swh.model.git_objects module#

Converts SWH model objects to git(-like) objects

Most of the functions in this module take as argument an object from swh.model.model, and format it like a git object.

They are the inverse functions of those in swh.loader.git.converters, but with extensions, as SWH’s model is a superset of Git’s:

  • extensions of existing types (eg. revision/commit and release/tag dates can be expressed with precision up to milliseconds, to support formatting Mercurial objects)

  • new types, for SWH’s specific needs (swh.model.model.RawExtrinsicMetadata and swh.model.model.ExtID)

  • support for somewhat corrupted git objects that we need to reproduce

This is used for two purposes:

  • Format manifests that can be hashed to produce intrinsic identifiers

  • Write git objects to reproduce git repositories that were ingested in the archive.

swh.model.git_objects.content_git_object(content: Content) bytes[source]#

Formats a content as a git blob.

A content’s identifier is the blob sha1 à la git of the tagged content.

swh.model.git_objects.directory_entry_sort_key(entry: DirectoryEntry)[source]#

The sorting key for tree entries


Escape the newlines present in snippet according to git rules.

New lines in git manifests are escaped by indenting the next line by one space.

swh.model.git_objects.format_date(date: Timestamp) bytes[source]#

Convert a date object into an UTC timestamp encoded as ascii bytes.

Git stores timestamps as an integer number of seconds since the UNIX epoch.

However, Software Heritage stores timestamps as an integer number of microseconds (postgres type “datetime with timezone”).

Therefore, we print timestamps with no microseconds as integers, and timestamps with microseconds as floating point values. We elide the trailing zeroes from microsecond values, to “future-proof” our representation if we ever need more precision in timestamps.


Normalize a time representation for processing by Software Heritage

This function supports a numeric timestamp (representing a number of seconds since the UNIX epoch, 1970-01-01 at 00:00 UTC), a datetime.datetime object (with timezone information), or a normalized Software Heritage time representation (idempotency).


time_representation – the representation of a timestamp


a normalized dictionary with three keys:

  • timestamp: a dict with two optional keys:

    • seconds: the integral number of seconds since the UNIX epoch

    • microseconds: the integral number of microseconds

  • offset: the timezone offset as a number of minutes relative to UTC

  • negative_utc: a boolean representing whether the offset is -0000 when offset = 0.

Return type:


swh.model.git_objects.directory_git_object(directory: Dict | Directory) bytes[source]#

Formats a directory as a git tree.

A directory’s identifier is the tree sha1 à la git of a directory listing, using the following algorithm, which is equivalent to the git algorithm for trees:

  1. Entries of the directory are sorted using the name (or the name with ‘/’ appended for directory entries) as key, in bytes order.

  2. For each entry of the directory, the following bytes are output:

  • the octal representation of the permissions for the entry (stored in the ‘perms’ member), which is a representation of the entry type:

    • b’100644’ (int 33188) for files

    • b’100755’ (int 33261) for executable files

    • b’120000’ (int 40960) for symbolic links

    • b’40000’ (int 16384) for directories

    • b’160000’ (int 57344) for references to revisions

  • an ascii space (b’ ‘)

  • the entry’s name (as raw bytes), stored in the ‘name’ member

  • a null byte (b’')

  • the 20 byte long identifier of the object pointed at by the entry, stored in the ‘target’ member:

    • for files or executable files: their blob sha1_git

    • for symbolic links: the blob sha1_git of a file containing the link destination

    • for directories: their intrinsic identifier

    • for revisions: their intrinsic identifier

(Note that there is no separator between entries)

swh.model.git_objects.format_git_object_from_headers(git_type: str, headers: Iterable[Tuple[bytes, bytes]], message: bytes | None = None) bytes[source]#

Format a git_object comprised of a git header and a manifest, which is itself a sequence of headers, and an optional message.

The git_object format, compatible with the git format for tag and commit objects, is as follows:

  • for each key, value in headers, emit:

    • the key, literally

    • an ascii space (\x20)

    • the value, with newlines escaped using escape_newlines(),

    • an ascii newline (\x0a)

  • if the message is not None, emit:

    • an ascii newline (\x0a)

    • the message, literally

  • headers – a sequence of key/value headers stored in the manifest;

  • message – an optional message used to trail the manifest.


the formatted git_object as bytes

swh.model.git_objects.format_git_object_from_parts(git_type: str, parts: Iterable[bytes]) bytes[source]#

Similar to format_git_object_from_headers(), but for manifests made of a flat list of entries, instead of key-value + message, ie. trees and snapshots.

swh.model.git_objects.format_author_data(author: Person, date_offset: TimestampWithTimezone | None) bytes[source]#

Format authorship data according to git standards.

Git authorship data has two components:

  • an author specification, usually a name and email, but in practice an arbitrary bytestring

  • optionally, a timestamp with a UTC offset specification

The authorship data is formatted thus:

`name and email`[ `timestamp` `utc_offset`]

The timestamp is encoded as a (decimal) number of seconds since the UNIX epoch (1970-01-01 at 00:00 UTC). As an extension to the git format, we support fractional timestamps, using a dot as the separator for the decimal part.

The utc offset is a number of minutes encoded as ‘[+-]HHMM’. Note that some tools can pass a negative offset corresponding to the UTC timezone (‘-0000’), which is valid and is encoded as such.


the byte string containing the authorship data

swh.model.git_objects.revision_git_object(revision: Dict | Revision) bytes[source]#

Formats a revision as a git tree.

The fields used for the revision identifier computation are:

  • directory

  • parents

  • author

  • author_date

  • committer

  • committer_date

  • extra_headers or metadata -> extra_headers

  • message

A revision’s identifier is the ‘git’-checksum of a commit manifest constructed as follows (newlines are a single ASCII newline character):

tree <directory identifier>
[for each parent in parents]
parent <parent identifier>
[end for each parents]
author <author> <author_date>
committer <committer> <committer_date>
[for each key, value in extra_headers]
<key> <encoded value>
[end for each extra_headers]


The directory identifier is the ascii representation of its hexadecimal encoding.

Author and committer are formatted using the Person.fullname attribute only. Dates are formatted with the format_offset() function.

Extra headers are an ordered list of [key, value] pairs. Keys are strings and get encoded to utf-8 for identifier computation. Values are either byte strings, unicode strings (that get encoded to utf-8), or integers (that get encoded to their utf-8 decimal representation).

Multiline extra header values are escaped by indenting the continuation lines with one ascii space.

If the message is None, the manifest ends with the last header. Else, the message is appended to the headers after an empty line.

The checksum of the full manifest is computed using the ‘commit’ git object type.

swh.model.git_objects.target_type_to_git(target_type: ObjectType) bytes[source]#

Convert a software heritage target type to a git object type

swh.model.git_objects.release_git_object(release: Dict | Release) bytes[source]#
swh.model.git_objects.snapshot_git_object(snapshot: Dict | Snapshot, *, ignore_unresolved: bool = False) bytes[source]#

Formats a snapshot as a git-like object.

Snapshots are a set of named branches, which are pointers to objects at any level of the Software Heritage DAG.

As well as pointing to other objects in the Software Heritage DAG, branches can also be alias*es, in which case their target is the name of another branch in the same snapshot, or *dangling, in which case the target is unknown (and represented by the None value).

A snapshot identifier is a salted sha1 (using the git hashing algorithm with the snapshot object type) of a manifest following the algorithm:

  1. Branches are sorted using the name as key, in bytes order.

  2. For each branch, the following bytes are output:

  • the type of the branch target:

    • content, directory, revision, release or snapshot for the corresponding entries in the DAG;

    • alias for branches referencing another branch;

    • dangling for dangling branches

  • an ascii space (\x20)

  • the branch name (as raw bytes)

  • a null byte (\x00)

  • the length of the target identifier, as an ascii-encoded decimal number (20 for current intrinsic identifiers, 0 for dangling branches, the length of the target branch name for branch aliases)

  • a colon (:)

  • the identifier of the target object pointed at by the branch, stored in the ‘target’ member:

    • for contents: their sha1_git

    • for directories, revisions, releases or snapshots: their intrinsic identifier

    • for branch aliases, the name of the target branch (as raw bytes)

    • for dangling branches, the empty string

Note that, akin to directory manifests, there is no separator between entries. Because of symbolic branches, identifiers are of arbitrary length but are length-encoded to avoid ambiguity.


ignore_unresolved – if False (the default), raises an exception when alias branches point to non-existing branches

swh.model.git_objects.raw_extrinsic_metadata_git_object(metadata: Dict | RawExtrinsicMetadata) bytes[source]#

Formats RawExtrinsicMetadata as a git-like object.

A raw_extrinsic_metadata identifier is a salted sha1 (using the git hashing algorithm with the raw_extrinsic_metadata object type) of a manifest following the format:

target $ExtendedSwhid
discovery_date $Timestamp
authority $StrWithoutSpaces $IRI
fetcher $Str $Version
format $StrWithoutSpaces
origin $IRI                         <- optional
visit $IntInDecimal                 <- optional
snapshot $CoreSwhid                 <- optional
release $CoreSwhid                  <- optional
revision $CoreSwhid                 <- optional
path $Bytes                         <- optional
directory $CoreSwhid                <- optional


$IRI must be RFC 3987 IRIs (so they may contain newlines, that are escaped as described below)

$StrWithoutSpaces and $Version are ASCII strings, and may not contain spaces.

$Str is an UTF-8 string.

$CoreSwhid are core SWHIDs, as defined in SoftWare Heritage persistent IDentifiers (SWHIDs). $ExtendedSwhid is a core SWHID, with extra types allowed (‘ori’ for origins and ‘emd’ for raw extrinsic metadata)

$Timestamp is a decimal representation of the rounded-down integer number of seconds since the UNIX epoch (1970-01-01 00:00:00 UTC), with no leading ‘0’ (unless the timestamp value is zero) and no timezone. It may be negative by prefixing it with a ‘-’, which must not be followed by a ‘0’.

Newlines in $Bytes, $Str, and $Iri are escaped as with other git fields, ie. by adding a space after them.

swh.model.git_objects.extid_git_object(extid: ExtID) bytes[source]#

Formats an extid as a gi-like object.

An ExtID identifier is a salted sha1 (using the git hashing algorithm with the extid object type) of a manifest following the format:

` extid_type $StrWithoutSpaces [extid_version $Str] extid $Bytes target $CoreSwhid [payload_type $StrWithoutSpaces] [payload $ContentIdentifier] `

$StrWithoutSpaces is an ASCII string, and may not contain spaces.

Newlines in $Bytes are escaped as with other git fields, ie. by adding a space after them.

The extid_version line is only generated if the version is non-zero.

The payload_type and payload lines are only generated if they are not None. $ContentIdentifier is the object ID of a content object.