path: root/info/ltx3pub/l3d002e.tex

%%% ====================================================================
%%%  @LaTeX3-article{ LaTeX3-LTX3-002e,
%%%  filename        = "l3d002e.tex",
%%%  archived        = "ctan:/tex-archive/info/ltx3pub/",
%%%  related-files   = "part of l3d002.tex",
%%%  author          = "David Rhead",
%%%  doc-group       = "Project core team",
%%%  title           = "Some ideas for improving {\LaTeX}\\ General",
%%%  version         = "1.1",
%%%  date            = "18-Mar-1993",
%%%  time            = "20:19:36 GMT",
%%%  status          = "public, official",
%%%  abstract        = "Ideas and suggestions from David Rhead for
%%%                     improving various areas in LaTeX",
%%%  note            = "prepared for the workshop at Dedham 91",
%%%  keywords        = "",
%%%  project-address = "LaTeX3 Project            \\
%%%                     c/o Dr. Chris Rowley      \\
%%%                     The Open University       \\
%%%                     Parsifal College          \\
%%%                     Finchley Road             \\
%%%                     London NW3 7BG, England, UK",
%%%  project-tel     = "+44 171 794 0575",
%%%  project-FAX     = "+44 171 433 6196",
%%%  project-email   = "LTX3-Mgr@SHSU.edu",
%%%  copyright       = "Copyright (C) 1993 LaTeX3 Project
%%%                     All rights reserved.
%%%
%%%                     Permission is granted to make and distribute
%%%                     verbatim copies of this publication or of
%%%                     coherent parts from this publication provided
%%%                     this copyright notice and this permission
%%%                     notice are preserved on all copies.
%%%
%%%                     Permission is granted to copy and distribute
%%%                     translations of this publication or of
%%%                     individual items from this publication into
%%%                     another language provided that the translation
%%%                     is approved by the original copyright holders.
%%%
%%%                     No other permissions to copy or distribute this
%%%                     publication in any form are granted and in
%%%                     particular no permission to copy parts of it
%%%                     in such a way as to materially change its
%%%                     meaning.",
%%%  generalinfo     = "To subscribe to the LaTeX3 discussion list:
%%%
%%%                      Send mail to listserv@vm.urz.uni-heidelberg.de
%%%                      with the following line as the body of the
%%%                      message (substituting your own name):
%%%
%%%                        subscribe LaTeX-L First-name Surname
%%%
%%%                     To find out about volunteer work:
%%%
%%%                      look at the document vol-task.tex which can
%%%                      be obtained electronically, see below.
%%%
%%%                     To retrieve project publications electronically:
%%%
%%%                      Project publications are available for
%%%                      retrieval by anonymous ftp from ctan hosts:
%%%                          ftp.tex.ac.uk
%%%                          ftp.dante.de
%%%                          ftp.shsu.edu
%%%                      in the directory /tex-archive/info/ltx3pub.
%%%
%%%                      The file ltx3pub.bib in that directory gives
%%%                      full bibliographical information including
%%%                      abstracts in BibTeX format.  A brief history
%%%                      of the project and a description of its aims
%%%                      is contained in l3d001.tex.
%%%
%%%                     If you only have access to email, and not ftp
%%%                      You may use the ftpmail service.
%%%                      Send a message just containg the word
%%%                          help
%%%                      to ftpmail@ftp.shsu.edu
%%%                      for more information about this service.
%%%
%%%                     For offers of financial contributions or
%%%                      contributions of computing equipment or
%%%                      software, contact the project at the above
%%%                      address, or the TeX Users Group.
%%%
%%%                     For offers of technical assistance, contact the
%%%                      project at the above address.
%%%
%%%                     For technical enquiries and suggestions, send
%%%                      e-mail to the latex-l list or contact the
%%%                      project at the above address.",
%%%  checksum        = "28553 1599 11400 79963",
%%%  docstring       = "The checksum field above contains a CRC-16
%%%                     checksum as the first value, followed by the
%%%                     equivalent of the standard UNIX wc (word
%%%                     count) utility output of lines, words, and
%%%                     characters.  This is produced by Robert
%%%                     Solovay's checksum utility.",
%%%  }
%%% ====================================================================

\chapter{Citations and reference-lists}
\label{citing}

\section{Some e-mail comments}

\begin{footnotesize}\begin{verbatim}
From:     Sebastian Rahtz <spqr@UK.AC.SOTON.ECS>
Date:     Tue, 7 Aug 90 15:33:13 bst

...  writes:
 > One of the most common mistakes that I see from LaTeX users is typing
 > \maketitle before \begin{document}; since this causes horrible things to
 > happen, perhaps we should make sure that this causes an error.
and why not put \bibliographystyle before \begin{document}, where it
belongs..
\end{verbatim}\end{footnotesize}
\begin{center} --- \end{center}
\begin{footnotesize}\begin{verbatim}
From      David Rhead ...
Date:     13 Aug 90 17:39:39

Sebastian suggested
     "and why not put \bibliographystyle before \begin{document}, where it
      belongs.."

This might need thinking through, bearing in mind that certain types of
document may have more than one list of references.  For example:
Manuals produced by software houses, e.g. the SPSS-X documentation.
      Such manuals may, in effect, divide citations into 2 categories:
      1.  references to other manuals produced by the software house,
          which might be cited using a "short title" scheme
      2.  references to other literature, which might be cited using
          an author-date scheme.
      E.g. the SPSS-X Introductory Statistics Guide generally uses
      author-date, but gives the full reference to the SPSS-X User Guide
      in its preface and thereafter refers to it as "SPSS-X Users Guide".
      The software house may list its own publications in the preface to
      its manual, putting the list of references to other literature
      at the back of the manual.  To support this sort of thing would
      require something like
            \documentstyle{manual}
            \begin{document}
            ...
            %  \xxx represents a command that embeds \bibliography-like
            %  information in a preface.  "software-house" represents
            %  a style that software houses seem to like where the
            %  reference is embedded in an explanatory paragraph.
            \xxx[Our other manuals]{software-house}{our-manuals}
            \chapter{...}
            ...
            \bibliography[Other people's stuff]{author-date}{other-literature}
      where \xxx and \bibliography are assumed to have
      *  an optional argument to specify a title
      *  an argument to specify the scheme (like \bibliographystyle does)
      *  an argument to specify the bib files.
Books.  It might be sensible to divide a book's references into 2 lists, e.g.
      "References" and "Further reading".  E.g. the draft revised British
      Standard for theses suggests having "Bibliography" as well as "List
      of references".  It might also be sensible to have different styles
      for the 2 lists:  perhaps a concise style for the "References",
      but a style that prints some extra information about the
      "Further reading".  The author might want to specify something like
            \documentstyle{...}
            \begin{document}
            ...
            \bibliography[References]{author-date}{specific-books}
            \bibliography[Further reading]{annotated}{general-books}
            ...
      [Such sub-division seems to be countenanced by the gurus, e.g. Chicago
      Manual of Style (p. 425) and Butcher's Copy-editing (p. 183,192).]
Unfortunately, such considerations lead to more questions like "How does
one arrange that \cite gives (...) around author-date citations, but not around
short-title citations?"  and "Can a root file have several bbl files and what
would they be called?", and I don't know the answers.

At the moment, I just think that it would be better to refrain from putting
\bibliographystyle before \begin{document} until the implications for
documents that have more than one list of references have been thought through.
Otherwise there might be a change to "\bibliographystyle before
\begin{document}" at LaTeX 3.0 followed by another change (e.g. to
"bibliography style as argument of \bibliography") in some subsequent LaTeX.
\end{verbatim}\end{footnotesize}
\begin{center} --- \end{center}
\begin{footnotesize}\begin{verbatim}
From:     Sebastian Rahtz <spqr@UK.AC.SOTON.ECS>
Date:     Tue, 14 Aug 90 10:00:33 bst

 >      "and why not put \bibliographystyle before \begin{document}, where it
 >       belongs.."
David Rhead has some sensible cautionary remarks, and I would concur
with his underlying thesis that the bibliography support in
LaTeX/BibTeX has definite lacunae. He identifies the problem

 > This might need thinking through, bearing in mind that certain types of
 > document may have more than one list of references.  For example:

and suggests a formulation along the lines of
 >             \bibliography[References]{author-date}{specific-books}
 >             \bibliography[Further reading]{annotated}{general-books}

this is interesting, but nobody will thank anyone for changing the
basic syntax of a LaTeX command in the near future. I would suggest
retaining \bibliography as it is documented, and implementing David's
suggestions with others as new commands. ... such as separate
bibliographies for chapters - whatever LL or OP say about it being so much
work to do a multi-chapter book that sets of bibliographies being not
much more work, I don't see why we shouldn't get it in one day. People
*do* want it. I am on my fourth conference proceedings in as many
years, and I dont enjoy building the slightly complicated Makefile to
get all the references in the right place up to date.

but I continue to say that the user who puts \bibliographystyle before
\begin{document} is behaving intuitively, and should therefore be
allowed to do it. lets keep `bibliography' for the relatively simple
academic type that LL and OP envisaged, and invent a new term
(\reference?) for what David Rhead is talking about.
\end{verbatim}\end{footnotesize}
\begin{center} --- \end{center}
\begin{footnotesize}\begin{verbatim}
From:     "Nelson H. F. Beebe" <Beebe@EDU.UTAH.SCIENCE>
Date:     Sat, 18 Aug 90 17:15:10 CET

I just hit what I view as a design flaw in thebibliography
environment.  It was precipitated by the following BibTeX
bibliography entry in the .bbl file:

\bibitem{Steele:floating-point-output}
Guy~L. {Steele Jr.} and Jon~L. White.
\newblock How to print floating-point numbers accurately.
\newblock {\em ACM SIG\-PLAN Notices}, 25(6):112--126, June 1990.
\newblock In electronic mail dated Wed, 27 Jun 90 11:55:36 EDT, Guy
Steele reported that an intrepid pre-SIGPLAN 90 conference
implementation of what is stated in the paper revealed 3 mistakes:

  \begin{enumerate}
    \item
         Table~9 (page 125):\par\noindent
        \begin{tabular}{ll}
          for & {\tt -1:USER!({"}{"});} \\
          substitute & {\tt -1:USER!({"}0{"});}
        \end{tabular} \par\noindent
       and delete the comment.
    \item
          Table~10 (page 125):\par\noindent
          \begin{tabular}{ll}
            for & {\tt fill(-k, {"}0{"})}\\
            substitute & {\tt fill(-k-1, {"}0{"})}
          \end{tabular}
    \item
          Table~5 (page 124):\par\noindent
          insert {\tt k <-- 0} after assertion, and also delete {\tt k
          <-- 0} from Table~6.
  \end{enumerate}

The effect of this is that the next bibliography entry gets
number 4, one more than the last enumerate counter.  The reason
lies in latex.tex:

%  The thebibliography environment is a list environment.  To save the
%  use of an extra counter, it should use  enumiv  as the item counter.

For now, I will switch to an itemize instead of enumerate.

The question for this list is, is the saving of an extra counter
at what is usually almost the end of a document anyway worth this
design gotcha?  I suggest not.
\end{verbatim}\end{footnotesize}
\begin{center} --- \end{center}
\begin{footnotesize}\begin{verbatim}
From:     MITTELBACH FRANK <PZF5HZ@EARN.DRUEDS2>
Date:     Fri, 5 Oct 90 17:43:46 CET

I would like to foward a mail from Nico which was a reaction
to some discussions at the Cork meeting.

------------------------ forwarded mail -------------
>
> here are some first thoughts on BibTeX and front matter.
>
> BibTeX
> ------
> BibTeX is unfortunately not a database management system --
> unfortunate in the sense that users have to think of a way to
> *really* manage the database, i.e. add, change, remove, sort, select,
> import and export entries. On my Atari I use a simple database
> program -- a real GEM program with mouse and buttons -- to maintain a
> literature database. I've written a conversion program that generates
> a .bib file from this database. This system works and is also used ny
> my ex-colleagues at the university.
>
> In general, working with BibTeX is cumbersome -- every time you add a
> reference to your document you have to run LaTeX, BibTeX, and LaTeX
> twice again -- and a lot of LaTeX users think it only pays off when
> your document/bibliography is above a certain length.
>
> What I would like to see in LaTeX is coding of the logical structure
> of every \bibitem. In version 2.09, BibTeX takes care of the logical
> structure and outputs formatted text. In other words: a part of the
> document, the one included by the \bibliography command, does not
> contain tags for the logical structure. Furthermore, if a user
> decides NOT to use BibTeX, he/she has to do the formatting completely
> by hand.
>
> Front matter
> ------------
> Currently, an article starts with \title, \author and \date
> instructions. In the Elsevier styles I have added \address,
> \received, \revised and \accepted commands, and also a keyword
> environment, similar to the abstract environment. By doing so, we can
> automatically convert a LaTeX-coded document to an SGML-coded
> document.
> I think it is a bad idea to put the front matter information in a
> .bib record for the above reasons and also for the following reason:
> some parts of the front matter do not have to be re-used again in
> bibliographical entries, and therefore do not belong in the .bib
> database.
> Examples: list pf previous books in a series, LCC data, dedication,
> motto (books), keywords, date of receipt, revision, acceptance
> (articles).
>
...

The last comment about the bad idea probably needs some
explanation: During the Cork conference we discussed the possibility
to use an extended version of bibtex, which is able to produce
several bibliogrphies at once, to format a title page (on request).

The idea was that for larger documents one could set up a bibliography
entry for the document itself with all relevant information in it.
In the document itself the title would then be formated by a command,
say \titlefrombib{<key>} which would make request bibtex to produce
a file containing all necessary information for producing the title.
This means that every journal that accepts LaTeX input would supply
a BibTeX style file which generates from a single bib entry such
a file. The advantages of this method would be that unused information
would produce no problem (the bst file would simply ignore them).
Another advantage would be that authors (for eample for TUGboat) would
already sending a bib entry along with his document.

Of course, a standard way for simple document that generates titles
without using BibTeX should be availabe too. And the standard document
styles should also have their bst files for producing the title.

This scheme if, of course only sensible (if at all) if \BibTeX is
able to produce several bbl files in one pass.

I still think that such a feature has some promising possibilites
and would like to hear other opinions about it.
\end{verbatim}\end{footnotesize}
\begin{center} --- \end{center}
\begin{footnotesize}\begin{verbatim}
From:     David Rhead
Date:     9 Oct 90 10:39:45

Frank forwarded some mail from Nico about BibTeX etc.


BIBLIOGRAPHIC DATABASE?

I agree with Nico's comment about BibTeX not being a database management
system.  I have the impression that some research workers really want a
system that will not only contain author, date, etc. for a paper/book,
but also a copy of the abstract and perhaps some reminders to themselves.
Thus, they want something that will both
(1) help (by supplying bibliographic details) when they are in the process
    of writing a paper
and
(2) help them to search through all the papers they've ever read
    until they come to something that is about a particular subject.
    E.g. "give me all the papers about X", "what was that paper about Y?".

It's unfortunate that Lamport/Patashnik used the term "bibliographic
database" (although I can't think of anything better offhand).  It leads
to people thinking that BibTeX will do the things that they associate
with "database systems" these days.

However, I think that to turn BibTeX into something that did more
than (1) would be too ambitious.  We're going to have enough problems
finding someone to make BibTeX do (1) better.

I think it would be better to treat (2) as a separate project, and to ask
"is (2) best done by adding database management features to BibTeX,
or would it be better done by adding BibTeX-like features to a
database management system?"  I don't know what the answer is.

For the time being, I think users will have to be left in an unsatisfactory
situation.  If the "raw data" is kept in a "real database", the .bib file
is just yet another intermediate file.  I seem to remember that
Sebastian Rahtz has set INGRES up so that INGRES can hold bibliographic
information and write it out in the form of a .bib file.  It sounds as
though Nico is doing something similar with a different database
system.

It seems sensible to use a "real database system" for what "real database
systems" are good at.  Although it is clearly unsatisfactory to have a .bib
file that merely "shadows" a "real database" I can't think of anything better
that could be done quickly.  [Perhaps a database expert might have some bright
ideas.  Could such an expert write software that took an .aux file, generated
instructions in a "query language" to select the \cite-ed references, and
then produced a .bbl file (or equivalent) without there ever being a .bib file?
I don't know:  I'm not a database expert.  Even if they could, it
would probably still be useful to have a standalone program like BibTeX
that did task (1) in a database-system-independent way.]

A salesman has sent me a leaflet about a piece of software called
EndNote which apparently seems to aim to do both (1) and (2) for Mac and PC
word-processor users.  (It's not public domain, but then neither is INGRES.)
I see that it can export information for troff's "refer".  He's coming to
see me at the beginning of November, so I'll ask if EndNote might be tailored
to read .aux files and produce .bbl files.  I expect the answer will be
"no" or "only after a lot of work", but it does no harm to ask!
Perhaps one could have a public-domain BibTeX for task (1) with tailored
proprietary software for people who want (1)+(2).

Conclusion:  I can't think of an easy way of improving the "3 times
through LaTeX and once through BibTeX" business at the moment.  I doubt
whether its worth making BibTeX into a proper database system, although
it might be worth making "a proper database system" do what BibTeX does.


CODING THE LOGICAL STRUCTURE OF EACH \BIBITEM?

I can see the attraction of this.  The list of references could be
held in the root file or \input or \include-ed from a .tex file
like everything else.  There wouldn't need to be any special treatment
of .bbl files.

However, in terms of project-management, it seems very convenient to regard
determination of the logical structure for bibliographic references
as a separate task which can be delegated to whoever volunteers
(if we can find them) to do another iteration on BibTeX.
It would be up to this person to specify the logical elements
for a reference, e.g. to agonise about whether ADDRESS is
the fundamental concept of whether it should be PLACEOFPUBLICATION.

All that needs agreeing between the LaTeX 3.0 people and the
BibTeX worker (if we find one) is the form of interface, i.e.
what BibTeX passes back to (a) go where the \cite was and (b) go
in the list of references.  There might need to be separate
interfaces for "reference by number", author-date and "short title"
[I'll have a go at suggesting what these might be in a future message]
but beyond that, the LaTeX 3.0 people need not be concerned
about the distinction between e.g. different author-date styles.
Thus, the BibTeX worker's considerations of matters such as
ADDRESS versus PLACEOFPUBLICATION can proceed in parallel with
the LaTeX 3.0 people's work on other matters.  It wouldn't matter
much whether both projects were complete at the same time.

If the LaTeX 3.0 people attempted to code the logical structure of
each \bibitem, this would mean that LaTeX 3.0 could not be finished
until the LaTeX 3.0 people had satisfied themselves about "What are
the fundamental types of publication (\bibitem{knuth-84}{book} or
\bibitem{knuth-84}{monograph}?), and what are the fundamental
items of bibliographic information about them?  Did Lamport/Patashnik
get it right?  E.g. ADDRESS versus PLACEOFPUBLICATION."  The LaTeX 3.0
people would presumably have to provide TeX code that sorted bibliographic
details out into the order required for a particular style e.g. "reference
by number in ACM style" (and perhaps provide a few style-options to
show how the details could be changed for a different convention,
e.g. author-date in APA style).  I think that the LaTeX 3.0 work
is ambitious enough without taking this analysis on too.  Attempting
to get the subdivision of \bibitem right could hold the rest of the
project up.

Continuing to delegate the work on the logical structure of each \bibitem
to BibTeX might not be as elegant from the user's point-of-view as
getting LaTeX to do all the work (using structure information from
subdivisions of \bibitem and bibliography-style information from
\documentstyle) but I think the result would be available sooner and that
the user might prefer to have something better than LaTeX 2.09 soon, rather
than to have perfection not so soon.

The two approaches (1) "put all the BibTeX work into LaTeX" (to take
account of the logical structure of each \bibitem), and (2) "make
BibTeX into a proper database management system" seem to be pulling
in different directions.  I don't think one can do both (otherwise
you'd end up with LaTeX as a bibliographic database management system),
although one could do neither.

For LaTeX 3.0, I'd be inclined to leave the contents of each \bibitem
(or the successor to \bibitem) as a "black box", to be filled in by the
user or by BibTeX.  If someone does the analysis for BibTeX 2.0 (say),
the question could be considered again if there is ever a LaTeX 4.0 (!)

[This all assumes that the SGML-ers have not analysed the structure
of a list of references and hence that someone has to do the analysis.
If the SGML-ers have done the anlysis (for a DTD, perhaps), could they
publish it?]

Conclusion:  I'd specify an interface between LaTeX 3.0 and BibTeX
that would support the "reference by number", author-date and (if
possible) "short title" schemes, but delegate the task of supplying
\bibitems (or whatever) to that specification to whoever updates BibTeX
and its .bst files.


FRONT MATTER INFORMATION IN A .BIB RECORD?

The gurus of "how to do a list of references" seem to agree that
bibliographic details should be as they appear on the title page of
the article, book, etc.  But there are many caveats:
*  Several books by the same author in one bibliography should
   follow the same style (Chicago Manual of Style, p. 441).
*  There are potential problems with names like Tchaikovsky,
   which may appear in different forms (Chaikovsky) on different
   title pages, even though the works are all by the same person.
   [British Standard BS 1629, p. 5]
*  The part of the name not on the title page may be enclosed
   in square brackets (Chicago, p. 441).
*  If the name on the title page is a pseudonym, the author's
   real name may be given in the bibliography in square brackets.
   (Chicago, p. 442).
*  Capitalization, punctuation, etc. of a title may be differ in a
   bibliography from the conventions on the title page (Chicago, p. 447).
   Similarly, compulsory line-breaks may be wanted on the title-page
   but not in the list of references (\\, Lamport's book, p. 164).
   For another example, consider "LaTeX: A Document Preparation System":
   that's not how it appears on the title page.
*  It may be necessary to use discretion about whether to regard
   a subtitle as part of a title or to abbreviate a long title
   (ISO standard 690, p. 5).
*  A bibliography may give "place of publication" in a form that is
   different to that on the title page, using discretion about:
   - whether to list all places where the publisher has offices
     or just one place
   - whether to give further information (if the place of publication
     is not widely known or could be ambiguous).
   (Chicago, p. 456).
*  A bibliography might give a publisher's name in a form that differs
   slightly from that shown on the title page (Chicago, p. 458).

I think that having a BibTeX that can produce several bibliographies at
once would be "a good thing".  For example:
*  for conference proceedings where each contribution may have its own
   list of references
*  for books that may have e.g. "References" and "Further reading"
*  things like the SPSS manual (and other things produced by software
   houses), which seem to give the software-house's related publications
   in a preface, but put "academic references" at the end.

However, I don't necessarily think that the same mechanism should
be used to "derive a publication's title-page from its .bib entry".
Traditionally, bibliography entries have been derived from title-pages
(with some human discretion), rather than the other way round, so its
probably safer to have software that imitates the tradition.  One might
think of having things like
\begin{titlepage}
   \author[bibliography-version]{titlepage-version}
   \title[bibliography-version]{titlepage-version}
   \place[bibliography-version]{titlepage-version}
   \publisher[bibliography-version]{titlepage-version}
\end{titlepage}
\begin{copyrightpage}
   \copyrightholder{...}
   \isbn{...}
\end{copyrightpage}
in the .tex file (where the optional arguments allow humans to exercise
the discretion recommended by the gurus) and having LaTeX produce
perhaps a .bibitem file that the user can append to a suitable .bib file
(maybe after exercising a bit more discretion).  LaTeX could put
information that might conceivably be used by a bibliographer in the
.bibitem file, but refrain from putting information that no bibliographer
would ever want to that file.  TUGboat would get a .bib entry
with each article (but it would be derived automatically from the
article-heading, rather than used to automatically produce the
article-heading).

[Conversely, if the Cork suggestion was adopted, and title-page
information was produced by some future BibTeX from a .bib file, there
would have to be some mechanism to allow for minor variations, e.g.
      TITLE = "LaTeX: A Document Preparation System",
      TITLEPAGETITLE = "LaTeX\\A Document Preparation System"]

The problems that I mentioned in the context of "logical structure
of \bibitem" arise here too.  To write .sty files for (say) book, report,
conference-proceedings and article, you only need to be clear about those
categories (as well as being a TeX wizard and having a lot of time,
perseverance and patience).  To write .bst files with entry-types of book,
report, conference-proceedings, article, you need to be clear whether
they are distinct entry-types or not.  [ISO 690 could be interpreted
as lumping books, reports and conference-proceedings all together
as "monographs".]   You also need to be clear about the fields:
e.g. PLACE or ADDRESS.  So the LaTeX 3.0 code for title-pages would
get held up (and hence LaTeX 3.0 as a whole would get held up)
while someone analysed "the structure of a \bibitem".
On the other hand, if LaTeX 3.0 wrote out a .bibitem file that wasn't
quite what some new BibTeX expected, it wouldn't matter very much,
and could be corrected once it was clear what was required.

Conclusion:  I'd like BibTeX to support multiple lists-of-references,
but think that "LaTeX producing .bib info from titlepage info"
might be better than "BibTeX producing titlepage info from .bib info".
\end{verbatim}\end{footnotesize}
\begin{center} --- \end{center}
\section{My e-mailed \lq\lq paper''}

\subsection{Introduction}

Since there are now plans for a new version of \LaTeX\
\cite{lamport-86,m+s-89}, this may be a good time to consider
how a future version should deal with citations and reference-lists.

My viewpoint is that of an advisor to authors who use \LaTeX\ 2.09.
As such, I'm often in the position of having to decide whether \LaTeX\
acts inappropriately or whether an  author is asking for something
inappropriate.  Generally, I have the impression that \LaTeX\ 2.09 sometimes
makes it unnecessarily difficult for people to comply with the conventions that
are standard in \lq\lq academic publishing''.

In making the suggestions that follow, I do realise that it's easier for me
to suggest that the volunteers developing \LaTeX\ 3.0 might provide some new
facilities than it is for a volunteer to find time to do the work!

I assume that the reader is familiar with the relevant sections of
\cite{lamport-86}.

\subsection{Conventions to be supported}
\label{conventions}

\subsubsection{Citation schemes}
\label{basic-schemes}

In mainstream publishing
\cite{bs-5605,bs-6371,butcher-81,chicago-82,gibaldi,huth,inter,iso-690,%
oconnor,page},
there seem to be three basic schemes for citations and the corresponding
reference-lists:
\begin{description}
\item[reference by number]  In this scheme, citations are normally numbered in
      order of first citation.  In particular, \lq\lq order of first citation''
      is used by over 300 biomedical journals \cite{huth,inter}, and is
      specified in the ISO standard \cite{iso-690}.  (Thus, in \BibTeX\ terms,
      the usual numbering sequence is that given by {\tt unsrt}.)  The number
      is used as a \lq\lq label'' in the reference-list.
\item[author-date] There are two main forms of citation, depending on whether
      or not the cited author's name occurs naturally in a sentence.  In the
      first case, the citation is of the form \lq\lq\ \dots\ (1972)
      \dots\ '', whereas in the second case it is of the form \lq\lq\ \dots\
      (Crane, 1972) \dots\ ''.  There are no \lq\lq labels'' in the
      reference-list, which is arranged in alphabetical order of authors'
      surnames (with supplementary rules for \lq\lq tie-breaking'').
\item[short form] The \lq\lq short form'' scheme is often used when citations
      occur in footnotes.  Typically, the first citation (or the first
      citation in the current chapter) gives a fairly full reference but
      subsequent citations use a short form.  The \lq\lq short form'' may be
      introduced within the first citation, or given in a table of
      abbreviations.  There are no \lq\lq labels'' in the reference-list,
      which may be subdivided by \lq\lq type of cited document''.
      The scheme is common in the humanities, but also seems to be used by
      some software-houses when referring to their own publications (see, for
      example, \cite{norusis}).
\end{description}
I think that, in order to make it straightforward to achieve, with \LaTeX,
the effects that people routinely achieve with traditional publishing
procedures, it is desirable that \LaTeX\ should provide explicit support for
all three citation schemes.

Notice that:
\begin{itemize}
\item It is not, in general, possible to convert a document from one scheme
      to another (e.g., from \lq\lq reference by number'' to author-date)
      automatically.  Some re-writing is required.
\item The number of items of information that need to be available
      differ between the schemes.  For a \lq\lq reference by number''
      citation, it is only necessary to keep track of one item
      (the number), whereas for the other schemes it is necessary to keep
      track of more than one item (i.e., the author and the date, or the
      \lq\lq fairly full form'' and the \lq\lq short form'')
      so that they can be used separately.
\item Occasionally, two different schemes may be used in parallel
      within the same document. (For example, in \cite{norusis},
      a software-house seems to use \lq\lq short form'' when citing
      its own publications, but author-date when citing other publications.)
\end{itemize}
Thus, it seems to me that the three schemes are best regarded as logically
distinct.

However, within a particular scheme, there are variations of punctuation, etc.,
that can be regarded as matters of \lq\lq house style''.
For example, some journals that have adopted a \lq\lq reference by number''
scheme use bracketed numerals for citations, while others use superscripts.
Such variations can be accomodated by differences between style files.

\subsubsection{Additional references}
\label{support-additional}

In addition to a list of \lq\lq works cited'', some documents
have an additional reference-list that specifies \lq\lq further reading''
or \lq\lq all sources consulted'' (see \cite[pp.\ 182 \& 192]{butcher-81},
\cite[p.\ 40]{huth} and \cite[p.\ 22]{bs-4821}).  There needs to be some
provision for typesetting such additional lists.


\subsection{Deficiencies in \LaTeX\ 2.09}
\label{2.09-problems}

\subsubsection{Citations}
\label{2.09-citations}

When used in conjunction with \BibTeX\ and \verb+\bibliographystyle{unsrt}+,
\LaTeX\ 2.09 makes a good job of organising citations (and sorting the
corresponding reference-list) according to the \lq\lq reference by number''
scheme as required by many journals.

Generally, however, \LaTeX\ 2.09 does not provide the breadth or depth
of facilities needed to support the variety of requirements for
mainstream publishing:
\begin{itemize}
\item It is unfortunate that the \lq\lq reference by number'' sequence obtained
      most naturally by the do-it-yourself-er \cite[p.\ 73]{lamport-86} from
      \LaTeX\ 2.09 (\lq\lq order of appearance within {\tt thebibliography}'')
      is unlikely to be what the do-it-yourself-er's journal-editor requires
      (which will usually be \lq\lq order of first citation'').  The
      do-it-yourself-er is given no warning (either in \cite{lamport-86} or by
      the software) that sorting is likely to be needed.
\item Although one can use style-options such as {\tt apalike} and
      {\tt aaai} to re-define \verb+\cite+ and {\tt thebibliography}
      for an author-date scheme, there are many obstacles placed in the
      user's way:
      \begin{itemize}
      \item The existence of the style-options is not documented in
            \cite{lamport-86}.
      \item If one finds a style-option  in a (software) archive, it
            may need modification to produce the precise effect required.
      \item It is not obvious how one should refer separately to two
            items (author and date) supplied via a \verb+\bibitem+ argument
            originally designed for one.  The do-it-yourself-er might
            have to study style files such as {\tt aaai.sty} and {\tt bbl}
            files such as those produced by {\tt aaai-named.bst} to
            deduce how to do this.
      \end{itemize}
\item The \lq\lq short form'' scheme seems unsupported.
\item At certain points in a document, an author my need to cite
      several works at once.  It may be necessary to specify a
      page (or section, etc\@.) for each work.  For instance,
      \cite[p.\ 404]{chicago-82} suggests references of the form
      \lq\lq (Kelley 1896a, 10; Kelley 1896b; Kelley 1907, 3)''.
      This is not easy in \LaTeX\ 2.09, since \verb+\cite+'s
      optional argument applies to the citation as a whole.  The
      author cannot supply a separate \lq\lq optional argument''
      for each work.  (By contrast, {\sf EndNote}'s analogous facilities
      \cite[p.\ 58]{endnote} seem to allow each individual work
      to be given its own \lq\lq additional text''.)
\item It does not seem easy to use different schemes in parallel
      within the same document.  If \verb+\cite+ and
      {\tt thebibliography} are defined as required for one scheme,
      they will usually be unsuitable for any other.
\end{itemize}

\subsubsection{Additional references}
\label{2.09-additional}

As stated in section \ref{support-additional} an author may need to
typeset a list of \lq\lq further reading'' etc., in addition to
the usual list of \lq\lq works cited''.

If using the {\tt thebibliography} environment from one of \LaTeX\
2.09's standard styles for such additional references, an author will be
faced with the following problems:
\begin{itemize}
\item the title will be the same as that for the list of \lq\lq works
      cited'', namely \lq\lq References'' for {\tt article} and
      \lq\lq Bibliography'' for {\tt report} and {\tt book}
\item the \lq\lq labels'' (which may be appropriate in the list
      of \lq\lq works cited'', particularly for the \lq\lq reference
      by number'' scheme) will also appear in the
      additional list (where they are inappropriate), because
      both lists use the same definition of \verb+\bibitem+
\item by default, the \lq\lq labels'' will not be unique, since
      the \lq\lq works cited'' list and the \lq\lq additional references''
      list will both be numbered from one
\item it will be necessary to supply dummy {\it cite-key\/}s,
      purely to satisfy the syntax required for a \verb+\bibitem+.
\end{itemize}

\subsubsection{Other problems}
\label{2.09-other}

The definitions of the {\tt thebibliography} environment in \LaTeX\
2.09's standard styles:
\begin{itemize}
\item issue either a \verb+\section*+ or a \verb+\chapter*+ command,
      using a {\it heading} of \lq\lq References'' for {\tt article}
      and \lq\lq Bibliography'' for {\tt report} and {\tt book}
\item set {\it left-head\/} and {\it right-head\/} to either
      \lq\lq REFERENCES'' or \lq\lq BIBLIOGRAPHY''
\item do not arrange for a table-of-contents entry.
\end{itemize}
These definitions can cause problems when the {\it heading}, etc.\
supplied by the standard style is inappropriate, or when a table-of-contents
entry is desired.
Admittedly, anyone who doesn't like the standard styles is free to
take copies of the style files and modify them to suit their requirements.
However, I have the impression that:
\begin{itemize}
\item among people who are competent to modify style files,
      modification of these aspects of the standard styles is \lq\lq the
      rule'' rather than \lq\lq the exception''
\item those \LaTeX\ users who aren't particularly computer-literate
      find the whole business mysterious, and seek out support staff
      who have to modify these aspects of the style files for them.
\end{itemize}
Overall, this seems to be an area where \LaTeX\ 2.09 is failing to
\lq\lq free people from formatting concerns to allow them to
concentrate on writing'' \cite[p.\ 8]{lamport-86}.

An associated problem is that modified style files may no longer be
compatible with standard utilities such as {\tt lablst}.

\subsection{Introduction of new facilities}
\label{new-facils}

It has been decided \cite{m+s-89} that \LaTeX\ 3.0 will be compatible with
\LaTeX\ 2.09 input files.  Thus, in particular, \LaTeX\ 3.0 must define
\verb+\cite+ and \verb+thebibliography+ so that they have the same effect on
\LaTeX\ 2.09 input files as the \LaTeX\ 2.09 definitions do.  This implies that
it would be difficult for \LaTeX\ 3.0 to (for example) define \verb+\cite+
so that there can be an optional argument for each work in a multiple citation
and define \verb+\bibitem+ so that it can have an \lq\lq author'' argument and
a \lq\lq date'' argument.

It therefore seems best to provide duplicates of the \LaTeX\ 2.09 facilities
in \LaTeX\ 3.0 (for \lq\lq backwards compatibility'') but to attempt to
provide new commands/environments in parallel so as to provide the required
functionality.  The new facilities would be regarded as the \lq\lq normal''
facilities, would be described in the body of the successor to \cite{lamport-86}, and
would be the natural choice for new users.  The old facilities would be
regarded as \lq\lq deprecated'' and relegated to an appendix of the successor
to \cite{lamport-86}.

Thus we can have both backwards compatibility and improved facilities
for the future.

\subsection{Division of labour}
\label{div-of-labour}

\subsubsection{Details needed for document \lq\lq as a whole''}

The three basic citation schemes mentioned in section \ref{basic-schemes}
determine certain details of a document \lq\lq as a whole''.  For each
citation, there must be an entry in a reference-list.  Each entry in
the relevant reference-list must have associated information that can be
used in citations.

\subsubsection{Details needed for reference-list, etc.}

There are a lot of other details that need to be resolved.

The information within each reference-list entry will probably need
formatting according to certain rules of \lq\lq house style''.
The information given in citations needs organising in a consistent way
(particularly for the \lq\lq short form'' scheme).

Different people may want to assemble their reference-lists in different ways.
Some people may wish to \lq\lq do it themselves'' \cite[p.\ 73]{lamport-86} from
a physical card-index, while some may prefer to use \BibTeX\ to get details
from a {\tt bib} file.  In some disciplines, proprietary systems such as
{\tt EndNote} \cite{endnote} seem popular (because they help the user to search
a database for literature to cite, as well as helping the user
incorporate details of the literature into a document).
Researchers may also wish to incorporate material obtained by searching
details held on a {\sc cd-rom}.

A reference-list generally needs sorting into a particular order.  Since the
list may occupy several pages, I assume that any sorting is best done outside
\LaTeX, either by other software (e.g., \BibTeX) or manually by the author.

\subsubsection{\LaTeX\ and other software}

It seems best to regard the movement of text-strings (e.g., an author's
surname) within the \lq\lq document as a whole'' as a task that is distinct
from the arrangement of details within the text-strings, and to assume a
\lq\lq division of labour'' in which the former task is performed by \LaTeX\
while the latter is performed by some other software or manually by the author.
The \lq\lq division of labour'' between \LaTeX\ 2.09 and \BibTeX\ seems to set
a good precedent.

This division of labour will lead to modular software.  Once the
interface between a reference-list and the rest of the document has
been defined, people can use \LaTeX\ for the body of their document,
but can:
\begin{itemize}
\item experiment with different software (\BibTeX, {\sf EndNote})
      for formatting the details of their reference-lists
\item enhance the other software (e.g., \BibTeX) independently
      of enhancements to \LaTeX.
\item lay their reference-lists out manually if they prefer.
\end{itemize}

\subsection{\protect\LaTeX\ 3.0: A possible user interface?}
\label{what-to-do}

\subsubsection{Specifications and names}

If the reasoning given in sections \ref{conventions}, \ref{new-facils}
and \ref{div-of-labour} is accepted, consideration needs to be given to
the form that new commands/environments should take in order to support the
three basic citation schemes, and to provide facilities for \lq\lq additional
references''.  In particular, it will be necessary to choose
names other than \verb+\cite+, {\tt thebibliography} and \verb+\bibitem+ (since
these names will be kept for the facilities provided for compatibility with
\LaTeX\ 2.09).


\subsubsection{Four sets of commands/environments}
\label{4-sets}

It seems to me that it would probably be convenient to have three sets of
commands/environments for dealing with citations and the corresponding
reference-lists, each set specifically designed to implement a particular
citation scheme.  Having three such sets gives scope for taking proper
account of the peculiarities of each scheme, without having one scheme
adversely affected by the peculiarities of another.

To avoid the difficulties mentioned in section \ref{2.09-additional},
it might also be worth having a specific environment for
\lq\lq additional references''.

\LaTeX\ 3.0 might, for example, have commands/environments as specified
in the following table.

\begin{center}
\begin{footnotesize}
\renewcommand{\arraystretch}{1.25}
\begin{tabular}{lccccccc}
\hline
                    &  Citation        & Environment for & Entry in           \\
                    &                  & reference-list  & reference-list     \\
\hline
Reference by number &  \verb+\numcite+ & {\tt numrefs}   & \verb+\numentry+    \\
\hline
Author-date         &  \verb+\dcite+
                                       & {\tt adrefs}    & \verb+\adentry+    \\
                    &  \verb+\adcite+ &                 &                 \\
\hline
Short form          &  \verb+\firstcite+  & {\tt sfrefs}    & \verb+\sfentry+     \\
                    &  \verb+\sfcite+&              &                     \\
\hline
Additional references&    ---         &  {\tt morerefs} & \verb+\moreentry+ \\
\hline
{\it Analogue at 2.09} &  \verb+\cite+ & {\tt thebibliography}
                                                          & \verb+\bibitem+   \\
\hline
\end{tabular}
\end{footnotesize}
\end{center}

Here it is assumed that:
\begin{itemize}
\item \verb+\numcite+ and \verb+\numentry+ have {\it key-list} and
      {\it cite-key} (respectively) as their only mandatory arguments.
\item \verb+\dcite+ and \verb+\adcite+ have {\it key-list} as argument.
      \verb+\dcite+ gives a citation of the form (1972),
      while \verb+\adcite+ gives a citation of the form (Crane, 1972).
      \verb+\adentry+ has three arguments: the {\it cite-key},
      the author (e.g., Crane), and the date (e.g., 1972).
\item \verb+\firstcite+ and \verb+\sfcite+ have {\it key-list} as argument.
      \verb+\firstcite+ gives the form of citation to be used when a work
      is first mentioned.  \verb+\sfcite+ gives the short form
      to be used in subsequent citations.
      The arguments of
      \verb+\sfentry+ might include: the {\it cite-key}, the form
      of reference to be used at the first citation, and the short form
      to be used subsequently.
      Whereas \verb+\numentry+ and \verb+\adentry+ can \lq\lq introduce''
      the full reference (like \verb+\item+ starts a new item
      \cite[p.\ 166]{lamport-86}), it may be better for \verb+\sfentry+
      to have the full reference as an argument, so that it can be used
      as the default \lq\lq form to be used at first citation''.
\end{itemize}

Although it would be desirable for the successors to the ``standard styles''
to define facilities for all three citation schemes, other
\verb+\documentstyle+s need not define facilities for all three.  For example,
a journal that wants its authors to use the author-date scheme would supply a
style file that only provides author-date facilities.

\subsubsection{Further details}
\label{further-details}

\paragraph{Reference by number}

The \verb+\numcite+ and \verb+\numentry+ commands might take the form
\verb+\numcite{+{\it key-list}\verb+}+ and
\verb+\numentry{+{\it cite-key}\verb+}+.  Notice that, since \verb+\numentry+
is specifically designed for \lq\lq reference by number'', there is no need to
allow an optional {\it label} argument like that for \verb+\bibitem+.

To conform to the ISO specification \cite{iso-690}, the successors to the
``standard styles'' would arrange for \verb+\numcite+ to give a citation of the form (24)
and for \verb+\numentry+ to give a reference-list entry of the form
\begin{description}
\item[{\rm 24.}] {\sc Crane, D.} {\it Invisible colleges.} \dots
\end{description}

Perhaps \LaTeX\ 3.0 could use the {\tt aux} file to refine an initial estimate
of the width of the final \verb+\numentry+'s \lq\lq label'', so that the
do-it-yourself-er wouldn't need to supply a {\it widest-label\/} argument.

\paragraph{Author-date}

The commands \verb+\dcite+, \verb+\adcite+ and \verb+\adentry+ might
be defined to have the forms \verb+\dcite{+{\it key-list}\verb+}+,
\verb+\adcite{+{\it key-list}\verb+}+ and
\verb+\adentry{+{\it cite-key}\verb+}{+{\it author}\verb+}{+{\it date}\verb+}+.
The {\tt adrefs} environment would {\it not} have a {\it widest-label}
argument, since in this scheme entries in the reference-list are unlabelled.

If such a definition of \verb+\adentry+ was documented in the successor to
\cite{lamport-86}, a do-it-yourself-er would be able to use the author-date
system just as easily as the \lq\lq reference by number'' system.

To conform to the ISO specification \cite{iso-690}, the successors to the
``standard styles'' would arrange for \verb+\dcite+ to give a citation of
the form (1972), for \verb+\adcite+ to give a citation of the form
(Crane, 1972), and for \verb+\adentry+ to give a reference-list entry with no
label.

\paragraph{Short form}

The \verb+\firstcite+, \verb+\sfcite+ and \verb+\sfentry+ commands might
be defined as \verb+\firstcite{+{\it key-list}\verb+}+,
\verb+\sfcite{+{\it key-list}\verb+}+ and
\verb+\sfentry{+{\it cite-key}\verb+}[+{\it
      fairly-full-form}\verb+]{+{\it short-form}\verb+}{+{\it
      full-reference}\verb+}+.
Such  definitions would, in effect, automate Butcher's manual method of
ensuring consistency \cite[p.\ 178]{butcher-81}.
Having {\it full-reference} as an argument means that the full reference
can be used as the default {\it fairly-full-form} (to be used when the work
is first cited \cite[p.\ 407]{chicago-82}).
The {\tt sfrefs} environment would {\em not} have a
{\it widest-label} argument.

The successors to the ``standard styles'' would arrange for \verb+\firstcite+
to produce the {\it fairly-full-form} and \verb+\sfcite+ to produce the
{\it short-form}.%
\footnote{%
This makes the pessimistic assumption that \LaTeX\ can not itself determine
whether a citation is the \lq\lq first citation'' of a particular work.
If someone was ingenious enough to produce code that determined whether a
citation is a \lq\lq first citation'', {\tt\ttbackslash firstcite} would be
unnecessary (except, perhaps, for the construction of tables of abbreviations
\cite[p.\ 414]{chicago-82}.)%
}
To conform to the ISO specification \cite{iso-690}, neither
\verb+\firstcite+ nor \verb+\sfcite+
would add any punctuation (but, for example, a style file that implemented
the MLA conventions would have to add brackets \cite[ch.\ 5]{gibaldi}).
In the successors to the ``standard styles'',
the \verb+\sfentry+ would produce an entry with no label.

People producing other style files would be free to implement other
conventions \cite[p.\ 168]{butcher-81}.

Some works may have no bibliography or may have just a \lq\lq select
bibliography'' \cite[p.\ 168]{butcher-81}.  For such works, it will still
be necessary to supply the details for use by \verb+\firstcite+ and
\verb+\sfcite+.  It may therefore be worth allowing a form of
{\tt sfrefs} (e.g., \verb+\begin{sfrefs}[null]+)
that holds details of cited works but does no typesetting.

\paragraph{Additional references}
\label{3.0-additional}

Since the entries in an \lq\lq additional list'' will not be cited as such
(although an \lq\lq all sources consulted'' list may contain a duplicate of
a cited entry in a \lq\lq works cited'' list), the list will be typeset
without \lq\lq labels''.  Even in a document that uses the \lq\lq reference by
number'' citation scheme (and so needs \lq\lq labels'' in the {\tt numrefs}
list), there will be no \lq\lq labels'' for the entries in an additional list.

It therefore seems likely that the {\tt morerefs} environment could
be implemented as a variation of {\tt adrefs} or {\tt sfrefs}, the main changes
being:
\begin{itemize}
\item a change of title (but see section \ref{reflist-scope})
\item absence of
      {\it cite-key, author, date, fairly-full-form} and {\it short-form}
      arguments.
\end{itemize}

\subsubsection{Order within the reference-list}

As stated in section \ref{div-of-labour}, it is probably best to leave
any sorting of the reference-list to some other software, or to the author.

However, it might be possible for \LaTeX\ to provide a warning if a
reference-list is obviously in the wrong order.  Perhaps:
\begin{itemize}
\item although there may be no easy alternative to numbering
      \verb+\numcite+s in order of appearance within {\tt numrefs}
      (even though \lq\lq order of first citation'' is usually what
      is required), \LaTeX\ could give a warning if a \verb+\numcite+
      gave a number that exceeded the \lq\lq biggest number produced
      by \verb+\numcite+ so far'' by more than one.
\item there could be a warning if an \verb+\adentry+ had an
      {\it author} whose first letter came before the first letter of the
      preceding \verb+\adentry+'s {\it author} in the alphabet.
\end{itemize}

\subsubsection{Citation of a specific division}
\label{division}

As stated in section \ref{2.09-citations}, provision needs to be made for
the citation of a particular division (e.g., page, section, chapter,
equation) of another work.  The syntax of citation commands should
not only allow several works to be cited simultaneously, but should also
allow the relevant division of each work to be specified.

From the author's point-of-view, there would be a variety of satisfactory
ways to specify citations that are to appear as
``[4, p.\ 10; 5; 6, p.\ 3]'', e.g.
\begin{verbatim}
    \numcite{smith[p. 10],brown,jones[p. 3]}
    \numcite{smith, p. 10; brown; jones, p. 3}
    \numcite{smith & p. 10; brown; jones & p. 3}
\end{verbatim}
The precise syntax would have to take
account of the practicalities of programming a command that has to be able
to accept pairs of arguments, where the second member of each pair is optional.

Analogous facilities would be needed for author-date and \lq\lq short form''
citations.

Incidentally, since abbreviations such as p., ch., sec., and fig.\ are
common when such divisions are specified, I think that citation commands
should arrange for the optional arguments to be typeset with
\verb+\frenchspacing+.

\subsection{Details of reference-lists}

\subsubsection{Variations within mainstream publishing practice}
\label{mainstream-lists}

Although many academic and technical publications involve only a single
undivided reference-list, some such publications involve:
\begin{description}
\item[more than one list]
      This situation can arise:
      \begin{itemize}
      \item when there is a list of \lq\lq further reading'' etc.,
            as well as the list of \lq\lq works cited''.
            This case has been covered in sections \ref{support-additional},
            \ref{2.09-additional}, \ref{4-sets} and \ref{3.0-additional}.
      \item when conference proceedings are produced, since each
            contribution may have its own reference-list.
      \item in manuals for software.  For example, in \cite{norusis}, a
            software house's own publications are introduced in the preface
            and cited (in effect) using a \lq\lq short form'' scheme, while
            other people's publications are listed at the end of the manual
            and are cited using the author-date scheme.
      \end{itemize}
\item[subdivisions within a list]
      Some reference-lists, particularly in the humanities, are subdivided
      according to the source of the cited documents (see
      \cite[p.\ 183]{butcher-81}, \cite[p.\ 425]{chicago-82} and
      \cite[p.\ 88]{gibaldi}).
\end{description}

In some cases, an author may wish to add explanatory paragraphs describing,
for example, how material was chosen for a \lq\lq select bibliography''
\cite[fig.\ 15.11]{chicago-82} or information about access to (document)
archives \cite[fig.\ 15.16]{chicago-82}.

\subsubsection{\protect\LaTeX\ 2.09}

\LaTeX\ 2.09 can cope with documents that have more than one
{\tt thebibliography} environment, and seems to deal satisfactorily
with a situation in which some \verb+\cite+ commands are to
one {\tt thebibliography} and some are to another (provided that
the {\it cite-key\/}s are unique).  The default effect is
to give {\it label\/}s that are not unique, which will be acceptable
when each contribution to a \lq\lq conference proceedings'' has its
references numbered from one, but not if \lq\lq works cited''
and \lq\lq additional references'' are both numbered from one
(see section \ref{2.09-additional}).

The specification of {\tt thebibliography} \cite[p.\ 187]{lamport-86}
does not allow anything other than \verb+\bibitem+s within a
{\tt thebibliography} environment.  Hence, it is not clear
how one can introduce subheadings within a reference-list.
(In practice, a \verb+\section*+ seems to work between
\verb+\bibitem+s, but I suspect that it puts \LaTeX\ 2.09 into a loop
if placed before the first \verb+\bibitem+.)

Anyone trying to add explanatory paragraphs (as in \cite[fig.\
15.11]{chicago-82} and \cite[fig.\ 15.16]{chicago-82}) will probably
find that \LaTeX\ 2.09 objects that
\begin{quote}
\lq\lq{\verb+Something's wrong--perhaps a missing \item+}''.
\end{quote}

\subsubsection{\protect\LaTeX\ 3.0}

Ideally, in order to provide support for the conventions that are routine
in mainstream publishing practice, \LaTeX\ 3.0 should be able to cope with
all the variations outlined in section \ref{mainstream-lists}.

\paragraph{Multiple lists}

The suggestions made in section \ref{what-to-do} would probably cater for most
situations where a document has more than one reference-list.

The distinction between {\tt morerefs} and the other environments for
reference-lists would take care of situations where there is a list of
\lq\lq additional references'' as well as a list of \lq\lq works cited''.
The distinction between the \lq\lq short form'' commands/environment
and the other commands/environments would take care of situations where a
software house uses \lq\lq short form'' for its own publications
and some other scheme for other publications.  Conference proceedings
will be able to have \lq\lq a reference-list for each contribution''
if \LaTeX\ 3.0 follows the \LaTeX\ 2.09 precedent that allows
more than one {\tt thebibliography} in a document.

\paragraph{Subdivisions and explanatory paragraphs}

I have the impression that, because the sciences have different conventions
from the humanities, people using the \lq\lq reference by number'' citation
scheme are unlikely to want the options of subdividing their reference-list
and inserting explanatory paragraphs.  Therefore, it would be legitimate to
say (for example) \lq\lq subdivisions and explanatory paragraphs are supported
within {\tt adrefs}, {\tt sfrefs} and {\tt morerefs} but not within
{\tt numrefs}'', if this made the programming task easier.

For example, it might be convenient to implement {\tt numrefs} as
a \lq\lq list-making environment'' (as in \LaTeX\ 2.09) but to implement
the other environments in some other way.  Lack of support for subdivisions
and explanatory paragraphs is unlikely to matter in {\tt numrefs}; the
\lq\lq other way'' (more like \verb+\paragraph+, perhaps?) might make it
easier to implement support for these facilities in the other environments.

\paragraph{Other problems}
\label{reflist-scope}

One approach to some of the problems mentioned in section \ref{2.09-other}
is for the standard styles to define the heading for the reference-list by,
for example, \verb+\def\numrefsheading{References}+, so that anyone who wants
to change the heading can do so by issuing a \verb+\renewcommand+ command
somewhere before the start of their reference-list.

Another approach is to work in terms of the standard publishing industry
concept of \lq\lq back matter'' \cite[p.\ 4]{chicago-82}.  Instead of having to
understand where the {\it heading}, {\it left-head}, {\it right-head} and
table-of-contents entry (or lack of it) originate for units such as the
glossary (if any), the reference-list(s) and the index (if any)
{\em separately}, an author would only have to understand how these features
are treated {\em consistently} within \lq\lq back matter''.

Although the \lq\lq back matter'' approach could be used if environments
such as {\tt numrefs} followed the {\tt thebibliography} precedent and issued
commands such as \verb+\chapter+ or \verb+\section+ themselves, authors
might find the way that {\it heading}, {\it left-head}, {\it right-head} and
table-of-contents entry materialize less mysterious if it was just the same
for a reference-list as for (say) a glossary.  This would imply that
{\tt numrefs}, {\tt adrefs}, {\tt sfrefs} and {\tt morerefs} should
not issue commands like \verb+\chapter+ or \verb+\section+ themselves.
As a bonus, it would then be possible for an author to insert an explanatory
paragraph before the reference-list, and to arrange for subdivisions.

For example, if there was a {\tt backmatter} environment within which
\verb+\chapter+ was treated as defining a unit of \lq\lq back matter'',
an author's file might contain commands such as
\begin{verbatim}
\begin{backmatter}
   \chapter{Glossary}
      ...
   \chapter{References}
      \section{Primary sources}
      \begin{sfrefs}
      ...
      \end{sfrefs}
      \section{Secondary sources}
      \begin{sfrefs}
      ...
      \end{sfrefs}
   \chapter{Further reading}
      \begin{morerefs}
      ...
      \end{morerefs}
\end{backmatter}
\end{verbatim}

\subsection{Conclusion}

\LaTeX\ has a large number of users, and potential users, who wish to produce
documents that conform to the conventions that are standard in academic
publishing.  One element of their requirement is the need to conform to the
conventions for citations and reference-lists that are usual in their
disciplines.

The choice for \LaTeX\ 3.0 may be between:
\begin{enumerate}
\item having more facilities for citations and reference-lists than
      \LaTeX\ 2.09, perhaps as suggested in section \ref{what-to-do}.
      This would imply an increase in:
      \begin{itemize}
      \item the amount of code needed to implement the facilities, and
            the guru time needed for writing the code
      \item the number of pages needed, in the successor to \cite{lamport-86},
            to describe the facilities --- perhaps 8 pages rather than
            the 2 pages in \cite{lamport-86}.
      \end{itemize}
\item no significant increase in the facilities provided for citations
      and reference-lists.  Contrary to the idea of \lq\lq freeing people
      from formatting concerns to allow them to concentrate on writing''
      \cite[p.\ 8]{lamport-86}, many authors (perhaps most authors) would be
      wasting time:
      \begin{itemize}
      \item hacking at style-files
      \item searching (software) archives for ready made solutions
      \item taking up support staff's time in the search for advice
            (with the support staff
            in turn taking up gurus' time in their search for solutions).
      \end{itemize}
      Moreover, most of this time would be wasted by (or on behalf of) authors
      who don't want anything at all exotic; they just want to conform
      to the conventions that are standard in traditional academic publishing.
\end{enumerate}
I'm inclined to think that the first option would be the lesser of the
two evils.

\subsection*{Appendix: Interface with \protect\BibTeX}
\addcontentsline{toc}{subsection}{Appendix: Interface with \protect\BibTeX}

The preceding sections make some suggestions for a \LaTeX\ 3.0 user
interface that would enable the do-it-yourself-er to conform to
the conventions that are usual in academic publishing.
It is also necessary to consider the implications for the interface to
\BibTeX.

\subsubsection*{Single reference-list}

Inspecting {\tt bst} files gives me the impression that, if it
was decided to adopt a scheme such as that described in section
\ref{what-to-do}, it would be fairly easy to produce new {\tt bst} files
to supersede existing ones.  For example, a {\tt bst} file
that implemented a \lq\lq reference by number'' scheme would write
\verb+\numentry+ commands rather than \verb+\bibitem+ commands.

If {\tt bst} files were created in this way, they would be able to
deal with the straightforward situation when there is a single
reference-list that contains all works cited (plus, possibly,
any works specified by a command like \verb+\nocite+).

\subsubsection*{Multiple reference-lists, all with the same style}

More complicated situations can arise in which a document involves
several reference-lists.  For example:
\begin{itemize}
\item The editor of the proceedings of a conference might want the
      published proceedings to have a reference-list at the end of each
      chapter.
\item If, as suggested in section \ref{reflist-scope}, {\tt sfrefs} was
      implemented in a way that allowed a sequence of {\tt sfrefs} environments
      to each be preceded by a \verb+\section+ command, then, as far as
      \BibTeX\ is concerned, each {\tt sfrefs} environment might be a separate
      reference-list.
\end{itemize}

In both these examples, the document would involve several reference-lists,
but each reference-list would need to be typeset in a common style.
I assume that the main problems would be in arranging:
\begin{itemize}
\item to have multiple {\tt bbl} files, or distinct divisions of a single
      {\tt bbl} file
\item that each reference-list takes its entries from the correct {\tt bbl}
      file, or from the correct division of a single {\tt bbl} file.
\end{itemize}

\subsubsection*{Two reference-lists, each with a different style}

People producing documents that have a second reference-list (e.g.,
\lq\lq further reading'') in addition to the list of \lq\lq works cited''
might want the first list typeset in one style and the second
list typeset in another.  (In particular, if the \lq\lq reference by number''
scheme is used, the first list will have \lq\lq labels'' but the second
list will have no \lq\lq labels''.)  If the commands/environments suggested
in section \ref{what-to-do} were implemented, the first list would use
{\tt numrefs}, {\tt adrefs} or {\tt sfrefs}, while the second list would
use {\tt morerefs}.

In this situation, it would be necessary to communicate to \BibTeX\
that two lists are required, but that they are to be typeset in different
styles.  Since the second list is to contain \lq\lq works {\em not\/}
cited'', it will also be necessary to specify the works to be shown in the
second list.

If it is decided to extend the interface between \LaTeX\ and \BibTeX\ to
cater for such situations, it will probably be necessary to consider
defining alternatives to \LaTeX\ 2.09's \verb+\bibliography+ and
\verb+\bibliographystyle+ commands, since it seems unlikely that the syntax
of the \LaTeX\ 2.09 commands could be extended so as to pass the necessary
information.  One might, for example, consider syntax such as
\verb+\bibtexcites[+{\it cites-style}\verb+]{+{\it bib-files}\verb+}+
and
\verb+\bibtexmore[+{\it more-style}\verb+]{+{\it bib-files}\verb+}{+{\it
      key-list}\verb+}+.
Here, it is assumed that:
\begin{itemize}
\item the \verb+\documentstyle+ would set defaults for the {\it cites-style}
      and {\it more-style} that are to be passed to \BibTeX, but the user
      can over-ride the defaults via the optional arguments to
      \verb+\bibtexcites+ and \verb+\bibtexmore+
\item \verb+\bibtexmore+'s {\it key-list} argument would be used to
      specify the works to be included in the list of \lq\lq additional
      references''.
\end{itemize}

If it is not practicable to extend the \LaTeX/\BibTeX\ interface to cater for
these situations automatically, it would presumably be a matter of
some \lq\lq human intervention'':
\begin{itemize}
\item to prepare a {\tt bbl} file for the \lq\lq additional references''
\item to \verb+\input+ the {\tt bbl} file.
\end{itemize}

\section{Further e-mail comments}

\begin{footnotesize}\begin{verbatim}
From:     David Rhead ...
Date:     4 Apr 91 12:16:14

Here are a few comments on Nico's comments (5th March) about my paper on
citations and reference-lists.


>>  2. The paper on reference lists concentrates too much on layout and not
>>  enough on structure. For a, in my humble opinion, much more valuable
>>  discussion of LaTeX 3.0, reference lists and BibTeX I'd like to refer to the
>>  talk Frank Mittelbach gave at the Cork conference last year.

Frank has sent me a copy of "BibTeX reconsidered", by Reinhard Wonneberger
and himself.  I think that this is the written version of the talk that
he gave at Cork.  There seems to be a lot of common ground between my paper
and "BibTeX reconsidered", although the former is looking at it from the LaTeX
point-of-view while the latter is looking at it from the BibTeX point-of-view.
[Both my paper and "BibTeX reconsidered" would like support for multiple
bibliographies, the short title/form scheme (often in footnotes) and
commented bibliographies.]

I'd like "document support" (as Reinhard/Frank put it) for reference-by-number,
author-date and short-form citation schemes.  My paper suggested one possible
user interface through which this support could be provided.  Of course,
other interfaces could be defined that would do the job.  I think that the
main thing is to provide satisfactory support for the 3 schemes.


>>  I'm totally opposed to the idea of having different coding schemes for
>>  different systems of citation. In my opinion, this goes completely against
>>  the basic idea behind LaTeX and SGML, namely separation of form and contents.
>>  Consider the amount of re-coding when switching from the number system to the
>>  name-year system!

There are differences between the schemes that may be so significant that
they could/should be regarded as "different in form".
*   For instance, ISO 690 gives the example
        The notion of an invisible college has been explored in the
        sciences (24).  Its absence among historians is noted by
        Steig (13, p. 556).  It may be as Burchard (8) points out ...
    which, if converted to author-date, would be
        The notion of an invisible college has been explored in the
        sciences (Crane, 1972).  Its absence among historians is noted by
        Steig (1981, p. 556).  It may be as Burchard (1965) points out ...
    where the substitution has to take account of whether the author's name
    does or does not occur naturally in the sentence.  I.e. the form of
    citation depends on what else is in the sentence, and conversion
    cannot easily be automated.  Similarly, I'd be surprised if one can
    guarantee to be able to change from other schemes to the
    short-form scheme without some re-writing.
*   Reference-by-number involves keeping track of one thing (the number),
    author-date involves keeping track of two (surname and date), short-form
    involves keeping track of at least two things (form for first citation,
    form for subsequent citations, perhaps also the full form as the default
    form for first citation).
If the schemes are different in form (i.e. involve different logical
structures), it may be legitimate to consider having different coding
schemes, while still aiming at separation of form from content (e.g.
whether reference-by-number uses superscripts or brackets).

However, if someone has sufficient insight to be able to propose a single user
interface that can cater for all three schemes (plus possible "additional
references") within one set of commands/environments, I agree that it
would be very nice.  My own attempts to define a single interface
that would cater for all three schemes have ended up being unsatisfactory.
Here's how they end up unsatisfactory.
                              ------------
For the purpose of this account I'll use \refentry to mean the successor
to \bibitem.  Presumably \refentry would have to have 2 or 3 arguments besides
the cite-key (rather than \bibitem's one extra argument) so that the arguments
could be used for:
- nothing, in the reference-by-number scheme (since LaTeX would supply
  the numbers)
- author and date, in the author-date scheme
- first-citation-form and subsequent-citation-form, for the short-form
  scheme.
Thus, for reference-by-number, the user (or BibTeX) would supply
\refentry{cite-key}{}{} ...
but for author-date they would supply
\refentry{cite-key}{author}{date} ...
while for short form they would supply
\refentry{cite-key}{fairly-full-form}{short-form} ...
[We're already in trouble.  How do we deal with what appears to be the tendency
in short form for the "full reference" to be the default "fairly full form"?
Do we go for \refentry{cite-key}[fairly-full-form]{short-form}{full reference},
and if so what are the implications for the other 2 schemes?]
Or should one go for
\refentry{cite-key}{author}{date}{fairly-full-form}{short-form} ...
so as to give LaTeX all the information it needs to allow completely
automatic switching between citation schemes, separating form from content
but placing a heavy burden on the user (who would have to provide
at least dummy information for both author-date and short-form
schemes, even if they are only going to use reference-by-number)?

For citation commands one might have \cite and \shortcite (following
precedents in the archives) [but see section 6.5 of my paper, about citation
of a specific devision].  By aiming at author-date, one might be able to
get a .tex file that also worked for reference-by-number.  For example,
if \cite and \shortcite both gave a number when a .sty file implemented
reference-by-number, while \cite gave (author, date) and \shortcite
gave (date) when a .sty file implemented author-date, the following
input might work for both schemes:
    The notion of an invisible college has been explored in the
    sciences \cite{crane-72}.  Its absence among historians is noted by
    Steig \shortcite{steig-81[p. 556]}.  It may be as Burchard
    \shortcite{burchard-65} points out ...
I don't think that the reverse would work, i.e. if you think in terms
of reference-by-number while writing, you'll do things like
    Steig \cite{steig-81[p. 556]}.  It may be as Burchard
which, if converted automatically to author-date, would give
    Steig (Steig, 1981; p. 556).  It may be as Burchard
which gives two occurences of Steig, which isn't right.
For short-form, you might interpret \cite as meaning "the form of citation
used at first citation" and \shortcite as meaning "the form to be used
subsequently".  But if you do this, you'll put your \shortcites in
places that are different from the places you'd put them for author-date
For example, you might have (using an ISO 690 example again)
    ... Steig \cite{steig-81}.  It may be as Burchard
    \cite{burchard-65} points out ... Steig \shortcite{steig-81}
    has further noted ...
for the MLA or
    ... Steig\footnote{\cite{steig-81}} ... Steig\footnote{\shortcite{steig-81}}
for most other publications that use short-form.  [This assumes that the first
of these citations of Steig above is actually the first citation of Steig
in the whole document (or chapter).]  But this would not give you the right
input file for author-date, for which you would want
    ... Steig \shortcite{steig-81}.  It may be as Burchard
Again, you might be able to convert automatically TO reference-by-number,
but it seems unlikely that you could convert automatically FROM
reference-by-number.  The problem might disappear if LaTeX itself could work
out which citations are "first citations".  Then someone can put \cites and
\shortcites in the places required for author-date but choose a .sty file
that gives short-form; the .sty file would ignore the distinction between
\cite and \shortcite and would use the fairly-full-form for "first citation"
and the short-form for subsequent citations.  But can LaTeX work out which
citations are "first citations"?
                              ------------
Although one can try (as above) to imagine a single set of
commands/environments that would support all three schemes, it seems to
lead to problems, and I worry that, in making things right for one scheme,
they'll be made wrong for another.  So I gave up on the idea of a single
set of commands, which was why my paper went for the idea of separate
commands/environments tailored to the 3 specific citation schemes.
But, if someone has more success than I did in imagining how a single set of
commands/environments could be simultaneously compatible with all three schemes,
I'd be interested to see their proposals.

Or should one be less ambitions, perhaps aiming to satisfy people who
want reference-by-number and author-date, and not bothering about people
who want short-form?  Things like \refentry{cite-key}{author}{date},
\cite and \shortcite might work for both reference-by-number and author-date.
(People who want reference-by-number would have to supply author and date
information that they might regard as redundant, unless they go
\refentry{cite-key}{}{} or the syntax is something like
\refentry{cite-key}[author][date].)


>>  To David's review I'd like to add that \bibitem's have no sub-division, at
>>  least not one that is indicated by explicit control sequences (`tags').
>>  Instead, the tagging of \bibitem's is done _outside_ LaTeX, which has always
>>  struck me as odd.

In theory, it would be nice if the \bibitem's did have subdivisions.  There
might be subdivisions for the do-it-yourself-er that were analogous to
the fields used by the BibTeX-er.

In practice, to sub-divide the \bibitems would involve deciding what the
subdivisions should be, which leads one into questions that cause difficulty
in BibTeX (e.g., "Should it be address or place-of-publication?",
"Is it really worth having booklet separate?", "Is it really worth
having phdthesis separate from mastersthesis?", "Does volume mean
number-in-series or subdivision-of-book?").  It may not be easy to
answer these questions.  I don't think that it is worth delaying
LaTeX 3.0 while answers are sought.  [One could always return to the
question for LaTeX 4.0, if there was one!]
\end{verbatim}\end{footnotesize}