tar {utils} | R Documentation |
Create a tar archive.
tar(tarfile, files = NULL, compression = c("none", "gzip", "bzip2", "xz"), compression_level = 6, tar = Sys.getenv("tar"), extra_flags = "")
tarfile |
The pathname of the tar file: tilde expansion (see
|
files |
A character vector of filepaths to be archived: the default is to archive all files under the current directory. |
compression |
character string giving the type of compression to be used (default none). Can be abbreviated. |
compression_level |
integer: the level of compression. Only used for the internal method. |
tar |
character string: the path to the command to be used. If
the command itself contains spaces it needs to be quoted (e.g., by
|
extra_flags |
Any extra flags for an external |
This is either a wrapper for a tar
command or uses an
internal implementation in R. The latter is used if tarfile
is a connection or if the argument tar
is "internal"
or
""
(the ‘factory-fresh’ default). Note that whereas
Unix-alike versions of R set the environment variable TAR, its
value is not the default for this function.
Argument extra_flags
is passed to an external tar
and
so is platform-dependent. Possibly useful values include -h
(follow symbolic links, also -L on some platforms),
--acls, --exclude-backups, --exclude-vcs (and
similar) and on Windows --force-local (so drives can be
included in filepaths: however, this is the default for the
Rtools
tar
). For GNU tar
,
--format=ustar forces a more portable format. (The default is
set at compilation and will be shown at the end of the output from
tar --help
: for version 1.30 ‘out-of-the-box’ it is
--format=gnu, but the manual says the intention is to change
to --format=posix which is the same as pax
–
it was never part of the POSIX standard for tar
and should
not be used.)
For libarchive's bsdtar
, --format=ustar is more
portable than the default.
One issue which can cause an external command to fail is a command
line too long for the system shell: as from R 3.5.0 this is worked
around if the external command is detected to be GNU tar
or
libarchive tar
(aka bsdtar
).
Note that files = '.'
will usually not work with an external
tar
as that would expand the list of files after
tarfile
is created. (It does work with the default internal
method.)
The return code from system
or 0
for the internal
version, invisibly.
The ‘tar’ format no longer has an agreed standard!
‘Unix Standard Tar’ was part of POSIX 1003.1:1998 but has been
removed in favour of pax
, and in any case many common
implementations diverged from the former standard.
Many R platforms use a version of GNU tar
(including
Rtools
on Windows), but the behaviour seems to be changed
with each version. macOS >= 10.6 and FreeBSD use bsdtar
from the libarchive project (but for macOS, a version from 2010), and
commercial Unixes will have their own versions. bsdtar
is
available for many other platforms: macOS up to at least 10.9 had GNU
tar
as gnutar
and other platforms,
e.g. Solaris, have it as gtar
: on a Unix-alike
configure
will try gnutar
and gtar
before tar
.
Known problems arise from
The handling of file paths of more than 100 bytes. These were
unsupported in early versions of tar
, and supported in one
way by POSIX tar
and in another by GNU tar
and
yet another by the POSIX pax
command which
recent tar
programs often support. The internal
implementation warns on paths of more than 100 bytes,
uses the ‘ustar’ way from the 1998 POSIX
standard which supports up to 256 bytes (depending on the path: in
particular the final component is limited to 100 bytes) if possible,
otherwise the GNU way (which is widely supported, including by
untar
).
Most formats do not record the encoding of file paths.
(File) links. tar
was developed on an OS that used
hard links, and physical files that were referred to more than once
in the list of files to be included were included only once, the
remaining instances being added as links. Later a means to include
symbolic links was added. The internal implementation supports
symbolic links (on OSes that support them), only. Of course, the
question arises as to how links should be unpacked on OSes that do
not support them: for regular files file copies can be used.
Names of links in the ‘ustar’ format are restricted to 100
bytes. There is an GNU extension for arbitrarily long link names,
but bsdtar
ignores it. The internal method uses the
GNU extension, with a warning.
Header fields, in particular the padding to be used when fields are not full or not used. POSIX did define the correct behaviour but commonly used implementations did (and still do) not comply.
File sizes. The ‘ustar’ format is restricted to 8GB per (uncompressed) file.
For portability, avoid file paths of more than 100 bytes and all links (especially hard links and symbolic links to directories).
The internal implementation writes only the blocks of 512 bytes
required (including trailing blocks of nuls), unlike GNU tar
which by default pads with nul to a multiple of 20 blocks
(10KB). Implementations which pad differ on whether the block padding
should occur before or after compression (or both): padding was
designed for improved performance on physical tape drives.
The ‘ustar’ format records file modification times to a resolution of 1 second: on file systems with higher resolution it is conventional to discard fractional seconds.
When an external tar
command is used, compressing the tar
archive requires that tar
supports the -z,
-j or -J flag, and may require the appropriate
command (gzip
, bzip2
or xz
) to be
available. For GNU tar
, further compression programs can be
specified by e.g. extra_flags = "-I lz4"
.
Some versions of bsdtar
accept options such as
--lz4, --lzop and --lrzip or an external
compressor via --use-compress-program lz4
: these could
be supplied in extra_flags
.
NetBSD prior to 8.0 used flag --xz rather than -J,
so this should be used via extra_flags = "--xz"
rather
than compression = "xz"
. The commands from OpenBSD and the
Heirloom Toolchest are not documented to support xz
.
The tar
programs in commercial Unixen such as AIX and
Solaris do not support compression.
For users of macOS (formerly OS X). Apple's HFS and HFS+ file systems
have a legacy concept of ‘resource forks’ dating from classic
Mac OS and rarely used nowadays. Apple's version of tar
stores these as separate files in the tarball with names prefixed by
‘._’, and unpacks such files into resource forks (if possible):
other ways of unpacking (including untar
in R) unpack
them as separate files.
When argument tar
is set to the command tar
on macOS,
environment variable COPYFILE_DISABLE=1 is set, which for the
system version of tar
prevents these separate files being
included in the tarball.
https://en.wikipedia.org/wiki/Tar_(file_format),
http://pubs.opengroup.org/onlinepubs/9699919799/utilities/pax.html#tag_20_92_13_06
for the way the POSIX utility pax
handles tar
formats.
https://github.com/libarchive/libarchive/wiki/FormatTar.