use the ‘format’ argument to define the kind of format. Use specialized ostreams to write any format supported by python
Raises GitCommandError:
 in case something went wrong
Returns:True if the repository is bare
blame(rev, file)

The blame information for the given file at the given revision.

Parm rev:revision specifier, see git-rev-parse for viable options.
Returns:list: [git.Commit, list: [<line>]] A list of tuples associating a Commit object with a list of lines that changed within the given commit. The Commit objects will be given in order of appearance.

A list of Head objects representing the branch heads in this repo

Returns:git.IterableList(Head, ...)
clone(path, progress=None, **kwargs)

Create a clone from this repository. :param path:

is the full path of the new repo (traditionally ends with ./<name>.git).
  • progress – See ‘git.remote.Remote.push’.
  • kwargs

    odbt = ObjectDatabase Type, allowing to determine the object database implementation used by the returned Repo instance

    All remaining keyword arguments are given to the git-clone command


git.Repo (the newly cloned repo)

classmethod clone_from(url, to_path, progress=None, **kwargs)

Create a clone from the given URL :param url: valid git url, see http://www.kernel.org/pub/software/scm/git/docs/git-clone.html#URLS :param to_path: Path to which the repository should be cloned to :param progress: See ‘git.remote.Remote.push’. :param kwargs: see the clone method :return: Repo instance pointing to the cloned directory


The Commit object for the specified revision :param rev: revision specifier, see git-rev-parse for viable options. :return: git.Commit

config_level = ('system', 'user', 'global', 'repository')
Returns:GitConfigParser allowing to read the full git configuration, but not to write it

The configuration will include values from the system, user and repository configuration files.

Parameters:config_level – For possible values, see config_writer method If None, all applicable levels will be used. Specify a level in case you know which exact file you whish to read to prevent reading multiple files for instance
Note:On windows, system configuration cannot currently be read as the path is unknown, instead the global path will be used.
Returns:GitConfigParser allowing to write values of the specified configuration file level. Config writers should be retrieved, used to change the configuration ,and written right away as they will lock the configuration file in question and prevent other’s to write it.
Parameters:config_level – One of the following values system = sytem wide configuration file global = user level configuration file repository = configuration file for this repostory only
create_head(path, commit='HEAD', force=False, logmsg=None)

Create a new head within the repository. For more documentation, please see the Head.create method.

Returns:newly created Head Reference
create_remote(name, url, **kwargs)

Create a new remote.

For more information, please see the documentation of the Remote.create methods

Returns:Remote reference
create_submodule(*args, **kwargs)

Create a new submodule

Note:See the documentation of Submodule.add for a description of the applicable parameters
Returns:created submodules
create_tag(path, ref='HEAD', message=None, force=False, **kwargs)

Create a new tag reference. For more documentation, please see the TagReference.create method.

Returns:TagReference object

If True, git-daemon may export this repository

delete_head(*heads, **kwargs)

Delete the given heads

Parameters:kwargs – Additional keyword arguments to be passed to git-branch

Delete the given remote.


Delete the given tag references


the project’s description

Returns:HEAD Object pointing to the current head reference

A list of Head objects representing the branch heads in this repo

Returns:git.IterableList(Head, ...)
Returns:IndexFile representing this repository’s index.
classmethod init(path=None, mkdir=True, **kwargs)

Initialize a git repository at the given path if specified

Parameters:path – is the full path to the repo (traditionally ends with /<name>.git) or None in which case the repository will be created in the current working directory
Parm mkdir:if specified will create the repository directory if it doesn’t already exists. Creates the directory with a mode=0755. Only effective if a path is explicitly given
Parm kwargs:keyword arguments serving as additional options to the git-init command
Returns:git.Repo (the newly created repo)
is_dirty(index=True, working_tree=True, untracked_files=False)
Returns:True, the repository is considered dirty. By default it will react like a git-status without untracked files, hence it is dirty if the index or the working copy have changes.
iter_commits(rev=None, paths='', **kwargs)

A list of Commit objects representing the history of a given ref/commit

Parm rev:revision specifier, see git-rev-parse for viable options. If None, the active branch will be used.
Parm paths:is an optional path or a list of paths to limit the returned commits to Commits that do not contain that path or the paths will not be returned.
Parm kwargs:Arguments to be passed to git-rev-list - common ones are max_count and skip
Note:to receive only commits between two named revisions, use the “revA..revB” revision specifier

:return git.Commit[]

iter_submodules(*args, **kwargs)

An iterator yielding Submodule instances, see Traversable interface for a description of args and kwargs :return: Iterator

iter_trees(*args, **kwargs)
Returns:Iterator yielding Tree objects
Note:Takes all arguments known to iter_commits method
re_author_committer_start = <_sre.SRE_Pattern object at 0x7f7a9bf4bc90>
re_hexsha_only = <_sre.SRE_Pattern object at 0x7f7a9be7c738>
re_hexsha_shortened = <_sre.SRE_Pattern object at 0x7f7a9be7cf08>
re_tab_full_line = <_sre.SRE_Pattern object at 0x7f7a9bf2f660>
re_whitespace = <_sre.SRE_Pattern object at 0x7f7a9be243c0>

A list of Reference objects representing tags, heads and remote references.

Returns:IterableList(Reference, ...)

A list of Reference objects representing tags, heads and remote references.

Returns:IterableList(Reference, ...)
Returns:Remote with the specified name
Raises ValueError:
 if no remote with such a name exists

A list of Remote objects allowing to access and manipulate remotes :return: git.IterableList(Remote, ...)

rev_parse(repo, rev)

Object at the given revision, either Commit, Tag, Tree or Blob


rev – git-rev-parse compatible revision specification as string, please see http://www.kernel.org/pub/software/scm/git/docs/git-rev-parse.html for details


Currently there is no access to the rev-log, rev-specs may only contain topological tokens such ~ and ^.

  • BadObject – if the given revision could not be found
  • ValueError – If rev couldn’t be parsed
  • IndexError – If invalid reflog index is specified
Returns:Submodule with the given name
Raises ValueError:
 If no such submodule exists
submodule_update(*args, **kwargs)

Update the submodules, keeping the repository consistent as it will take the previous state into consideration. For more information, please see the documentation of RootModule.update

Returns:git.IterableList(Submodule, ...) of direct submodules available from the current head
Returns:TagReference Object, reference pointing to a Commit or Tag
Parameters:path – path to the tag reference, i.e. 0.1.5 or tags/0.1.5

A list of Tag objects that are available in this repo :return: git.IterableList(TagReference, ...)


The Tree object for the given treeish revision Examples:

Parameters:rev – is a revision pointing to a Treeish ( being a commit or tree )
Note:If you need a non-root level tree, find it by iterating the root tree. Otherwise it cannot know about its path relative to the repository root and subsequent operations might have unexpected results.

Files currently untracked as they have not been staged yet. Paths are relative to the current working directory of the git command.

Note:ignored files will not appear here, i.e. files mentioned in .gitignore
Returns:The working tree directory of our git repository
Raises AssertionError:
 If we are a bare repository


Package with general repository related functions

git.repo.fun.rev_parse(repo, rev)

Object at the given revision, either Commit, Tag, Tree or Blob


rev – git-rev-parse compatible revision specification as string, please see http://www.kernel.org/pub/software/scm/git/docs/git-rev-parse.html for details


Currently there is no access to the rev-log, rev-specs may only contain topological tokens such ~ and ^.

  • BadObject – if the given revision could not be found
  • ValueError – If rev couldn’t be parsed
  • IndexError – If invalid reflog index is specified

This is taken from the git setup.c:is_git_directory function.

git.repo.fun.name_to_object(repo, name, return_ref=False)
Returns:object specified by the given name, hexshas ( short and long ) as well as references are supported
Parameters:return_ref – if name specifies a reference, we will return the reference instead of the object. Otherwise it will raise BadObject or BadName
git.repo.fun.short_to_long(odb, hexsha)
Returns:long hexadecimal sha1 from the given less-than-40 byte hexsha or None if no candidate could be found.
Parameters:hexsha – hexsha with less than 40 byte

Recursively dereerence a tag and return the resulting object


Convert the given object to a commit if possible and return it


git.util.stream_copy(source, destination, chunk_size=524288)

Copy all data from the source stream into the destination stream in chunks of size chunk_size

Returns:amount of bytes written
git.util.join_path(a, *p)

Join path tokens together similar to os.path.join, but always use ‘/’ instead of possibly ‘’ on windows.

git.util.join_path_native(a, *p)
As join path, but makes sure an OS native path is returned. This is only
needed to play it safe on my dear windows and to assure nice paths that only use ‘’
class git.util.Stats(total, files)

Represents stat information as presented by git at the end of a merge. It is created from the output of a diff operation.


c = Commit( sha1 )
s = c.stats
s.total         # full-stat-dict
s.files         # dict( filepath : stat-dict )


A dictionary with the following keys and values:

deletions = number of deleted lines as int
insertions = number of inserted lines as int
lines = total number of lines changed as int, or deletions + insertions


In addition to the items in the stat-dict, it features additional information:

files = number of changed files as int
class git.util.IndexFileSHA1Writer(f)

Wrapper around a file-like object that remembers the SHA1 of the data written to it. It will write a sha when the stream is closed or if the asked for explicitly usign write_sha.

Only useful to the indexfile

Note:Based on the dulwich project
class git.util.Iterable

Defines an interface for iterable items which is to assure a uniform way to retrieve and iterate items within the git repository

classmethod iter_items(repo, *args, **kwargs)

For more information about the arguments, see list_items :return: iterator yielding Items

classmethod list_items(repo, *args, **kwargs)

Find all items of this type - subclasses can specify args and kwargs differently. If no args are given, subclasses are obliged to return all items if no additional arguments arg given.

Note:Favor the iter_items method as it will

:return:list(Item,...) list of item instances

class git.util.IterableList(id_attr, prefix='')

List of iterable objects allowing to query an object by id or by named index:

heads = repo.heads

It requires an id_attribute name to be set which will be queried from its contained items to have a means for comparison.

A prefix can be specified which is to be used in case the id returned by the items always contains a prefix that does not matter to the user, so it can be left out.

class git.util.BlockingLockFile(file_path, check_interval_s=0.3, max_block_time_s=9223372036854775807)

The lock file will block until a lock could be obtained, or fail after a specified timeout.

Note:If the directory containing the lock was removed, an exception will be raised during the blocking period, preventing hangs as the lock can never be obtained.
class git.util.LockFile(file_path)

Provides methods to obtain, check for, and release a file based lock which should be used to handle concurrent access to the same file.

As we are a utility class to be derived from, we only use protected methods.

Locks will automatically be released on destruction

class git.util.Actor(name, email)

Actors hold information about a person acting on the repository. They can be committers and authors or anything with a name and an email as mentioned in the git log entries.

classmethod author(config_reader=None)

Same as committer(), but defines the main author. It may be specified in the environment, but defaults to the committer

classmethod committer(config_reader=None)
Returns:Actor instance corresponding to the configured committer. It behaves similar to the git implementation, such that the environment will override configuration values of config_reader. If no value is set at all, it will be generated
Parameters:config_reader – ConfigReader to use to retrieve the values from in case they are not set in the environment
conf_email = 'email'
conf_name = 'name'
env_author_email = 'GIT_AUTHOR_EMAIL'
env_author_name = 'GIT_AUTHOR_NAME'
env_committer_email = 'GIT_COMMITTER_EMAIL'
env_committer_name = 'GIT_COMMITTER_NAME'
name_email_regex = <_sre.SRE_Pattern object at 0x7f7a9c78c468>
name_only_regex = <_sre.SRE_Pattern object at 0x7f7a9c400e90>
Returns:string identifying the currently active system user as name@node
git.util.assure_directory_exists(path, is_file=False)

Assure that the directory pointed to by path exists.

Parameters:is_file – If True, path is assumed to be a file and handled correctly. Otherwise it must be a directory
Returns:True if the directory was created, False if it already existed
class git.util.RemoteProgress

Handler providing an interface to parse progress information emitted by git-push and git-fetch and to dispatch callbacks allowing subclasses to react to the progress.

END = 2
OP_MASK = -4

Called whenever a line could not be understood and was therefore dropped.

Returns:a progress handler suitable for handle_process_output(), passing lines on to this Progress

handler in a suitable format

re_op_absolute = <_sre.SRE_Pattern object at 0x7f7a9c3ccca8>
re_op_relative = <_sre.SRE_Pattern object at 0x2d16820>
update(op_code, cur_count, max_count=None, message='')

Called whenever the progress changes

  • op_code

    Integer allowing to be compared against Operation IDs and stage IDs.

    Stage IDs are BEGIN and END. BEGIN will only be set once for each Operation ID as well as END. It may be that BEGIN and END are set at once in case only one progress message was emitted due to the speed of the operation. Between BEGIN and END, none of these flags will be set

    Operation IDs are all held within the OP_MASK. Only one Operation ID will be active per call.

  • cur_count – Current absolute count of items
  • max_count – The maximum count of items we expect. It may be None in case there is no maximum number of items or if it is (yet) unknown.
  • message – In case of the ‘WRITING’ operation, it contains the amount of bytes transferred. It may possibly be used for other purposes as well.

You may read the contents of the current line in self._cur_line

x = 6

Remove the given recursively. :note: we use shutil rmtree but adjust its behaviour to see whether files that

couldn’t be deleted are read-only. Windows will not remove them in that case
class git.util.WaitGroup

WaitGroup is like Go sync.WaitGroup.

Without all the useful corner cases. By Peter Teichman, taken from https://gist.github.com/pteichman/84b92ae7cef0ab98f5a8