.TH SPLIT 1 2026-10-19 "w64devkit" "w64devkit Manual"
.SH NAME
split \- split a file into pieces
.SH SYNOPSIS
.B split
.RB [ \-b
.IR count ]
.RB [ \-l
.IR count ]
.RB [ \-a
.IR length ]
.RI [ input
.RI [ prefix ]]
.SH DESCRIPTION
.B split
copies
.I input
to a series of new files, called pieces, each holding a fixed number of
lines or bytes.
By default a piece holds 1000 lines.
With no
.IR input ,
or when it is
.BR \- ,
.B split
reads standard input.
To give a
.I prefix
and still read standard input, name the input
.BR \- .
This
.B split
is a busybox applet; see
.BR busybox (1).
.PP
A piece is created when there is data to put in it, and replaces a file
of the same name without warning.
Empty input creates no pieces.
Data is copied byte for byte, and
.B split
writes nothing to standard output.
.SS Piece names
A piece is named
.I prefix
followed by a suffix of lowercase letters.
The prefix is
.B x
by default, and the suffix is 2 letters long unless
.B \-a
says otherwise.
The suffixes run in order from
.B aa
to
.BR az ,
then
.B ba
to
.BR bz ,
and so on to
.BR zz ,
so the names sort in the order of the data.
.PP
Each letter of suffix multiplies the number of pieces by 26: 26 pieces
for 1 letter, 676 for 2, and 17576 for 3.
If there is more data after the last name is used,
.B split
reports
.B suffixes exhausted
and exits with status 1.
The pieces written until then remain.
.PP
The
.I prefix
may include a directory, which must exist.
The prefix and the suffix together must not be longer than 255 bytes,
directory included.
Otherwise
.B split
reports
.B suffix too long
and writes nothing.
.SS Counts
A count is a decimal number.
Hexadecimal, a sign, and white space are not accepted, and leading zeros
are allowed.
A count for
.B \-b
may be followed by one of the lowercase suffixes
.BR b ,
.BR k ,
.BR m ,
or
.BR g ,
which multiply it by 512, 1024, 1048576, or 1073741824.
A count for
.B \-l
cannot have a suffix.
.PP
A count that is malformed or out of range is an error, reported before
the input is opened.
The largest count for
.B \-l
is 9223372036854775807.
A
.B \-b
count, after its suffix multiplies it, may be up to
18446744073709551615, but one above 9223372036854775807 does not work:
.B split
writes empty or partial pieces, and may hang.
.PP
A count of 0 is accepted and is not useful.
.B "\-b\ 0"
creates empty pieces until the names run out, and then fails.
.B "\-l\ 0"
puts all the input in one piece.
.PP
With
.BR \-l ,
a piece holds
.I count
lines, and a line ends at an LF.
A CR is an ordinary byte.
The last piece holds what remains, and a last line without an LF is
written as it is.
With
.BR \-b ,
every piece except the last holds exactly
.I count
bytes, and a piece may end in the middle of a line or of a UTF-8
character.
.SH OPTIONS
Options may follow the operands, unless
.B POSIXLY_CORRECT
is set.
An option's argument may be attached, as in
.BR \-l100 ,
or be the next argument.
.B \-\-
ends the options.
Long options are not accepted.
An unknown option, a missing option argument, or more than two operands
prints the usage message on standard error and exits with status 1.
.TP
.BI "\-b " count
Split by bytes: each piece except the last holds
.I count
bytes.
The count may have a suffix; see
.BR Counts .
.IP
If both
.B \-b
and
.B \-l
are given, the input is split by bytes, and the count is the argument of
whichever came last.
That argument is also checked as a line count, so it cannot have a
suffix if
.B \-l
is given anywhere: in
.BR "\-l\ 5\ \-b\ 1k" ,
the
.B 1k
is an error, and
.B "\-b\ 1k\ \-l\ 5"
splits into 5-byte pieces.
.TP
.BI "\-l " count
Split by lines: each piece except the last holds
.I count
lines.
The default, when neither
.B \-b
nor
.B \-l
is given, is 1000 lines.
.TP
.BI "\-a " length
Make the suffix
.I length
letters long.
The default is 2.
It is a decimal number from 0 to 2147483647, and an error if the name
would be too long; see
.BR "Piece names" .
A length of 0 is not rejected, but there is then no suffix: the first
piece is named
.I prefix
itself, and each later one steps the last character of
.I prefix
up to
.BR z .
With the default prefix that gives at most three pieces, named
.BR x ,
.BR y ,
and
.BR z .
.TP
.B \-\-help
Print a usage message on standard output and exit with status 0.
It is recognized only as the first argument.
.SH ENVIRONMENT
.TP
.B BB_GLOBBING
Unless it is
.BR 0 ,
the C runtime expands wildcards in the operands when
.B split
starts as a new program, which serves commands typed in
.BR cmd.exe .
A busybox process started with it unset sets it to
.B 0
for its children, so commands run from the kit's shell receive their
operands as the shell expanded them.
.TP
.B BB_UMASK
A file mode creation mask in octal.
When it includes the owner write bit,
.BR 200 ,
the pieces that
.B split
creates are read-only.
A later run cannot replace them: it fails with
.BR "Permission denied" .
.TP
.B POSIXLY_CORRECT
If set, the first operand ends the options.
.PP
Variables that affect every busybox applet are described in
.BR busybox (1).
.SH EXIT STATUS
.TP
.B 0
Success.
.TP
.B 1
An invalid option, count, or number of operands, the
.I input
could not be opened or read, the name would be too long, a piece could
not be created or written, or the suffixes ran out.
.PP
Errors are reported on standard error as
.PP
.RS 4
.EX
split: \fImessage\fR
.EE
.RE
.PP
Pieces written before an error remain.
.SH EXAMPLES
Split a log into pieces of 10000 lines, named
.BR part_aa ,
.BR part_ab ,
and so on:
.PP
.RS 4
.EX
$ split \-l 10000 big.log part_
.EE
.RE
.PP
Split a file into 1 MiB pieces, then join them again:
.PP
.RS 4
.EX
$ split \-b 1m archive.zip archive.zip.
$ cat archive.zip.?? >copy.zip
.EE
.RE
.PP
Split standard input into pieces of 3 lines.
The input is named
.B \-
so that a prefix can follow it:
.PP
.RS 4
.EX
$ seq 1 7 | split \-l 3 \- n_
$ cat n_ab
4
5
6
.EE
.RE
.PP
Put the pieces in an existing directory, with 3 letters of suffix:
.PP
.RS 4
.EX
$ split \-a 3 \-l 500 data.csv out/data_
.EE
.RE
.PP
Split a large file into 2 GiB pieces:
.PP
.RS 4
.EX
$ split \-b 2g backup.tar backup.tar.
.EE
.RE
.SH NOTES
Differences from GNU
.BR split :
.IP \(bu 2
The only options are
.BR \-a ,
.BR \-b ,
.BR \-l ,
and
.BR \-\-help .
There are no long options, and none of GNU's
.BR \-C ,
.BR \-d ,
.BR \-e ,
.BR \-n ,
.BR \-t ,
.BR \-u ,
.BR \-x ,
.BR \-\-additional\-suffix ,
.BR \-\-filter ,
or
.BR \-\-verbose .
.IP \(bu 2
Suffixes are always lowercase letters.
Without
.BR \-a ,
GNU
.B split
lengthens them when the names run out, but this one fails after 676
pieces.
.IP \(bu 2
.B "\-a\ 0"
means no suffix, not the default length.
.IP \(bu 2
The only
.B \-b
suffixes are lowercase
.BR b ,
.BR k ,
.BR m ,
and
.BR g .
GNU suffixes such as
.BR K ,
.BR KB ,
and
.B MB
are errors.
.IP \(bu 2
Counts of 0 are not rejected.
.PP
Files and standard streams are binary, so CR LF line endings and Ctrl-Z
characters are copied unchanged, and lines end only at LF.
.PP
File names may use drive letters, backslashes, and forward slashes, in
the
.I input
and in the
.IR prefix .
They may contain any Unicode characters on Windows 10 version 1903 and
later, and are limited to the system code page on earlier versions.
A full path of 260 characters or more, counting the current directory
for a relative name, fails with
.BR "No such file or directory" ,
even when the name passes the 255-byte check.
The device names
.IR /dev/null ,
.IR /dev/zero ,
.IR /dev/urandom ,
.IR /dev/stdin ,
.IR /dev/tty
(the console), and
.I /dev/fd/N
work as an
.I input
on Windows as on Unix.
A directory as the
.I input
fails with
.BR "Permission denied" .
.SH SEE ALSO
.BR busybox (1),
.BR cat (1),
.BR dd (1),
.BR head (1)