refer(1) General Commands Manual refer(1)
Name
refer - process bibliographic references for groff
Synopsis
refer [-bCenPRS] [-a n] [-B field.macro] [-c fields] [-f n] [-i fields]
[-k field] [-l range-expression] [-p database-file] [-s fields]
[-t n] [file ...]
refer --help
refer -v
refer --version
Description
The GNU implementation of refer is part of the groff(1) document
formatting system. refer is a troff(1) preprocessor that prepares
bibliographic citations by looking up keywords specified in a roff(7)
input document, obviating the need to type such annotations, and
permitting the citation style in formatted output to be altered
independently and systematically. It copies the contents of each file
to the standard output stream, looking up each citation between lines
starting with .[ and .] and replacing it with a bibliographic
reference. GNU refer furthermore interprets lines between those
starting with .R1 and .R2 as instructions directing how citations are
to be processed. refer interprets and generates roff lf requests so
that file names and line numbers in messages produced by commands that
read its output correctly describe the source document. Normally,
refer is not executed directly by the user, but invoked by specifying
the -R option to groff(1). If no file operands are present, or if file
is "-", refer reads the standard input stream.
A citation identifies a work by reference to a bibliographic record
detailing it. Select a work from a database of records by listing
keywords that uniquely identify its entry. Alternatively, a document
can specify a record for the work at the point its citation occurs. A
document can use either or both strategies as desired.
For each citation, refer produces a mark in the text, like a
superscripted footnote number or "[Lesk1978a]". A mark consists of a
label between brackets. The mark can be separated from surrounding
text and from other labels in various ways. refer produces roff
language requests usable by a document or a macro package such as me,
mm, mom, or ms to produce a formatted reference for each citation. A
citation's reference can be output immediately after it occurs (as with
footnotes), or references may accumulate, with corresponding output
appearing later in the document (as with endnotes). When references
accumulate, multiple citations of the same reference produce a single
formatted entry.
Interpretation of lines between .R1 and .R2 tokens as preprocessor
commands is a GNU refer extension. Documents employing this feature
can still be processed by AT&T refer by adding the lines
.de R1
.ig R2
..
to the beginning of the document. The foregoing input causes troff to
ignore everything between .R1 and .R2. The effects of some refer
commands can be achieved by command-line options; these are supported
for compatibility with AT&T refer. It is usually more convenient to
use commands.
Bibliographic records
A bibliographic record describes a referenced work in sufficient detail
that it may be cited to accepted standards of scholarly and
professional clarity. The record format permits annotation and
extension that a document may use or ignore. A record is a plain text
sequence of fields, one per line, each consisting of a percent sign %,
an alphanumeric character classifying it, one space, and its contents.
If a field's contents are empty, the field is ignored.
Frequently, such records are organized into a bibliographic database,
with each entry separated by blank lines or file boundaries. This
practice relieves documents of the need to maintain bibliographic data
themselves. The programs lookbib(1) and lkbib(1) consult a
bibliographic database, and indxbib(1) indexes one to speed retrieval
from it, reducing document processing time. Use of these tools is
optional.
The conventional uses of the bibliographic field entries are as
follows. Within a record, fields other than %A and %E replace previous
occurrences thereof. The ordering of multiple %A and %E fields is
significant.
%A names an author. If the name contains a suffix such as "Jr." or
"III", it should be separated from the surname by a comma. We
recommend always supplying an %A field or a %Q field.
%B records the title of the book within which a cited article is
collected. See %J and %T.
%C names the city or other place of publication.
%D indicates the date of publication. Specify the year in full.
If the month is specified, use its name rather than its number;
only the first three letters are required. We recommend always
supplying a %D field; if the date is unknown, use "in press" or
"unknown" as its contents.
%E names an editor of the book within which a cited article is
collected. Where a work has editors but no authors, name the
editors in %A fields and append ", (ed.)" or ", (eds.)" to the
last of these.
%G records the U.S. government ordering number, ISBN, DOI, or other
unique identifier.
%I names the publisher (issuer).
%J records the name of the journal within which a cited article is
collected. See %B and %T.
%K lists keywords intended to aid searches.
%L is a label; typically unused in database entries, it can
override the label format otherwise determined.
%N records the issue number of the journal within which a cited
article is collected.
%O presents additional ("other") information, typically placed at
the end of the reference.
%P lists the page numbers of a cited work that is part of a larger
collection. Specify a range with m-n.
%Q names an institutional author when no %A fields are present.
Only one %Q field is permitted.
%R is an identifier for a report, thesis, memorandum, or other
unpublished work.
%S records the title of a series to which the cited work belongs.
%T is the work's title. See %B and %J.
%V is the volume number of the journal or book containing the cited
work.
%X is an annotation. By convention, it is not formatted in the
citing document.
If the obsolescent "accent strings" feature of the ms or me macro
packages is used, such strings should follow the character to be
accented; an ms document must call the AM macro before using them. Do
not quote accent strings: use one \ rather than two. See groff_char(7)
for a modern approach to the problem of diacritics.
Citations
Citations have a characteristic format.
.[opening-text
flags keyword ...
field
...
.]closing-text
opening-text, closing-text, and flags are optional, and only one
keyword or field need be specified. If keywords are present, refer
searches the bibliographic database(s) for a unique reference matching
them. Multiple matches are an error; add more keywords to disambiguate
the reference. In the absence of keywords, fields constitute the
bibliographic record. Otherwise, fields specify additional data to
replace or supplement those in the reference. When references are
accumulating and keywords are present, specify additional fields at
most on the first citation of a particular reference; they apply to all
further citations thereof.
opening-text and closing-text are roff input used to bracket the label,
overriding the bracket-label command. Leading and trailing spaces are
significant. If either of these is non-empty, the corresponding
argument to the bracket-label command is not used; alter this behavior
with the [ and ] flags.
flags is a list of non-alphanumeric characters each of which modifies
the treatment of the particular citation. AT&T refer ignores them
since they are non-alphanumeric. They direct GNU refer as follows.
# Use the label specified by the short-label command, if any.
refer otherwise uses the normal label. Typically, a short label
implements author-date citation styles consisting of a name, a
year, and a disambiguating letter if necessary. "#" is meant to
suggest such a (quasi-)numeric label.
[ Precede opening-text with the first argument given to the
bracket-label command.
] Follow closing-text with the second argument given to the
bracket-label command.
An advantage of the [ and ] flags over use of opening-text and
closing-text is that you can update the document's bracketing style in
one place using the bracket-label command. Another is that sorting and
merging of citations is not necessarily inhibited if the flags are
used.
refer appends any label resulting from a citation to the roff input
line preceding the .[ token. If there is no such line, refer issues a
warning diagnostic.
There is no special notation for citing multiple references in series.
Use a sequence of citations, one for each reference, with nothing
between them. refer attaches all of their labels to the line preceding
the first. These labels may be sorted or merged. See the description
of the <> label expression, and of the sort-adjacent-labels and
abbreviate-label-ranges commands. A label is not merged if its
citation has a non-empty opening-text or closing-text. However, the
labels for two adjacent citations, the former using the ] flag and
without any closing-text, and the latter using the [ flag and without
any opening-text, may be sorted and merged even if the former's
opening-text or the latter's closing-text is non-empty. (To prevent
these operations, use the dummy character escape sequence \& as the
former's closing-text.)
Commands
Commands are contained between lines starting with .R1 and .R2. The -R
option prevents recognition of these lines. When refer encounters a
.R1 line, it flushes any accumulated references. Neither .R1 nor .R2
lines, nor anything between them, is output.
Commands are separated by newlines or semicolons. A number sign (#)
introduces a comment that extends to the end of the line, but does not
conceal the newline. Each command is broken up into words. Words are
separated by spaces or tabs. A word that begins with a (neutral)
double quote (") extends to the next double quote that is not followed
by another double quote. If there is no such double quote, the word
extends to the end of the line. Pairs of double quotes in a word
beginning with a double quote collapse to one double quote. Neither a
number sign nor a semicolon is recognized inside double quotes. A line
can be continued by ending it with a backslash "\"; this works
everywhere except after a number sign.
Each command name that is marked with * has an associated negative
command no-name that undoes the effect of name. For example, the
no-sort command specifies that references should not be sorted. The
negative commands take no arguments.
In the following description each argument must be a single word; field
is used for a single upper or lower case letter naming a field; fields
is used for a sequence of such letters; m and n are used for a non-
negative numbers; string is used for an arbitrary string; file is used
for the name of a file.
abbreviate* fields string1 string2 string3 string4
Abbreviate the first names of fields. An initial letter will be
separated from another initial letter by string1, from the
surname by string2, and from anything else (such as "von" or
"de") by string3. These default to a period followed by a
space. In a hyphenated first name, the initial of the first
part of the name will be separated from the hyphen by string4;
this defaults to a period. No attempt is made to handle any
ambiguities that might result from abbreviation. Names are
abbreviated before sorting and before label construction.
abbreviate-label-ranges* string
Three or more adjacent labels that refer to consecutive
references will be abbreviated to a label consisting of the
first label, followed by string, followed by the last label.
This is mainly useful with numeric labels. If string is
omitted, it defaults to "-".
accumulate*
Accumulate references instead of writing out each reference as
it is encountered. Accumulated references will be written out
whenever a reference of the form
.[
$LIST$
.]
is encountered, after all input files have been processed, and
whenever a .R1 line is recognized.
annotate* field string
field is an annotation; print it at the end of the reference as
a paragraph preceded by the line
.string
If string is omitted, it will default to AP; if field is also
omitted it will default to X. Only one field can be an
annotation.
articles string ...
Each string is a definite or indefinite article, and should be
ignored at the beginning of T fields when sorting. Initially,
"a", "an", and "the" are recognized as articles.
bibliography file ...
Write out all the references contained in each bibliographic
database file. This command should come last in an .R1/.R2
block.
bracket-label string1 string2 string3
In the text, bracket each label with string1 and string2. An
occurrence of string2 immediately followed by string1 will be
turned into string3. The default behavior is as follows.
bracket-label \*([. \*(.] ", "
capitalize fields
Convert fields to caps and small caps.
compatible*
Recognize .R1 and .R2 even when followed by a character other
than space or newline.
database file ...
Search each bibliographic database file. For each file, if an
index file.i created by indxbib(1) exists, then it will be
searched instead; each index can cover multiple databases.
date-as-label* string
string is a label expression that specifies a string with which
to replace the D field after constructing the label. See
subsection "Label expressions" below for a description of label
expressions. This command is useful if you do not want explicit
labels in the reference list, but instead want to handle any
necessary disambiguation by qualifying the date in some way.
The label used in the text would typically be some combination
of the author and date. In most cases you should also use the
no-label-in-reference command. For example,
date-as-label D.+yD.y%a*D.-y
would attach a disambiguating letter to the year part of the D
field in the reference.
default-database*
The default database should be searched. This is the default
behavior, so the negative version of this command is more
useful. refer determines whether the default database should be
searched on the first occasion that it needs to do a search.
Thus a no-default-database command must be given before then, in
order to be effective.
discard* fields
When the reference is read, fields should be discarded; no
string definitions for fields will be output. Initially, fields
are XYZ.
et-al* string m n
Configure use of "et al" in the evaluation of @ expressions in
label expressions. If u is the number of authors needed to make
the author sequence unambiguous and the total number of authors
is t, then the last t-u authors will be replaced by string
provided that t-u is not less than m and t is not less than n.
The default behavior is as follows.
et-al " et al" 2 3
Note the absence of a dot from the end of the abbreviation,
which is arguably not correct. (Et al[.] is short for et alli,
as etc. is short for et cetera.)
include file
Include file and interpret the contents as commands.
join-authors string1 string2 string3
Join multiple authors together with strings. When there are
exactly two authors, they will be joined with string1. When
there are more than two authors, all but the last two will be
joined with string2, and the last two authors will be joined
with string3. If string3 is omitted, it will default to
string1; if string2 is also omitted it will also default to
string1. For example,
join-authors " and " ", " ", and "
will restore the default method for joining authors.
label-in-reference*
When outputting the reference, define the string [F to be the
reference's label. This is the default behavior, so the
negative version of this command is more useful.
label-in-text*
For each reference output a label in the text. The label will
be separated from the surrounding text as described in the
bracket-label command. This is the default behavior, so the
negative version of this command is more useful.
label string
string is a label expression describing how to label each
reference.
separate-label-second-parts string
When merging two-part labels, separate the second part of the
second label from the first label with string. See the
description of the <> label expression.
move-punctuation*
In the text, move any punctuation at the end of line past the
label. We recommend employing this command unless you are using
superscripted numbers as labels.
reverse* string
Reverse the fields whose names are in string. An optional
integer after a field name limits the number of such fields to
the given count; no integer means no limit.
search-ignore* fields
While searching for keys in databases for which no index exists,
ignore the contents of fields. Initially, fields XYZ are
ignored.
search-truncate* n
Only require the first n characters of keys to be given. In
effect when searching for a given key words in the database are
truncated to the maximum of n and the length of the key.
Initially, n is 6.
short-label* string
string is a label expression that specifies an alternative
(usually shorter) style of label. This is used when the # flag
is given in the citation. When using author-date style labels,
the identity of the author or authors is sometimes clear from
the context, and so it may be desirable to omit the author or
authors from the label. The short-label command will typically
be used to specify a label containing just a date and possibly a
disambiguating letter.
sort* string
Sort references according to string. References will
automatically be accumulated. string should be a list of field
names, each followed by a number, indicating how many fields
with the name should be used for sorting. "+" can be used to
indicate that all the fields with the name should be used. Also
. can be used to indicate the references should be sorted using
the (tentative) label. (Subsection "Label expressions" below
describes the concept of a tentative label.)
sort-adjacent-labels*
Sort labels that are adjacent in the text according to their
position in the reference list. This command should usually be
given if the abbreviate-label-ranges command has been given, or
if the label expression contains a <> expression. This has no
effect unless references are being accumulated.
Label expressions
Label expressions can be evaluated both normally and tentatively. The
result of normal evaluation is used for output. The result of
tentative evaluation, called the tentative label, is used to gather the
information that normal evaluation needs to disambiguate the label.
Label expressions specified by the date-as-label and short-label
commands are not evaluated tentatively. Normal and tentative
evaluation are the same for all types of expression other than @, *,
and % expressions. The description below applies to normal evaluation,
except where otherwise specified.
field [n]
is the nth part of field. If n is omitted, it defaults to 1.
'string'
The characters in string literally.
@ All authors joined as specified by the join-authors command.
The whole of each author's name is used. However, if the
references are sorted by author (that is, the sort
specification starts with "A+"), then authors' surnames will be
used instead, provided that this does not introduce ambiguity,
and also an initial subsequence of the authors may be used
instead of all the authors, again provided that this does not
introduce ambiguity. Given any two referenced works with n
authors, the use of only the surname for the nth author of a
reference is regarded as ambiguous if the other reference
shares the first n-1 authors, the nth authors of each reference
are not identical, but the nth authors' surnames are the same.
A proper initial subsequence of the sequence of authors for
some reference is considered to be ambiguous if there is a
reference with some other sequence of authors which also has
that subsequence as a proper initial subsequence. When an
initial subsequence of authors is used, the remaining authors
are replaced by the string specified by the et-al command; this
command may also specify additional requirements that must be
met before an initial subsequence can be used. @ tentatively
evaluates to a canonical representation of the authors, such
that authors that compare equally for sorting purposes have the
same representation.
%n
%a
%A
%i
%I The serial number of the reference formatted according to the
character following the %. The serial number of a reference
is 1 plus the number of earlier references with same tentative
label as this reference. These expressions tentatively
evaluate to an empty string.
expr* If there is another reference with the same tentative label as
this reference, then expr, otherwise an empty string. It
tentatively evaluates to an empty string.
expr+n
expr-n The first (+) or last (-) n upper or lower case letters or
digits of expr. roff special characters (such as \('a) count
as a single letter. Accent strings are retained but do not
count toward the total.
expr.l expr converted to lowercase.
expr.u expr converted to uppercase.
expr.c expr converted to caps and small caps.
expr.r expr reversed so that the surname is first.
expr.a expr with first names abbreviated. Fields specified in the
abbreviate command are abbreviated before any labels are
evaluated. Thus .a is useful only when you want a field to be
abbreviated in a label but not in a reference.
expr.y The year part of expr.
expr.+y The part of expr before the year, or the whole of expr if it
does not contain a year.
expr.-y The part of expr after the year, or an empty string if expr
does not contain a year.
expr.n The surname part of expr.
expr1~expr2
expr1 except that if the last character of expr1 is - then it
will be replaced by expr2.
expr1 expr2
The catenation of expr1 and expr2.
expr1|expr2
If expr1 is non-empty then expr1 otherwise expr2.
expr1&expr2
If expr1 is non-empty then expr2 otherwise an empty string.
expr1?expr2:expr3
If expr1 is non-empty then expr2 otherwise expr3.
<expr> The label is in two parts, which are separated by expr. Two
adjacent two-part labels which have the same first part will be
merged by appending the second part of the second label onto
the first label separated by the string specified in the
separate-label-second-parts command (initially, a comma
followed by a space); the resulting label will also be a two-
part label with the same first part as before merging, and so
additional labels can be merged into it. It is permissible for
the first part to be empty; this may be desirable for
expressions used in the short-label command.
(expr) The same as expr. Used for grouping.
The above expressions are listed in order of precedence (highest
first); & and | have the same precedence.
Macro interface
Each reference starts with a call to the macro ]-. The string [F will
be defined to be the label for this reference, unless the
no-label-in-reference command has been given. There then follows a
series of string definitions, one for each field: string [X corresponds
to field X. The register [P is set to 1 if the P field contains a
range of pages. The [T, [A and [O registers are set to 1 according as
the T, A and O fields end with any of .?! (an end-of-sentence
character). The [E register will be set to 1 if the [E string contains
more than one name. The reference is followed by a call to the ][
macro. The first argument to this macro gives a number representing
the type of the reference. If a reference contains a J field, it will
be classified as type 1, otherwise if it contains a B field, it will be
type 3, otherwise if it contains a G or R field it will be type 4,
otherwise if it contains an I field it will be type 2, otherwise it
will be type 0. The second argument is a symbolic name for the type:
other, journal-article, book, article-in-book, or tech-report. Groups
of references that have been accumulated or are produced by the
bibliography command are preceded by a call to the ]< macro and
followed by a call to the ]> macro.
Options
--help displays a usage message, while -v and --version show version
information; all exit afterward.
-R Don't recognize lines beginning with .R1/.R2.
Other options are equivalent to refer commands.
-a n reverse An
-b no-label-in-text; no-label-in-reference
-B See below.
-c fields capitalize fields
-C compatible
-e accumulate
-f n label %n
-i fields search-ignore fields
-k label L~%a
-k field label field~%a
-l label A.nD.y%a
-l m label A.n+mD.y%a
-l ,n label A.nD.y-n%a
-l m,n label A.n+mD.y-n%a
-n no-default-database
-p db-file database db-file
-P move-punctuation
-s spec sort spec
-S label "(A.n|Q) ', ' (D.y|D)"; bracket-label " (" ) "; "
-t n search-truncate n
The B option has command equivalents with the addition that the file
names specified on the command line are processed as if they were
arguments to the bibliography command instead of in the normal way.
-B annotate X AP; no-label-in-reference
-B field.macro annotate field macro; no-label-in-reference
Exit status
refer exits with status 0 on successful operation, status 2 if the
program cannot interpret its command-line arguments, and status 1 if it
encounters an error during operation.
Environment
REFER Assign this variable a file name to override the default
database.
Files
/usr/dict/papers/Ind
Default database.
file.i Index files.
/opt/local/share/groff/1.24.1/tmac/refer.tmac
defines macros and strings facilitating integration with macro
packages that wish to support refer.
refer uses temporary files. See the groff(1) man page for details of
where such files are created.
Bugs
In label expressions, <> expressions are ignored inside .char
expressions.
Examples
We can illustrate the operation of refer with a sample bibliographic
database containing one entry and a simple roff document to cite that
entry.
$ cat > my-db-file
%A Daniel P.\& Friedman
%A Matthias Felleisen
%C Cambridge, Massachusetts
%D 1996
%I The MIT Press
%T The Little Schemer, Fourth Edition
$ refer -p my-db-file
Read the book
.[
friedman
.]
on your summer vacation.
<Control+D>
.lf 1 -
Read the book\*([.1\*(.]
.ds [F 1
.]-
.ds [A Daniel P. Friedman and Matthias Felleisen
.ds [C Cambridge, Massachusetts
.ds [D 1996
.ds [I The MIT Press
.ds [T The Little Schemer, Fourth Edition
.nr [T 0
.nr [A 0
.][ 2 book
.lf 5 -
on your summer vacation.
The foregoing shows us that refer (a) produces a label "1"; (b)
brackets that label with interpolations of the "[." and ".]" strings;
(c) calls a macro "]-"; (d) defines strings and registers containing
the label and bibliographic data for the reference; (e) calls a macro
"]["; and (f) uses the lf request to restore the line numbers of the
input. As noted in subsection "Macro interface" above, it is the
document's responsibility to employ and format the information
usefully. Let us see how we might turn groff_ms(7) to this task.
$ REFER=my-db-file groff -R -ms
.LP
Read the book
.[
friedman
.]
on your summer vacation.
Commentary is available.\*{*\*}
.FS \*{*\*}
Space reserved for penetrating insight.
.FE
ms's automatic footnote numbering mechanism is not aware of refer's
label numbering, so we have manually specified a (superscripted)
symbolic footnote for our non-bibliographic aside.
See also
"Refer -- A Bibliography System", by Bill Tuthill, 1983, Computing
Services, University of California, Berkeley.
"Some Applications of Inverted Indexes on the Unix System", by M. E.
Lesk, 1978, AT&T Bell Laboratories Computing Science Technical Report
No. 69.
indxbib(1), lookbib(1), lkbib(1)
groff 1.24.1 2026-05-15 refer(1)
groff 1.24.1 - Generated Mon May 18 15:13:13 CDT 2026