## ## Utility functions for Man->HTML converter. Note that these ## functions are specifically intended to work with the format as used ## by Tcl and Tk; they do not cope with arbitrary nroff markup. ## ## Copyright (c) 1995-1997 Roger E. Critchlow Jr ## Copyright (c) 2004-2010 Donal K. Fellows set ::manual(report-level) 1 proc manerror {msg} { global manual set name {} set subj {} set procname [lindex [info level -1] 0] if {[info exists manual(name)]} { set name $manual(name) } if {[info exists manual(section)] && [string length $manual(section)]} { puts stderr "$name: $manual(section): $procname: $msg" } else { puts stderr "$name: $procname: $msg" } } proc manreport {level msg} { global manual if {$level < $manual(report-level)} { uplevel 1 [list manerror $msg] } } proc fatal {msg} { global manual uplevel 1 [list manerror $msg] exit 1 } ## ## templating ## proc indexfile {} { if {[info exists ::TARGET] && $::TARGET eq "devsite"} { return "index.tml" } else { return "contents.htm" } } proc copyright {copyright {level {}}} { # We don't actually generate a separate copyright page anymore #set page "${level}copyright.htm" #return "Copyright © [htmlize-text [lrange $copyright 2 end]]" # obfuscate any email addresses that may appear in name set who [string map {@ (at)} [lrange $copyright 2 end]] return "Copyright © [htmlize-text $who]" } proc copyout {copyrights {level {}}} { set out "
" foreach c $copyrights { append out "[copyright $c $level]\n" } append out "
" return $out } proc CSS {{level ""}} { return "\n" } proc DOCTYPE {} { return "" } proc htmlhead {title header args} { set level "" if {[lindex $args end] eq "../[indexfile]"} { # XXX hack - assume same level for CSS file set level "../" } set out "[DOCTYPE]\n\n$title\n[CSS $level]\n" foreach {uptitle url} $args { set header "$uptitle > $header" } append out "

$header

" global manual if {[info exists manual(subheader)]} { set subs {} foreach {name subdir} $manual(subheader) { if {$name eq $title} { lappend subs $name } else { lappend subs "$name" } } append out "\n

[join $subs { | }]

" } return $out } ## ## parsing ## proc unquote arg { return [string map [list \" {}] $arg] } proc parse-directive {line codename restname} { upvar 1 $codename code $restname rest return [regexp {^(\.[.a-zA-Z0-9]*) *(.*)} $line all code rest] } proc htmlize-text {text {charmap {}}} { # contains some extras for use in nroff->html processing # build on the list passed in, if any lappend charmap \ {&} {&} \ {\\} "\" \ {\e} "\" \ {\ } { } \ {\|} { } \ {\0} { } \ \" {"} \ {<} {<} \ {>} {>} \ \u201c "“" \ \u201d "”" return [string map $charmap $text] } proc process-text {text} { global manual # preprocess text; note that this is an incomplete map, and will probably # need to have things added to it as the manuals expand to use them. set charmap [list \ {\&} "\t" \ {\%} {} \ "\\\n" "\n" \ {\(+-} "±" \ {\(co} "©" \ {\(em} "—" \ {\(fm} "′" \ {\(mu} "×" \ {\(mi} "−" \ {\(->} "" \ {\fP} {\fR} \ {\.} . \ {\(bu} "•" \ ] lappend charmap {\o'o^'} {ô} ; # o-circumflex in re_syntax.n lappend charmap {\-\|\-} -- ; # two hyphens lappend charmap {\-} - ; # a hyphen set text [htmlize-text $text $charmap] # General quoted entity regsub -all {\\N'(\d+)'} $text "\\&#\\1;" text while {[string first "\\" $text] >= 0} { # C R if {[regsub {^([^\\]*)\\fC([^\\]*)\\fR(.*)$} $text \ {\1\2\3} text]} continue # B R if {[regsub {^([^\\]*)\\fB([^\\]*)\\fR(.*)$} $text \ {\1\2\3} text]} continue # B I if {[regsub {^([^\\]*)\\fB([^\\]*)\\fI(.*)$} $text \ {\1\2\\fI\3} text]} continue # I R if {[regsub {^([^\\]*)\\fI([^\\]*)\\fR(.*)$} $text \ {\1\2\3} text]} continue # I B if {[regsub {^([^\\]*)\\fI([^\\]*)\\fB(.*)$} $text \ {\1\2\\fB\3} text]} continue # B B, I I, R R if { [regsub {^([^\\]*)\\fB([^\\]*)\\fB(.*)$} $text \ {\1\\fB\2\3} ntext] || [regsub {^([^\\]*)\\fI([^\\]*)\\fI(.*)$} $text \ {\1\\fI\2\3} ntext] || [regsub {^([^\\]*)\\fR([^\\]*)\\fR(.*)$} $text \ {\1\\fR\2\3} ntext] } { manerror "impotent font change: $text" set text $ntext continue } # unrecognized manerror "uncaught backslash: $text" set text [string map [list "\\" "\"] $text] } return $text } ## ## pass 2 text input and matching ## proc open-text {} { global manual set manual(text-length) [llength $manual(text)] set manual(text-pointer) 0 } proc more-text {} { global manual return [expr {$manual(text-pointer) < $manual(text-length)}] } proc next-text {} { global manual if {[more-text]} { set text [lindex $manual(text) $manual(text-pointer)] incr manual(text-pointer) return $text } manerror "read past end of text" error "fatal" } proc is-a-directive {line} { return [string match .* $line] } proc split-directive {line opname restname} { upvar 1 $opname op $restname rest set op [string range $line 0 2] set rest [string trim [string range $line 3 end]] } proc next-op-is {op restname} { global manual upvar 1 $restname rest if {[more-text]} { set text [lindex $manual(text) $manual(text-pointer)] if {[string equal -length 3 $text $op]} { set rest [string range $text 4 end] incr manual(text-pointer) return 1 } } return 0 } proc backup-text {n} { global manual if {$manual(text-pointer)-$n >= 0} { incr manual(text-pointer) -$n } } proc match-text args { global manual set nargs [llength $args] if {$manual(text-pointer) + $nargs > $manual(text-length)} { return 0 } set nback 0 foreach arg $args { if {![more-text]} { backup-text $nback return 0 } set arg [string trim $arg] set targ [string trim [lindex $manual(text) $manual(text-pointer)]] if {$arg eq $targ} { incr nback incr manual(text-pointer) continue } if {[regexp {^@(\w+)$} $arg all name]} { upvar 1 $name var set var $targ incr nback incr manual(text-pointer) continue } if {[regexp -nocase {^(\.[A-Z][A-Z])@(\w+)$} $arg all op name]\ && [string equal $op [lindex $targ 0]]} { upvar 1 $name var set var [lrange $targ 1 end] incr nback incr manual(text-pointer) continue } backup-text $nback return 0 } return 1 } proc expand-next-text {n} { global manual return [join [lrange $manual(text) $manual(text-pointer) \ [expr {$manual(text-pointer)+$n-1}]] \n\n] } ## ## pass 2 output ## proc man-puts {text} { global manual lappend manual(output-$manual(wing-file)-$manual(name)) $text } ## ## build hypertext links to tables of contents ## proc long-toc {text} { global manual set here M[incr manual(section-toc-n)] set manual($manual(name)-id-$text) $here set there L[incr manual(long-toc-n)] lappend manual(section-toc) \ "
$text" return "$text" } proc option-toc {name class switch} { global manual # Special case handling, oh we hate it but must do it if {[string match "*OPTIONS" $manual(section)]} { if {$manual(name) ne "ttk_widget" && ($manual(name) ne "ttk_entry" || ![string match validate* $name])} { # link the defined option into the long table of contents set link [long-toc "$switch, $name, $class"] regsub -- "$switch, $name, $class" $link "$switch" link return $link } } elseif {"$manual(name):$manual(section)" ne "options:DESCRIPTION"} { error "option-toc in $manual(name) section $manual(section)" } # link the defined standard option to the long table of contents and make # a target for the standard option references from other man pages. set first [lindex $switch 0] set here M$first set there L[incr manual(long-toc-n)] set manual(standard-option-$manual(name)-$first) \ "$switch, $name, $class" lappend manual(section-toc) \ "
$switch, $name, $class" return "$switch" } proc std-option-toc {name page} { global manual if {[info exists manual(standard-option-$page-$name)]} { lappend manual(section-toc)
$manual(standard-option-$page-$name) return $manual(standard-option-$page-$name) } manerror "missing reference to \"$name\" in $page.n" set here M[incr manual(section-toc-n)] set there L[incr manual(long-toc-n)] set other M$name lappend manual(section-toc) "
$name" return "$name" } ## ## process the widget option section ## in widget and options man pages ## proc output-widget-options {rest} { global manual man-puts
lappend manual(section-toc)
backup-text 1 set para {} while {[next-op-is .OP rest]} { switch -exact -- [llength $rest] { 3 { lassign $rest switch name class } 5 { set switch [lrange $rest 0 2] set name [lindex $rest 3] set class [lindex $rest 4] } default { fatal "bad .OP $rest" } } if {![regexp {^(<.>)([-\w ]+)()$} $switch \ all oswitch switch cswitch]} { if {![regexp {^(<.>)([-\w ]+) or ([-\w ]+)()$} $switch \ all oswitch switch1 switch2 cswitch]} { error "not Switch: $switch" } set switch "$switch1$cswitch or $oswitch$switch2" } if {![regexp {^(<.>)([\w]*)()$} $name all oname name cname]} { error "not Name: $name" } if {![regexp {^(<.>)([\w]*)()$} $class all oclass class cclass]} { error "not Class: $class" } man-puts "$para
Command-Line Name: $oswitch[option-toc $name $class $switch]$cswitch" man-puts "
Database Name: $oname$name$cname" man-puts "
Database Class: $oclass$class$cclass" man-puts
[next-text] set para

if {[next-op-is .RS rest]} { while {[more-text]} { set line [next-text] if {[is-a-directive $line]} { split-directive $line code rest switch -exact -- $code { .RE { break } .SH - .SS { manerror "unbalanced .RS at section end" backup-text 1 break } default { output-directive $line } } } else { man-puts $line } } } } man-puts

lappend manual(section-toc)
} ## ## process .RS lists ## proc output-RS-list {} { global manual if {[next-op-is .IP rest]} { output-IP-list .RS .IP $rest if {[match-text .RE .sp .RS @rest .IP @rest2]} { man-puts

$rest output-IP-list .RS .IP $rest2 } if {[match-text .RE .sp .RS @rest .RE]} { man-puts

$rest return } if {[next-op-is .RE rest]} { return } } man-puts

while {[more-text]} { set line [next-text] if {[is-a-directive $line]} { split-directive $line code rest switch -exact -- $code { .RE { break } .SH - .SS { manerror "unbalanced .RS at section end" backup-text 1 break } default { output-directive $line } } } else { man-puts $line } } man-puts
} ## ## process .IP lists which may be plain indents, ## numeric lists, or definition lists ## proc output-IP-list {context code rest} { global manual if {![string length $rest]} { # blank label, plain indent, no contents entry man-puts
while {[more-text]} { set line [next-text] if {[is-a-directive $line]} { split-directive $line code rest if {$code eq ".IP" && $rest eq {}} { man-puts "

" continue } if {$code in {.br .DS .RS}} { output-directive $line } else { backup-text 1 break } } else { man-puts $line } } man-puts

} else { # labelled list, make contents if {$context ne ".SH" && $context ne ".SS"} { man-puts

} set dl "

" man-puts $dl lappend manual(section-toc) $dl backup-text 1 set accept_RE 0 set para {} while {[more-text]} { set line [next-text] if {[is-a-directive $line]} { split-directive $line code rest switch -exact -- $code { .IP { if {$accept_RE} { output-IP-list .IP $code $rest continue } if {$manual(section) eq "ARGUMENTS" || \ [regexp {^\[\d+\]$} $rest]} { man-puts "$para
$rest
" } elseif {"•" eq $rest} { man-puts "$para
$rest " } else { man-puts "$para
[long-toc $rest]
" } } .sp - .br - .DS - .CS { output-directive $line } .RS { if {[match-text .RS]} { output-directive $line incr accept_RE 1 } elseif {[match-text .CS]} { output-directive .CS incr accept_RE 1 } elseif {[match-text .PP]} { output-directive .PP incr accept_RE 1 } elseif {[match-text .DS]} { output-directive .DS incr accept_RE 1 } else { output-directive $line } } .PP { if {[match-text @rest1 .br @rest2 .RS]} { # yet another nroff kludge as above man-puts "$para
[long-toc $rest1]" man-puts "
[long-toc $rest2]
" incr accept_RE 1 } elseif {[match-text @rest .RE]} { # gad, this is getting ridiculous if {!$accept_RE} { man-puts "

$rest

" backup-text 1 set para {} break } else { man-puts "

$rest" incr accept_RE -1 } } elseif {$accept_RE} { output-directive $line } else { backup-text 1 break } } .RE { if {!$accept_RE} { backup-text 1 break } incr accept_RE -1 } default { backup-text 1 break } } } else { man-puts $line } set para

} man-puts "$para

" lappend manual(section-toc) if {$accept_RE} { manerror "missing .RE in output-IP-list" } } } ## ## handle the NAME section lines ## there's only one line in the NAME section, ## consisting of a comma separated list of names, ## followed by a hyphen and a short description. ## proc output-name {line} { global manual # split name line into pieces regexp {^([^-]+) - (.*)$} [regsub -all {[ \n\r\t]+} $line " "] -> head tail # output line to manual page untouched man-puts "$head — $tail" # output line to long table of contents lappend manual(section-toc) "
$head — $tail
" # separate out the names for future reference foreach name [split $head ,] { set name [string trim $name] if {[llength $name] > 1} { manerror "name has a space: {$name}\nfrom: $line" } lappend manual(wing-toc) $name lappend manual(name-$name) $manual(wing-file)/$manual(name) } } ## ## build a cross-reference link if appropriate ## proc cross-reference {ref} { global manual remap_link_target global ensemble_commands exclude_refs_map exclude_when_followed_by_map set lref [string tolower $ref] if {[string match "Tcl_*" $ref] || [string match "Tk_*" $ref]} { set lref $ref } elseif {$ref eq "Tcl"} { set lref $ref } elseif { [regexp {^[A-Z0-9 ?!]+$} $ref] && [info exists manual($manual(name)-id-$ref)] } { return "$ref" } ## ## apply a link remapping if available ## if {[info exists remap_link_target($lref)]} { set lref $remap_link_target($lref) } ## ## nothing to reference ## if {![info exists manual(name-$lref)]} { foreach name $ensemble_commands { if {[regexp "^$name \[a-z0-9]*\$" $lref] && \ [info exists manual(name-$name)] && \ $manual(tail) ne "$name.n"} { return "$ref" } } if {$lref in {end}} { # no good place to send this tcl token? } return $ref } ## ## would be a self reference ## foreach name $manual(name-$lref) { if {"$manual(wing-file)/$manual(name)" in $name} { return $ref } } ## ## multiple choices for reference ## if {[llength $manual(name-$lref)] > 1} { set tcl_i [lsearch -glob $manual(name-$lref) *TclCmd*] set tcl_ref [lindex $manual(name-$lref) $tcl_i] set tk_i [lsearch -glob $manual(name-$lref) *TkCmd*] set tk_ref [lindex $manual(name-$lref) $tk_i] if {$tcl_i >= 0 && $manual(wing-file) eq "TclCmd" || $manual(wing-file) eq "TclLib"} { return "$ref" } if {$tk_i >= 0 && $manual(wing-file) eq "TkCmd" || $manual(wing-file) eq "TkLib"} { return "$ref" } if {$lref eq "exit" && $manual(tail) eq "tclsh.1" && $tcl_i >= 0} { return "$ref" } puts stderr "multiple cross reference to $ref in $manual(name-$lref) from $manual(wing-file)/$manual(tail)" return $ref } ## ## exceptions, sigh, to the rule ## if {[info exists exclude_when_followed_by_map($manual(tail))]} { upvar 1 tail tail set following_word [lindex [regexp -inline {\S+} $tail] 0] foreach {this that} $exclude_when_followed_by_map($manual(tail)) { # only a ref if $this is not followed by $that if {$lref eq $this && [string match $that* $following_word]} { return $ref } } } if { [info exists exclude_refs_map($manual(tail))] && $lref in $exclude_refs_map($manual(tail)) } { return $ref } ## ## return the cross reference ## return "$ref" } ## ## reference generation errors ## proc reference-error {msg text} { global manual puts stderr "$manual(tail): $msg: {$text}" return $text } ## ## insert as many cross references into this text string as are appropriate ## proc insert-cross-references {text} { global manual ## ## we identify cross references by: ## ``quotation'' ## emboldening ## Tcl_ prefix ## Tk_ prefix ## [a-zA-Z0-9]+ manual entry ## and we avoid messing with already anchored text ## ## ## find where each item lives ## array set offset [list \ anchor [string first {} $text] \ quote [string first {``} $text] \ end-quote [string first {''} $text] \ bold [string first {} $text] \ end-bold [string first {} $text] \ tcl [string first {Tcl_} $text] \ tk [string first {Tk_} $text] \ Tcl1 [string first {Tcl manual entry} $text] \ Tcl2 [string first {Tcl overview manual entry} $text] \ ] ## ## accumulate a list ## foreach name [array names offset] { if {$offset($name) >= 0} { set invert($offset($name)) $name lappend offsets $offset($name) } } ## ## if nothing, then we're done. ## if {![info exists offsets]} { return $text } ## ## sort the offsets ## set offsets [lsort -integer $offsets] ## ## see which we want to use ## switch -exact -- $invert([lindex $offsets 0]) { anchor { if {$offset(end-anchor) < 0} { return [reference-error {Missing end anchor} $text] } set head [string range $text 0 $offset(end-anchor)] set tail [string range $text [expr {$offset(end-anchor)+1}] end] return $head[insert-cross-references $tail] } quote { if {$offset(end-quote) < 0} { return [reference-error "Missing end quote" $text] } if {$invert([lindex $offsets 1]) eq "tk"} { set offsets [lreplace $offsets 1 1] } if {$invert([lindex $offsets 1]) eq "tcl"} { set offsets [lreplace $offsets 1 1] } switch -exact -- $invert([lindex $offsets 1]) { end-quote { set head [string range $text 0 [expr {$offset(quote)-1}]] set body [string range $text [expr {$offset(quote)+2}] \ [expr {$offset(end-quote)-1}]] set tail [string range $text \ [expr {$offset(end-quote)+2}] end] return "$head``[cross-reference $body]''[insert-cross-references $tail]" } bold - anchor { set head [string range $text \ 0 [expr {$offset(end-quote)+1}]] set tail [string range $text \ [expr {$offset(end-quote)+2}] end] return "$head[insert-cross-references $tail]" } } return [reference-error "Uncaught quote case" $text] } bold { if {$offset(end-bold) < 0} { return $text } if {$invert([lindex $offsets 1]) eq "tk"} { set offsets [lreplace $offsets 1 1] } if {$invert([lindex $offsets 1]) eq "tcl"} { set offsets [lreplace $offsets 1 1] } switch -exact -- $invert([lindex $offsets 1]) { end-bold { set head [string range $text 0 [expr {$offset(bold)-1}]] set body [string range $text [expr {$offset(bold)+3}] \ [expr {$offset(end-bold)-1}]] set tail [string range $text \ [expr {$offset(end-bold)+4}] end] return "$head[cross-reference $body][insert-cross-references $tail]" } anchor { set head [string range $text \ 0 [expr {$offset(end-bold)+3}]] set tail [string range $text \ [expr {$offset(end-bold)+4}] end] return "$head[insert-cross-references $tail]" } } return [reference-error "Uncaught bold case" $text] } tk { set head [string range $text 0 [expr {$offset(tk)-1}]] set tail [string range $text $offset(tk) end] if {![regexp {^(Tk_\w+)(.*)$} $tail all body tail]} { return [reference-error "Tk regexp failed" $text] } return $head[cross-reference $body][insert-cross-references $tail] } tcl { set head [string range $text 0 [expr {$offset(tcl)-1}]] set tail [string range $text $offset(tcl) end] if {![regexp {^(Tcl_\w+)(.*)$} $tail all body tail]} { return [reference-error {Tcl regexp failed} $text] } return $head[cross-reference $body][insert-cross-references $tail] } Tcl1 - Tcl2 { set off [lindex $offsets 0] set head [string range $text 0 [expr {$off-1}]] set body Tcl set tail [string range $text [expr {$off+3}] end] return $head[cross-reference $body][insert-cross-references $tail] } end-anchor - end-bold - end-quote { return [reference-error "Out of place $invert([lindex $offsets 0])" $text] } } } ## ## process formatting directives ## proc output-directive {line} { global manual # process format directive split-directive $line code rest switch -exact -- $code { .BS - .BE { # man-puts
} .SH - .SS { # drain any open lists # announce the subject set manual(section) $rest # start our own stack of stuff set manual($manual(name)-$manual(section)) {} lappend manual(has-$manual(section)) $manual(name) if {$code ne ".SS"} { man-puts "

[long-toc $manual(section)]

" } else { man-puts "

[long-toc $manual(section)]

" } # some sections can simply free wheel their way through the text # some sections can be processed in their own loops switch -exact -- [string index $code end]:$manual(section) { H:NAME { set names {} while {1} { set line [next-text] if {[is-a-directive $line]} { backup-text 1 output-name [join $names { }] return } lappend names [string trim $line] } } H:SYNOPSIS { lappend manual(section-toc)
while {1} { if { [next-op-is .nf rest] || [next-op-is .br rest] || [next-op-is .fi rest] } { continue } if { [next-op-is .SH rest] || [next-op-is .SS rest] || [next-op-is .BE rest] || [next-op-is .SO rest] } { backup-text 1 break } if {[next-op-is .sp rest]} { #man-puts

continue } set more [next-text] if {[is-a-directive $more]} { manerror "in SYNOPSIS found $more" backup-text 1 break } foreach more [split $more \n] { regexp {^(\s*)(.*)} $more -> spaces more set spaces [string map {" " " "} $spaces] if {[string length $spaces]} { set spaces $spaces } man-puts $spaces$more
if {$manual(wing-file) in {TclLib TkLib}} { lappend manual(section-toc)

$more } } } lappend manual(section-toc)
return } {H:SEE ALSO} { while {[more-text]} { if {[next-op-is .SH rest] || [next-op-is .SS rest]} { backup-text 1 return } set more [next-text] if {[is-a-directive $more]} { manerror "$more" backup-text 1 return } set nmore {} foreach cr [split $more ,] { set cr [string trim $cr] if {![regexp {^.*$} $cr]} { set cr $cr } if {[regexp {^(.*)\([13n]\)$} $cr all name]} { set cr $name } lappend nmore $cr } man-puts [join $nmore {, }] } return } H:KEYWORDS { while {[more-text]} { if {[next-op-is .SH rest] || [next-op-is .SS rest]} { backup-text 1 return } set more [next-text] if {[is-a-directive $more]} { manerror "$more" backup-text 1 return } set keys {} foreach key [split $more ,] { set key [string trim $key] lappend manual(keyword-$key) [list $manual(name) \ $manual(wing-file)/$manual(name).htm] set initial [string toupper [string index $key 0]] lappend keys "
$key" } man-puts [join $keys {, }] } return } } if {[next-op-is .IP rest]} { output-IP-list $code .IP $rest return } if {[next-op-is .PP rest]} { return } return } .SO { # When there's a sequence of multiple .SO chunks, process into one set optslist {} while 1 { if {[match-text @stuff .SE]} { foreach opt [split $stuff \n\t] { lappend optslist [list $opt $rest] } } else { manerror "unexpected .SO format:\n[expand-next-text 2]" } if {![next-op-is .SO rest]} { break } } output-directive {.SH STANDARD OPTIONS} man-puts
lappend manual(section-toc)
foreach optionpair [lsort -dictionary -index 0 $optslist] { lassign $optionpair option targetPage man-puts "
[std-option-toc $option $targetPage]" } man-puts
lappend manual(section-toc)
} .OP { output-widget-options $rest return } .IP { output-IP-list .IP .IP $rest return } .PP { man-puts

} .RS { output-RS-list return } .RE { manerror "unexpected .RE" return } .br { man-puts
return } .DE { manerror "unexpected .DE" return } .DS { if {[next-op-is .ta rest]} { # skip the leading .ta directive if it is there } if {[match-text @stuff .DE]} { set td "

" set bodyText [string map [list \n $td \t $td] \n$stuff] man-puts "

$bodyText
" #man-puts
$stuff
} elseif {[match-text .fi @ul1 @ul2 .nf @stuff .DE]} { man-puts "
[lindex $ul1 1][lindex $ul2 1]\n$stuff
" } else { manerror "unexpected .DS format:\n[expand-next-text 2]" } return } .CS { if {[next-op-is .ta rest]} { # ??? } if {[match-text @stuff .CE]} { man-puts
$stuff
} else { manerror "unexpected .CS format:\n[expand-next-text 2]" } return } .CE { manerror "unexpected .CE" return } .sp { man-puts

} .ta { manerror "ignoring $line" } .nf { if {[match-text @more .fi]} { foreach more [split $more \n] { man-puts $more
} } elseif {[match-text .RS @more .RE .fi]} { man-puts

foreach more [split $more \n] { man-puts $more
} man-puts
} elseif {[match-text .RS @more .RS @more2 .RE .RE .fi]} { man-puts
foreach more [split $more \n] { man-puts $more
} man-puts
foreach more2 [split $more2 \n] { man-puts $more2
} man-puts
} elseif {[match-text .RS @more .RS @more2 .RE @more3 .RE .fi]} { man-puts
foreach more [split $more \n] { man-puts $more
} man-puts
foreach more2 [split $more2 \n] { man-puts $more2
} man-puts
foreach more3 [split $more3 \n] { man-puts $more3
} man-puts
} elseif {[match-text .sp .RS @more .RS @more2 .sp .RE .RE .fi]} { man-puts

foreach more [split $more \n] { man-puts $more
} man-puts
foreach more2 [split $more2 \n] { man-puts $more2
} man-puts

} elseif {[match-text .RS .sp @more .sp .RE .fi]} { man-puts

foreach more [split $more \n] { man-puts $more
} man-puts

} else { manerror "ignoring $line" } } .fi { manerror "ignoring $line" } .na - .ad - .UL - .ne { manerror "ignoring $line" } default { manerror "unrecognized format directive: $line" } } } ## ## merge copyright listings ## proc merge-copyrights {l1 l2} { set merge {} set re1 {^Copyright +(?:\(c\)|\\\(co|©) +(\w.*?)(?:all rights reserved)?(?:\. )*$} set re2 {^(\d+) +(?:by +)?(\w.*)$} ;# date who set re3 {^(\d+)-(\d+) +(?:by +)?(\w.*)$} ;# from to who set re4 {^(\d+), *(\d+) +(?:by +)?(\w.*)$} ;# date1 date2 who foreach copyright [concat $l1 $l2] { if {[regexp -nocase -- $re1 $copyright -> info]} { set info [string trimright $info ". "] ; # remove extra period if {[regexp -- $re2 $info -> date who]} { lappend dates($who) $date continue } elseif {[regexp -- $re3 $info -> from to who]} { for {set date $from} {$date <= $to} {incr date} { lappend dates($who) $date } continue } elseif {[regexp -- $re3 $info -> date1 date2 who]} { lappend dates($who) $date1 $date2 continue } } puts "oops: $copyright" } foreach who [array names dates] { set list [lsort -dictionary $dates($who)] if {[llength $list] == 1 || [lindex $list 0] eq [lrange $list end end]} { lappend merge "Copyright © [lindex $list 0] $who" } else { lappend merge "Copyright © [lindex $list 0]-[lrange $list end end] $who" } } return [lsort -dictionary $merge] } proc makedirhier {dir} { try { if {![file isdirectory $dir]} { file mkdir $dir } } on error msg { return -code error "cannot create directory $dir: $msg" } } proc addbuffer {args} { global manual if {$manual(partial-text) ne ""} { append manual(partial-text) \n } append manual(partial-text) [join $args ""] } proc flushbuffer {} { global manual if {$manual(partial-text) ne ""} { lappend manual(text) [process-text $manual(partial-text)] set manual(partial-text) "" } } return