diff options
author | Karl Berry <karl@freefriends.org> | 2006-01-11 23:51:02 +0000 |
---|---|---|
committer | Karl Berry <karl@freefriends.org> | 2006-01-11 23:51:02 +0000 |
commit | 43bf9b8eb1efa49bd26f252b98a248de4cf6ae3b (patch) | |
tree | e0caf47b2991c7ed7f679f375f1d9c2e4e74ecf1 /Master | |
parent | 19b13f9a9ce2c738bf2c0d1045f40b82aa3c817a (diff) |
trunk/Master/texmf-dist/source/latex/authorindex
git-svn-id: svn://tug.org/texlive/trunk@149 c570f23f-e606-0410-a88d-b1316a301751
Diffstat (limited to 'Master')
-rw-r--r-- | Master/texmf-dist/source/latex/authorindex/authorindex | 510 |
1 files changed, 510 insertions, 0 deletions
diff --git a/Master/texmf-dist/source/latex/authorindex/authorindex b/Master/texmf-dist/source/latex/authorindex/authorindex new file mode 100644 index 00000000000..d0287335ab2 --- /dev/null +++ b/Master/texmf-dist/source/latex/authorindex/authorindex @@ -0,0 +1,510 @@ +#!/usr/bin/perl -w +# This file is part of the authorindex package for LaTeX + BibTeX. + +use strict; + +# configuration: apart from the command to invoke perl above you might want to +# change: + +my $bstenv="BSTINPUTS"; # Environment Variable holding .bst search path +my $bibenv="BIBINPUTS"; # Env Variable holding search path for BibTeX databases +my $tmp="_autidx_"; # Name base for temporary files + +# This script takes LaTeX .aux files as input. It extracts all citations made +# with page number information. These, together with the bibliography data base +# extracted from the input and a .bst file are processed by bibtex to get a +# file that associates each citation label with corresponding authors. Together +# with the knowledge of which work is cited on which page, this is used to +# compute which author is cited on which page. This information is written to +# the output in form suitable to be included in a LaTeX document. +# Alternatively, the script can also extract the label that appears in the +# references for each work instead of the pages. Also, output for further +# processing by makeindex can be generated instead of a 'ready' LaTeX file. + +# examine command line for options + +my %opt; # hash to contain options found +use Getopt::Std; +getopts('dhikpr',\%opt); + +# handle -h option: a short description of the script. + +exists $opt{'h'} && die <<END; +Usage: $0 [-d] [-h] [-i] [-k] [-p] [-r] [filename ...] + +-d generate additional statistical information as comments +-i generate file (with extension .ain) suitable as input for makeindex +-k keep auxiliary bibliography style file $tmp.bst after script finishes +-h print this help and exit +-r suppresses automatic inclusion of .aux generated by \\include-ed files +-p send result to standard output + +Any number of file names can be given. If none is given, input is read from +standard input. Otherwise, all files specified are read, after the extension +.aux has been appended to their name where necessary. +END + +# Make sure that all input files get .aux extension. + +grep(s/$/\.aux/,grep(!/\.aux$/,@ARGV)); +my @SAVEARGV=@ARGV; + +my $usenum=0; # default: put page numbers into index +my $PageTypeOrder="rRnAa"; # default order of number types +my $altedit="'skip\$"; # no editors if no authors present +my $addedit="'skip\$"; # no editors if authors present + +my $citationcount=0; # used for statistics +my $workscount=0; +my $explicits=0; +my $authorcount=0; + +my $miniindex=0; # flag: is 1 if mini-index is requested +my $nocompress=0; # flag: is 1 if page range compression is prohibited + +my %bib; # collect .bib-file names as defined keys of hash +my %pnlist; # collect occuring pages/bib.numbers as keys of hash + +my $editors=0; # number: 0 no editors; 1 editors if no authors + # 2: editors whenever present +my $nameformat=""; # string to format/sort names +my $maxnames=999; # max. number of author names per entry +my $truncnames=999; # how many authors to take if max number is exceeded +my $labeltype="labels"; # wether bibliography labels or pages are indexed + +my %Lab2Pag; # for each label gives array of pages where work cited +my %Lab2Num; # for each label give its bibliography number +my %Lab2Bib; # ditto, but for occurrence in bibliography +my %Aut2Pag; # for each author gives array of pages cited/in bilio. +my %Aut2Bib; # ditto, but only occurrence in bibliography +my %Aut2Lab; # list of BibTeX entry labels an author occurs in +my %LeadAutPag; # author -> hash marking pages with work author is 1st + +my %printname; # sort-key -> printed representation of author +my %plainname; # sort-key -> author name + +my %PageTypeOrder; # Page type code -> number giving relative order +my %PageOrder; # page string -> array used to sort pages + +my $see=""; # string to separate other and first author, or undef. +my $bst=""; # name of BibTeX program to extract author names +my $output=""; # name of file to which author index is written +my $twoabbrev=""; # string to append to page for 2 subsequent pages + +# scan input files and +# - build the file later to be processed by BibTeX, +# - generate a temporary bibtex database of the explicit author names given, +# - assemble for each citation the page where it was referenced and +# - look for data base specification, output file name, and so on. + +open(AUXFILE,">$tmp.aux") || die "Can't open temporary file $tmp.aux\n"; +open(BIBFILE,">$tmp.bib") || die "Can't open temporary file $tmp.bib\n"; + +while(<>){ + if(/^\\citationpage\{([^{]+)\}\{(.+)\}$/){ + $citationcount++; # used for statistics only. + $workscount++ unless ($1 eq '*') or (exists $Lab2Pag{$1}) + or (exists $Lab2Bib{$1}) or (exists $Lab2Num{$1}); + $pnlist{$2}=""; + push @{$Lab2Pag{$1}},$2; + print AUXFILE "\\citation{$1}\n"; + }elsif(/^\\bibcite\{([^{]+)\}\{(.+)\}$/){ + $workscount++ unless ($1 eq '*') or (exists $Lab2Pag{$1}) + or (exists $Lab2Bib{$1}) or (exists $Lab2Num{$1}); + $pnlist{$2}=""; + push @{$Lab2Num{$1}},$2; + print AUXFILE "\\citation{$1}\n"; + }elsif(/^\\bibpage\{([^{]+)\}\{(.+)\}$/){ + $workscount++ unless (exists $Lab2Pag{$1}) + or (exists $Lab2Bib{$1}) or (exists $Lab2Num{$1}); + $pnlist{$2}=""; + push @{$Lab2Bib{$1}},$2; + print AUXFILE "\\citation{$1}\n"; + }elsif(/^\\aiexplicit\{(.+)\}\{(.+)\}$/ and !$usenum){ + # above: page number might not contain '}{' + print BIBFILE "\@MISC{$tmp$explicits,author=\"$1\"}\n"; + $bib{$tmp}=""; + $pnlist{$2}=""; + push @{$Lab2Pag{"$tmp$explicits"}},$2; + print AUXFILE "\\citation{$tmp$explicits}\n"; + $explicits++; + }elsif(/^\\bibdata\{(.+)\}$/){ + @bib{split(",",$1)}=""; + }elsif(/^\\aistyle\{(.+)\}$/){ + warn "Multiple \\authorindexstyle\n" if $bst && ($bst ne $1); + $bst=$1; + }elsif(/^\\aioptions\{(.*)\}$/){ + ($editors,$nameformat,$maxnames,$truncnames,$labeltype)=split /\|/,$1; + $usenum|=($labeltype eq "labels"); + if($editors>0){ + $altedit="{ editor format }"; + $addedit=$altedit if $editors==2; + } + }elsif(/^\\aifilename\{(.+)\}$/){ + warn "Warning: Multiple authorindices\n" if $output; + $output=$1; + }elsif(/^\\\@input\{(.+)\}$/){ + push(@ARGV,$1) unless exists $opt{'r'}; + }elsif(/^\\pagetypeorder\{([rRaAn]+)\}$/){ + $PageTypeOrder=$1; + }elsif(/^\\aiseestring\{(.+)\}$/){ + $see=$1; + }elsif(/^\\aitwostring\{(.+)\}$/){ + $twoabbrev=$1; + }elsif(/^\\aiinbibflag$/){ + $miniindex=1; + }elsif(/^\\ainocompressflag$/){ + $nocompress=1; + } +} + +close BIBFILE; + +# output can go to stdout or a filename found in the input files. + +$output || die "You have to include .aux file produced by .tex file containing \\begin{document}\nin the argument list and you have to \\usepackage{authorindex}!\n"; +$output="-" if exists $opt{'p'}; + +# We need at least one BibTeX database + +my $bib=join(",",keys %bib) + || die "You must specify at least one BibTeX database\n"; +print AUXFILE "\\bibdata{$bib}\n"; + +# if the user hasn't explicitly given a .bst style for formatting author names, +# we generate our own based on the style options found in the input files. + +unless($bst){ + my @nameformat=split /;/,$nameformat; + my ($printkey,$namefmtcmd)=("cite\$ write\$ newline\$\n",""); + for (@nameformat){ + my ($namerep,$sortrep)=split /:/; + $sortrep=$namerep unless $sortrep; + $namefmtcmd.= + "duplicate\$ names swap\$ \"$namerep\" format.name\$ " . + "write\$ newline\$\n" . + "duplicate\$ names swap\$ \"$sortrep\" format.name\$ " . + "purify\$ \"u\" change.case\$ write\$ newline\$\n$printkey"; + $printkey="newline\$\n"; + } + + $ENV{$bstenv}=".:" . (exists $ENV{$bstenv} ? $ENV{$bstenv} : ""); + $bst=$tmp; + open(BSTFILE,">$bst.bst") || die "Can't open $bst.bst\n"; + print BSTFILE <<END; # Now comes the BibTeX programm inlined... +% Temporary file generated by $0 +entry{author editor}{}{} strings{names} integers{numnames} +function{format} +{ duplicate\$ empty\$ % field present? + { pop\$ } % no: do nothing but cleanup + { duplicate\$ 'names := % memorise namelist in variable "names" + num.names\$ duplicate\$ + #$maxnames > % too many names in list? + { pop\$ #$truncnames } % yes, truncate. + 'skip\$ % no, keep them all + if\$ + 'numnames := % save number of names + #0 % start index + { duplicate\$ numnames < } % test for "while\$" + { #1 + % next name + duplicate\$ names swap\$ % get name list and index + "{ll}" format.name\$ % format curr. name + "others" = % et al part? + 'skip\$ % yes, do not output + { $namefmtcmd } % no: format all + if\$ + } while\$ pop\$ % loop until index is 0 + } + if\$ +} +function{default.type}{author format author empty\$ $altedit $addedit if\$ } +function{article}{default.type} function{book}{default.type} +function{booklet}{default.type} function{inbook}{default.type} +function{incollection}{default.type} function{inproceedings}{default.type} +function{conference}{default.type} function{manual}{default.type} +function{mastersthesis}{default.type} function{misc}{default.type} +function{phdthesis}{default.type} function{proceedings}{default.type} +function{techreport}{default.type} function{unpublished}{default.type} +read iterate{call.type\$} +END + # ... and here comes perl again. + close BSTFILE; +} + +# Now we have decided on our .bst file and can finish the temporary .aux file +# we prepared for BibTeX. + +print AUXFILE "\\bibstyle{$bst}\n"; +close AUXFILE; + +# if we have written to the temporary database, make sure BibTeX can find it. + +$ENV{$bibenv}=".:" . (exists $ENV{$bibenv} ? $ENV{$bibenv} : "") if $explicits; + +# We now give BibTeX all the citation labels. In return we get a file whose +# lines in turn contain an author name and a label of a work of that author. +# The format the author names are given are determined by the BibTeX style file +# $bst.bst. + +print STDERR `bibtex $tmp`; +die "BibTeX error. Aborting leaving all temporary files $tmp.*\n" if $?; + +# if things went well, we can delete all these temporary files made for BibTeX. +# The generated .bst file is kept if the user wishes so (-k option). + +unlink "$bst.bst" if ($bst eq $tmp && !(exists $opt{'k'})); +unlink "$tmp.aux","$tmp.bib"; + +# Decide wether pages or citation labels go to the index + +my %Lab2Ent=%Lab2Pag; +if($usenum){ + %Lab2Ent=%Lab2Num; + %Lab2Bib=(); +}else{ + %Lab2Num=(); +} + +# We have now labels associated with page numbers and labels associated with +# author names (in the file generated by the BibTeX run). Now we can bring +# together the previous two main steps and compute for each author the pages +# where she is cited. In draft mode, we also remember for each author the +# labels of her works and the pages where these works are cited. + +my ($firstauthor,$firstsortname,$Lab,$PrevLab)=("","","",""); +open(BIBFILE,"$tmp.bbl") || die "Can't open $tmp.bbl\n"; +while(<BIBFILE>){ + chop; s/[\[\]]//g; + my $author=$_; + my ($sortname,$LabOrEmpty); + chop($sortname=<BIBFILE>); + chop($LabOrEmpty=<BIBFILE>); + if($LabOrEmpty){ + $PrevLab=$Lab; + $Lab=$LabOrEmpty; + } + my $printname=$author; + if($Lab ne $PrevLab){ + @{$LeadAutPag{$author}}{@{$Lab2Ent{$Lab}}}="" if exists $Lab2Ent{$Lab}; + @{$LeadAutPag{$author}}{@{$Lab2Ent{'*'}}}="" if exists $Lab2Ent{'*'}; + @{$LeadAutPag{$author}}{@{$Lab2Bib{$Lab}}}="" if exists $Lab2Bib{$Lab}; + $firstauthor=$author; + $firstsortname=$sortname; + }elsif($see){ + $printname="{$author}$see\\aifirst{$firstauthor}"; + $sortname="$sortname$see$firstsortname"; + } + @{$Aut2Pag{$printname}}{@{$Lab2Ent{$Lab}}}="" if exists $Lab2Ent{$Lab}; + @{$Aut2Pag{$printname}}{@{$Lab2Ent{'*'}}}="" if exists $Lab2Ent{'*'}; + @{$Aut2Pag{$printname}}{@{$Lab2Bib{$Lab}}}="" if exists $Lab2Bib{$Lab}; + @{$Aut2Bib{$printname}}{@{$Lab2Bib{$Lab}}}="" if exists $Lab2Bib{$Lab}; + push @{$Aut2Lab{$printname}},$Lab if (exists $opt{'d'}) and $LabOrEmpty; + $printname{$sortname}=$printname; + $plainname{$sortname}=$author; +} +close BIBFILE; +unlink "$tmp.blg","$tmp.bbl"; + +# Last not least, output the results, properly sorted if needed. + +open(AIFILE,">$output") || die "Can't create author index file $output\n"; + +# convert page type order into numerical values +my ($i,$page); +$PageTypeOrder{$i}=length($PageTypeOrder) while($i=chop $PageTypeOrder); +# create table that relates page to page order info +for $page (keys %pnlist){ + $PageOrder{$page}=&parse_pagenumber($page); +} + +if(exists $opt{'i'}){ + # generate file for makeindex: leave the work for makeindex. + my $name; + for $name (keys %printname){ + my $author=$printname{$name}; + my $page; + for $page (keys %{$Aut2Pag{$author}}){ + print AIFILE "\\indexentry{$name\@$author}{$page}\n"; + } + } +}else{ + # sort result, throw away duplicate page numbers and generate LaTeX file. + print AIFILE "\\begin{theauthorindex}\n"; + my ($prevfirstchar,$prevplain,$name)=("","",""); + for $name (sort keys %printname){ + my $thisfirstchar=substr($name,0,1); + if($thisfirstchar ne $prevfirstchar){ + print AIFILE "\\indexspace\n" if $prevfirstchar; + $prevfirstchar=$thisfirstchar; + } + my $author=$printname{$name}; + my $plain=$plainname{$name}; + my $rep=$author; + print AIFILE "% @{$Aut2Lab{$author}}\n" if (exists $opt{'d'}); + $rep=($plain eq $prevplain) ? "\\airep$author" : "\\aitop$author" + if $plain ne $author; + print AIFILE "\\item[$rep]"; + $Aut2Bib{$author}={} unless exists $Aut2Bib{$author}; + $LeadAutPag{$author}={} unless exists $LeadAutPag{$author}; + my %b2p=%{$Aut2Bib{$author}}; + my %lp=%{$LeadAutPag{$author}}; + my %pagerep; + my $page; + for $page (keys %{$Aut2Pag{$author}}){ + my $prep=$page; + $prep="\\aifirstpage{$prep}" if exists $lp{$page}; + $prep="\\aibibpage{$prep}" if exists $b2p{$page}; + $pagerep{$page}=$prep; + } + my $res=&compressed_pages($Aut2Pag{$author},\%pagerep); + print AIFILE " \\aipages{$res}\n"; + $prevplain=$plain; + $authorcount++; + } + print AIFILE "\\end{theauthorindex}\n"; + if(exists $opt{'d'}){ # in draft mode, include some statistics + print AIFILE "%\n% $citationcount citations "; + print AIFILE "of $workscount distinct works\n"; + print AIFILE "% $explicits times \\aimention\n" if $explicits; + print AIFILE "% $authorcount different authors\n"; + } +} +close AIFILE; + +# merge mini indices into the .bbl-Files if it was requested. + +if($miniindex){ + map s/aux$/bbl/,@SAVEARGV; + my $file; + for $file (@SAVEARGV){ + open(BBLINPUT,$file) || next; + open(BBLHELP,">$tmp.bbl") || die "Can't create temp file $tmp.bbl\n"; + my $currlabel=""; + while(<BBLINPUT>){ + if(/\\bibitem(\[.*\])*\{(.*)\}|\\end\{thebibliography\}/){ + if($currlabel){ + my $pagelist=&pages_for_label($currlabel); + print BBLHELP "\\bibindex{$pagelist}\n"; + } + $currlabel=$2; + print BBLHELP "$_"; + }elsif(/\\bibindex\{(.*)\}/){ + if($currlabel){ + my $pagelist=&pages_for_label($currlabel); + print BBLHELP "$`\\bibindex{$pagelist}$'"; + } + $currlabel=""; + }else{ + print BBLHELP "$_" if "$_" ne "\n"; + } + } + close BBLHELP; + close BBLINPUT; + rename "$tmp.bbl","$file" || die "Can't replace old $file\n"; + } +} + +# auxiliary functions + +# convert roman numeral string to integer + +sub romanvalue +{ + local($_)=shift; tr/IVXLCDM/ivxlcdm/; + my %romandigits = ("i", 1, "v", 5, "x", 10, "l", 50, + "c", 100, "d", 500, "m", 1000); + my ($i,$sum,$prev)=("",0,1); + while($i=chop){ + my $this=$romandigits{$i}; + $sum=$sum+(($this<$prev) ? -$this : $this); + $prev=$this; + } + return $sum; +} + +# convert letter to numeric value + +sub alphavalue +{ + local($_)=@_; tr/A-Z/a-z/; + return ord($_)-ord("a"); +} + +# split page number in components and replace each component by a number for +# the page type and the page number as an integer. + +sub parse_pagenumber +{ + local($_)=@_; + my $res=""; + while($_){ + s/^[^\\A-Za-z0-9]*//; + if(exists $PageTypeOrder{'n'} && s/(^\d+)//){ + $res.="$PageTypeOrder{'n'}".sprintf "%0.6d",$1; + }elsif(exists $PageTypeOrder{'R'} && + s/^\\uppercase\s*\{([ivxlcdm]+)\}//){ + $res.="$PageTypeOrder{'R'}".sprintf "%0.4d",&romanvalue($1); + }elsif(exists $PageTypeOrder{'R'} && s/(^[IVXLCDM]+)//){ + $res.="$PageTypeOrder{'R'}".sprintf "%0.4d",&romanvalue($1); + }elsif(exists $PageTypeOrder{'A'} && s/(^[A-Z])//){ + $res.="$PageTypeOrder{'A'}".sprintf "%0.2d",&alphavalue($1); + }elsif(exists $PageTypeOrder{'r'} && s/(^[ivxlcdm]+)//){ + $res.="$PageTypeOrder{'r'}".sprintf "%0.4d",&romanvalue($1); + }elsif(exists $PageTypeOrder{'a'} && s/(^[a-z])//){ + $res.="$PageTypeOrder{'a'}".sprintf "%0.2d",&alphavalue($1); + }else{ + s/^.//; + } + } + return $res; +} + +# test wether 2 pages are subsequent + +sub a_follows_b +{ + my ($i,$j)=@PageOrder{@_}; + $i++; + return($i eq $j); +} + +# make a sorted, maybe compressed, list of pages + +sub compressed_pages +{ + my ($A,$B)=@_; + my %pages=%{$A}; + my %pagerep=%{$B}; + my ($prevpage,$pendrep,$res,$pagepending,$page)=("","","","",""); + for $page (sort { $PageOrder{$a} cmp $PageOrder{$b} } keys %pages){ + # handle compression of page ranges. At the moment, we + # also compress ranges that might be displayed in different faces + my $pagerep=(exists $pagerep{$page}) ? $pagerep{$page} : $page; + if($prevpage){ + if(!$nocompress and &a_follows_b($prevpage,$page)){ + $pendrep=$pagepending ? "--$pagerep" + : ($twoabbrev ? "$twoabbrev" : ", $pagerep"); + $pagepending=1; + }else{ + $res.=($pagepending ? "$pendrep" : "").", $pagerep"; + $pagepending=0; + } + }else{ + $res.="$pagerep"; + } + $prevpage=$page; + } + $res.="$pendrep" if $pagepending; + return $res; +} + +sub pages_for_label +{ + my ($label)=@_; + if(exists $Lab2Pag{$label}){ + my (%pages,%empty); + @pages{@{$Lab2Pag{$label}}}=""; + return &compressed_pages(\%pages,\%empty); + } + return ""; +} |