UNIQ(1) General Commands Manual UNIQ(1)

uniq — merge or filter adjacent identical lines

uniq [-c] [-u|-d] [-iz] [-f fields] [-s skip] [-w limit] [from [to]]
uniq [-D|-G|--{all-repeated|group}=none|separate|prepend|append|both] [-iz] [-f fields] [-s skip] [-w limit] [from [to]]

Copies consecutive lines from from (the standard input stream if "-", the default) to to (the standard output stream if "-", the default; otherwise created a=rw - umask and truncated – equivalent to shell >):

by default
writing only the first line of each equal sequence,
with -u
writing only locally-unique lines,
with -d
writing only the first of each sequence of duplicates,
with -D
writing each duplicate value in a sequence, potentially separated by empty lines,
with -G
separating equal sequences with empty lines.

By default, the entire line is compared; -f slices off fields leading fields (defined as a maximal series of blanks (spaces or tabs in the C locale) followed by a maximal series of nonblanks), then -s slices off skip leading characters, then -w yields a maximum of limit characters.
The line is always written.

Unless -i, comparisons are byte-wise; otherwise, they're case-insensitive across characters in the current locale (invalid sequences are assumed to have a length of 1 byte and yield the maximum character).

The last of -udDG specified, if any, applies.

, --count
Prepend each written line with the number of lines it had coalesced.

, --unique
Only write lines that are non-equal to their neighbours, i.e. are the sole members of a sequence of length 1.
, --repeated
Write only the first line of each equal sequence longer than 1.
, --all-repeated, --all-repeated=none
Write all lines of each equal sequence longer than 1.
=separate
Likewise, but separate sequences with an empty line.
=prepend
Likewise, but prefix each sequence with an empty line.
=append
Likewise, but suffix each sequence with an empty line.
=both
Likewise, but prefix suffix the first such sequence, suffixing the subsequent ones.
, --group, --group=separate
Write all lines of all sequences, separating sequences with an empty line.
=none
Likewise, but don't insert empty lines. This is mostly equivalent to cat.
=prepend, --group=append, --group=both
Analogous to --all-repeated=.
All --all-repeated and --group values are prefix-matched (--group=b is equivalent to --group=both, &c.).

, --ignore-case
Compare lines case-insensitively according to the current locale.
, --zero-terminated
Line separator is NUL instead of newline.

, --skip-fields=fields
Skip the first fields (decimal) maximal series of blanks then nonblanks for comparison.
, --skip-chars=skip
Skip the first skip (decimal) characters for comparison.
, --check-chars=limit
Compare up to limit (decimal) characters.

1 if from or to couldn't be opened.

Exercise all slicing/comparison options:

$ printf '%s\n' 'a 0ąQ' ' b 1ĄWo' |
  uniq -ci -f1 -s2 -w1
      2 a 0ąQ

sort(1) to make equivalent lines adjacent, or its -u flag, which can uniquify lines based on collation sequence instead of equality.

iswblank(3), mbrtowc(3)

Conforms to IEEE Std 1003.1-2008 (“POSIX.1”), except is allowed for -fs; the standard allows any (or no) number alignment for the -c column — this implementation matches the GNU system at columns and a space, deviating from the AT&T UNIX of 4 and a space. The input file is specified to be a text file, which must not contain NULs: most other implementations terminate the line at the first NUL.

-Dizw, --group are extensions, originating from the GNU system; the -G spelling is an extension; the GNU system forbids --all-repeated=append, --all-repeated=both, and --group=none.

Because -fsw operate on characters, they are not suitable for slicing arbitrary data: set LC_ALL=C (LC_CTYPE, ) to slice by byte (this also replicates the (broken) behaviour of the GNU system; the same applies to -i, questionable though its usefulness in that domain may be).

Appeared in Version 3 AT&T UNIX as uniq(I):

uniq [ -ud ] [ input [ output ] ]
With the default case and both flags described as present-day.

Version 4 AT&T UNIX sees a SYNOPSIS of

uniq [ -udc [ +n ] [ -n ] ] [ input [ output ] ]
with -c always applying the default filter, overriding -ud (if specified), the count aligned to 4 columns, followed by a space, -n is equivalent to present-day -f n, and +n to -s n (though, expectedly, byte-wise). The maximal line size is bytes, unprotected against overflows, and terminating at a NUL.

Version 7 AT&T UNIX exits 1 on failure to open either file and writes the error to the standard error stream.

4.4BSD sees a rewrite, citing IEEE Std 1003.2 (“POSIX.2”), with a SYNOPSIS of

uniq [-c | -d | -u] [-f fields] [-s chars] [input_file [output_file]]
but a much more representative usage string of
usage: uniq [-c | -du] [-f fields] [-s chars] [input [output]]
insofar as -c excludes either of -du, and specifying both -du is equivalent to the default output (curiously, this matches all prior manuals, which read
Note that the normal mode output is the union of the -u and -d mode outputs.
but is unnoted in the rewritten one). The line sizes are now KiB and protected, and the "historic" -n and +n options are undocumented beyond a COMPATIBILITY mention, but recognised for compatibility. Fields are separated not by blanks (isblank(): space (), tab ()) but by whitespace (isspace(): also the vertical tab (), form-feed (), and carriage return ()).

X/Open Portability Guide Issue 2 (“XPG2”) includes Version 4 AT&T UNIX uniq verbatim.

X/Open Portability Guide Issue 3 (“XPG3”) adds APPLICATION USAGE, entirely shaded IN ("Internationalised functionality", defined as optional), of:

In an internationalised environment, the value of the LC_COLLATE environment variable must be equal to the value it had when the input files were sorted.

If uniq does not support selection of collating sequences via LC_COLLATE, the input files must be sorted according to the collating sequence of the "C" locale (see Volume 3, XSI Supplementary Definitions, Chapter 7, C Program Locale).

— indeed, specifying the comparison as maybe current collation, maybe not, and limiting the domain to 7 bits if not, and also weirdly discounting all uses of uniq that aren't in consort with sort. Unsurprisingly, no implementation does this.

IEEE Std 1003.2-1992 (“POSIX.2”) sees largely-present-day uniq with -cdufs, the -n +m syntax marked obsolete, -f defined in terms of blanks from the current locale and -s in terms of characters, likewise, and "-"-as-standard-input-stream is allowed for from, but not for to. from must be a text file — no embedded NULs, lines of up to LINE_MAX bytes, and must end in a newline. No mention is made of collation.

Version 3 of the Single UNIX Specification (“SUSv3”) removes the obsolescent syntax and requires, in ENVIRONMENT VARIABLES:

Determine the locale for ordering rules.
For no apparent reason, considering that the wording remains "repeated", which implies equality, not equivalence, and no mention of ordering is made in the rest of the uniq section.

IEEE Std 1003.1-2008 (“POSIX.1”) allows the obsolete syntax by allowing the option delimiter to be +, allows to being "-" to mean the standard output stream, explicitly discards newlines for comparison (matching existing practice), removes the LC_COLLATE mention and clarifies in EXAMPLES the current guidance that

To remove duplicate lines based on whether they collate equally instead of whether they are identical, applications should use:
sort -u
instead of:
sort | uniq
June 3, 2023 voreutils pre-v0.0.0-latest