NAME
got-worktree
—
Game of Trees work tree
format
DESCRIPTION
A Got
work tree
stores a file hierarchy which corresponds to a versioned snapshot stored in
a Git repository. The work tree's meta data is stored in the
.got directory. A work tree is created with
got checkout
and is required to make changes to a
Git repository with got(1).
A work tree stores the path to its Git repository, the name of a reference to the branch which files were checked out from, and the ID of a commit on this branch known as the base commit.
File meta-data is stored in a structured file called the file index which tracks the status of file modifications, additions, and deletions, relative to the base commit in the repository. The file index contains a series of records, and each such record contains the following status information for a particular file:
- Copy of filesystem meta-data
- Timestamp, file size, and file ownership information from stat(2). This is only used to detect file modifications and is never applied back to the filesystem. File permissions are not tracked, except for the executable bit. When versioned files are checked out into the work tree, the current umask(2) is heeded.
- Blob object ID
- The hash of the blob object which corresponds to the contents of this file in the repository. The hash is stored as binary data. The size of the hash depends on the hashing algorithm used in the repository.
- Commit object ID
- The hash of the commit object the file was checked out from. The hash is stored as binary data. This data is used to detect past incomplete update operations. Entries which do not match the work tree's base commit may still need to be updated to match file content stored in the base commit.
- Flags
- This field contains the length, according to
strlen(3), of path data
which follows, and the following flags:
- STAGE
- Reflects the added, modified, or deleted staged state of a path staged
with
got stage
. - NOT_FLUSHED
- The entry was added to the file index in memory and does not exist in
file index data read from disk. This happens to files which are added
to the work tree while operations such as
got checkout
,got update
,got cherrypick
,got backout
,got rebase
, andgot histedit
are in progress. This flag is always cleared before the entry is written to disk. - NO_BLOB
- The entry's on-disk file content in the work tree is not based on a
blob in the repository. The blob object ID of this entry must be
considered invalid. This happens when unversioned files are added with
got add
and when files are added to the work tree by operations such asgot cherrypick
,got backout
,got rebase
, andgot histedit
. - NO_COMMIT
- The entry is not based on a commit in the repository. The commit
object ID of this entry must be considered invalid. This happens when
unversioned files are added with
got add
and when files are added to the work tree by operations such asgot cherrypick
,got backout
,got rebase
, andgot histedit
. - NO_FILE_ON_DISK
- The entry has no corresponding on-disk file in the work tree. This
happens when files are removed with
got remove
.
- Path data
- The path of the entry, relative to the work tree root. Path data is of variable length and NUL-padded to a multiple of 8 bytes.
- Staged blob object ID
- The hash of a blob object containing file content which has been staged
for commit. The hash is stored as binary data. Only present if a file
addition or modification has been staged with
got stage
.
A corrupt or missing file index can be recreated on demand as follows:
$ mv .got/file-index
.got/file-index.bad
$ got update # re-create
.got/file-index
$ find . -type f -exec touch
{} + # update timestamp of all files
$ got update # sync
timestamps
When the file index is modified, it is read into memory in its entirety, modified in place, and written to a temporary file. This temporary file is then moved on top of the old file index with rename(2). This ensures that no other processes see an inconsistent file index which is in the process of being written.
Work tree meta data must only be modified while the work tree's lock file has been exclusively locked with lockf(3).
Each work tree has a universal unique identifier. When a work tree
is checked out or updated, this identifier is used to create a reference to
the current base commit in the Git repository. The presence of this
reference prevents the Git garbage collector and gotadmin
cleanup
from discarding the base commit and any objects it refers to.
When a work tree is no longer needed, its reference can be deleted from the
Git repository with got ref -d
.
FILES
- .got
- Meta-data directory where all files listed below reside.
- base-commit
- hash digest hex-string representation of the current base commit.
- file-index
- File status information.
- format
- Work tree format number.
- got.conf
- Configuration file for got(1). See got.conf(5).
- head-ref
- Name of the reference to the current branch.
- lock
- Lock file to obtain exclusive write access to meta data.
- path-prefix
- Path inside repository the work tree was checked out from.
- repository
- Path to the repository the work tree was checked out from.
- uuid
- A universal unique identifier for the work tree.
SEE ALSO
got(1), rename(2), stat(2), umask(2), flock(3), git-repository(5), got.conf(5)