Scroll to navigation

refer(1) General Commands Manual refer(1)


refer - process bibliographic references for groff


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


The GNU implementation of refer is part of the groff(1) document formatting system. refer is a troff(1) preprocessor that prepares bibilographic 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, except that it interprets lines between .[ and .] as citations to be translated into groff input, and lines between .R1 and .R2 as instructions regarding how citations are to be processed. Normally, refer is not executed directly by the user, but invoked by specifying the -R option to groff(1). If no file operands are given on the command line, or if file is “-”, the standard input stream is read.

Each citation specifies a reference. The citation can specify a reference that is contained in a bibliographic database by giving a set of keywords that only that reference contains. Alternatively it can specify a reference by supplying a database record in the citation. A combination of these alternatives is also possible.

For each citation, refer can produce a mark in the text. This mark consists of some label which can be separated from the text and from other labels in various ways. For each reference it also outputs groff(7) language commands that can be used by a macro package to produce a formatted reference for each citation. The output of refer must therefore be processed using a suitable macro package, such as me, mm, mom, or ms. The commands to format a citation's reference can be output immediately after the citation, or the references may be accumulated, and the commands output at some later point. If the references are accumulated, then multiple citations of the same reference will produce a single formatted reference.

The interpretation of lines between .R1 and .R2 as prepreocessor commands is a feature of GNU refer. Documents making use of this feature can still be processed by AT&T refer just by adding the lines

.de R1
.ig R2
to the beginning of the document. This will cause troff(1) to ignore everything between .R1 and .R2. The effect of some commands can also be achieved by options. These options are supported mainly for compatibility with AT&T refer. It is usually more convenient to use commands.

refer generates .lf requests so that file names and line numbers in messages produced by commands that read refer output will be correct; it also interprets lines beginning with .lf so that file names and line numbers in the messages and .lf lines that it produces will be accurate even if the input has been preprocessed by a command such as soelim(1).

Bibliographic databases

The bibliographic database is a text file consisting of records separated by one or more blank lines. Within each record fields start with a % at the beginning of a line. Each field has a one character name that immediately follows the %. It is best to use only upper and lower case letters for the names of fields. The name of the field should be followed by exactly one space, and then by the contents of the field. Empty fields are ignored. The conventional meaning of each field is as follows:

The name of an author. If the name contains a suffix such as “Jr.”, it should be separated from the last name by a comma. There can be multiple occurrences of the %A field. The order is significant. It is a good idea always to supply an %A field or a %Q field.
For an article that is part of a book, the title of the book.
The place (city) of publication.
The date of publication. The year should be specified in full. If the month is specified, the name rather than the number of the month should be used, but only the first three letters are required. It is a good idea always to supply a %D field; if the date is unknown, a value such as in press or unknown can be used.
For an article that is part of a book, the name of an editor of the book. Where the work has editors and no authors, the names of the editors should be given as %A fields and “, (ed.)” or “, (eds.)” should be appended to the last author.
U.S. government ordering number.
The publisher (issuer).
For an article in a journal, the name of the journal.
Keywords to be used for searching.
Journal issue number.
Other information. This is usually printed at the end of the reference.
Page number. A range of pages can be specified as m-n.
The name of the author, if the author is not a person. This will only be used if there are no %A fields. There can only be one %Q field.
Technical report number.
Series name.
Title. For an article in a book or journal, this should be the title of the article.
Volume number of the journal or book.

For all fields except %A and %E, if there is more than one occurrence of a particular field in a record, only the last such field will be used.

If accent strings are used, they should follow the character to be accented. This means that an ms document must call the .AM macro when it initializes. Accent strings should not be quoted: use one \ rather than two. Accent strings are an obsolescent feature of the me and ms macro packages; modern documents should use groff special character escape sequences instead; see groff_char(7).


Citations have a characteristic format.

flags keywords

The opening-text, closing-text, and flags components are optional. Only one of the keywords and fields components need be specified.

The keywords component says to search the bibliographic databases for a reference that contains all the words in keywords. It is an error if more than one reference is found.

The fields components specifies additional fields to replace or supplement those specified in the reference. When references are being accumulated and the keywords component is non-empty, then additional fields should be specified only on the first occasion that a particular reference is cited, and will apply to all citations of that reference.

The opening-text and closing-text components specify strings to be used to bracket the label instead of those in the bracket-label command. If either of these components is non-empty, the strings specified in the bracket-label command will not be used; this behavior can be altered using the [ and ] flags. Leading and trailing spaces are significant for these components.

The flags component is a list of non-alphanumeric characters each of which modifies the treatment of this particular citation. AT&T refer will treat these flags as part of the keywords and so will ignore them since they are non-alphanumeric. The following flags are currently recognized.

Use the label specified by the short-label command, instead of that specified by the label command. If no short label has been specified, the normal label will be used. Typically the short label is used with author-date labels and consists of only the date and possibly a disambiguating letter; the “#” is supposed to be suggestive of a numeric type of label.
Precede opening-text with the first string specified in the bracket-label command.
Follow closing-text with the second string specified in the bracket-label command.

An advantage of using the [ and ] flags rather than including the brackets in opening-text and closing-text is that you can change the style of bracket used in the document just by changing the bracket-label command. Another is that sorting and merging of citations will not necessarily be inhibited if the flags are used.

If a label is to be inserted into the text, it will be attached to the line preceding the .[ line. If there is no such line, then an extra line will be inserted before the .[ line and a warning will be given.

There is no special notation for making a citation to multiple references. Just use a sequence of citations, one for each reference. Don't put anything between the citations. The labels for all the citations will be attached to the line preceding the first citation. The labels may also be sorted or merged. See the description of the <> label expression, and of the sort-adjacent-labels and abbreviate-label-ranges commands. A label will not be merged if its citation has a non-empty opening-text or closing-text. However, the labels for a citation using the ] flag and without any closing-text immediately followed by a citation using the [ flag and without any opening-text may be sorted and merged even though the first citation's opening-text or the second citation's closing-text is non-empty. (If you wish to prevent this, use the dummy character escape sequence \& as the first citation's closing-text.)


Commands are contained between lines starting with .R1 and .R2. Recognition of these lines can be prevented by the -R option. When a .R1 line is recognized any accumulated references are flushed out. 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 the first names of fields. An initial letter will be separated from another initial letter by string1, from the last name 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.
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 references instead of writing out each reference as it is encountered. Accumulated references will be written out whenever a reference of the form
is encountered, after all input files have been processed, and whenever a .R1 line is recognized.
field is an annotation; print it at the end of the reference as a paragraph preceded by the line

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.

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.
Write out all the references contained in each bibliographic database file. This command should come last in an .R1/.R2 block.
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 \*([. \*(.] ", "
Convert fields to caps and small caps.
Recognize .R1 and .R2 even when followed by a character other than space or newline.
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.
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.
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.
When the reference is read, fields should be discarded; no string definitions for fields will be output. Initially, fields are XYZ.
Control use of et al. in the evaluation of @ expressions in label expressions. If the number of authors needed to make the author sequence unambiguous is u 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 and interpret the contents as commands.
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.
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.
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.
string is a label expression describing how to label each reference.
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.
In the text, move any punctuation at the end of line past the label. It is usually a good idea to give this command unless you are using superscripted numbers as labels.
Reverse the fields whose names are in string. Each field name can be followed by a number which says how many such fields should be reversed. If no number is given for a field, all such fields will be reversed.
While searching for keys in databases for which no index exists, ignore the contents of fields. Initially, fields XYZ are ignored.
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.
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 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 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 will have 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
The n-th part of field. If n is omitted, it defaults to 1.
The characters in string literally.
All the authors joined as specified by the join-authors command. The whole of each author's name will be used. However, if the references are sorted by author (that is, the sort specification starts with “A+”), then authors' last names 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. The use of only the last name for the i-th author of some reference is considered to be ambiguous if there is some other reference, such that the first i-1 authors of the references are the same, the i-th authors are not the same, but the i-th authors last names 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 purpose will have the same representation.
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.
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.
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 towards the total.
expr converted to lowercase.
expr converted to uppercase.
expr converted to caps and small caps.
expr reversed so that the last name is first.
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.
The year part of expr.
The part of expr before the year, or the whole of expr if it does not contain a year.
The part of expr after the year, or an empty string if expr does not contain a year.
The last name part of expr.
expr1 except that if the last character of expr1 is - then it will be replaced by expr2.
expr1 expr2
The concatenation of expr1 and expr2.
If expr1 is non-empty then expr1 otherwise expr2.
If expr1 is non-empty then expr2 otherwise an empty string.
If expr1 is non-empty then expr2 otherwise expr3.
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.
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.


--help displays a usage message, while -v and --version show version information; all exit afterward.

Don't recognize lines beginning with .R1/.R2.

Other options are equivalent to refer commands.

reverse An
no-label-in-text; no-label-in-reference
See below.
capitalize fields
label %n
search-ignore fields
label L~%a
-k field
label field~%a
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
database db-file
sort spec
label "(A.n|Q) ', ' (D.y|D)"; bracket-label " (" ) "; "
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.

annotate X AP; no-label-in-reference
-B field.macro
annotate field macro; no-label-in-reference


If set, overrides the default database.


Default database.
Index files.
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.


In label expressions, <> expressions are ignored inside .char expressions.


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
on your summer vacation.
.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 original input. As discussed in subsection “Macro interface” above, it is up to the document or a macro package to employ and format this information usefully. Let us see how we might turn groff_ms(7) to this task.

REFER=my-db-file groff -R -ms
Read the book
on your summer vacation.
Commentary is available.\*{*\*}
.FS \*{*\*}
Space reserved for penetrating insight.

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

“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)

4 October 2023 groff 1.23.0