diff --git a/Makefile.org b/Makefile.org index 8bb7e01773a26305f9e540b3d83b2daeceae9d66..7c802e8dfe7b408e2ca95b20b9f146e121d22f1b 100644 --- a/Makefile.org +++ b/Makefile.org @@ -716,7 +716,6 @@ install_docs: $(INSTALL_PREFIX)$(MANDIR)/man3 \ $(INSTALL_PREFIX)$(MANDIR)/man5 \ $(INSTALL_PREFIX)$(MANDIR)/man7 - @pod2man="`cd ./util; ./pod2mantest $(PERL)`"; \ here="`pwd`"; \ filecase=; \ case "$(PLATFORM)" in DJGPP|Cygwin*|mingw*|darwin*-*-cc) \ @@ -727,9 +726,9 @@ install_docs: sec=`$(PERL) util/extract-section.pl 1 < $$i`; \ echo "installing man$$sec/$$fn.$${sec}$(MANSUFFIX)"; \ (cd `$(PERL) util/dirname.pl $$i`; \ - sh -c "$$pod2man \ + pod2man \ --section=$$sec --center=OpenSSL \ - --release=$(VERSION) `basename $$i`") \ + --release=$(VERSION) `basename $$i`) \ > $(INSTALL_PREFIX)$(MANDIR)/man$$sec/$$fn.$${sec}$(MANSUFFIX); \ $(PERL) util/extract-names.pl < $$i | \ (grep -v $$filecase "^$$fn\$$"; true) | \ @@ -744,9 +743,9 @@ install_docs: sec=`$(PERL) util/extract-section.pl 3 < $$i`; \ echo "installing man$$sec/$$fn.$${sec}$(MANSUFFIX)"; \ (cd `$(PERL) util/dirname.pl $$i`; \ - sh -c "$$pod2man \ + pod2man \ --section=$$sec --center=OpenSSL \ - --release=$(VERSION) `basename $$i`") \ + --release=$(VERSION) `basename $$i`) \ > $(INSTALL_PREFIX)$(MANDIR)/man$$sec/$$fn.$${sec}$(MANSUFFIX); \ $(PERL) util/extract-names.pl < $$i | \ (grep -v $$filecase "^$$fn\$$"; true) | \ diff --git a/util/pod2man.pl b/util/pod2man.pl deleted file mode 100755 index 025d914f2e8be2f8deef2f3b1abd9166056be811..0000000000000000000000000000000000000000 --- a/util/pod2man.pl +++ /dev/null @@ -1,1184 +0,0 @@ -: #!/usr/bin/perl-5.005 - eval 'exec /usr/bin/perl -S $0 ${1+"$@"}' - if $running_under_some_shell; - -$DEF_PM_SECTION = '3pm' || '3'; - -=head1 NAME - -pod2man - translate embedded Perl pod directives into man pages - -=head1 SYNOPSIS - -B -[ B<--section=>I ] -[ B<--release=>I ] -[ B<--center=>I ] -[ B<--date=>I ] -[ B<--fixed=>I ] -[ B<--official> ] -[ B<--lax> ] -I - -=head1 DESCRIPTION - -B converts its input file containing embedded pod directives (see -L) into nroff source suitable for viewing with nroff(1) or -troff(1) using the man(7) macro set. - -Besides the obvious pod conversions, B also takes care of -func(), func(n), and simple variable references like $foo or @bar so -you don't have to use code escapes for them; complex expressions like -C<$fred{'stuff'}> will still need to be escaped, though. Other nagging -little roffish things that it catches include translating the minus in -something like foo-bar, making a long dash--like this--into a real em -dash, fixing up "paired quotes", putting a little space after the -parens in something like func(), making C++ and PI look right, making -double underbars have a little tiny space between them, making ALLCAPS -a teeny bit smaller in troff(1), and escaping backslashes so you don't -have to. - -=head1 OPTIONS - -=over 8 - -=item center - -Set the centered header to a specific string. The default is -"User Contributed Perl Documentation", unless the C<--official> flag is -given, in which case the default is "Perl Programmers Reference Guide". - -=item date - -Set the left-hand footer string to this value. By default, -the modification date of the input file will be used. - -=item fixed - -The fixed font to use for code refs. Defaults to CW. - -=item official - -Set the default header to indicate that this page is of -the standard release in case C<--center> is not given. - -=item release - -Set the centered footer. By default, this is the current -perl release. - -=item section - -Set the section for the C<.TH> macro. The standard conventions on -sections are to use 1 for user commands, 2 for system calls, 3 for -functions, 4 for devices, 5 for file formats, 6 for games, 7 for -miscellaneous information, and 8 for administrator commands. This works -best if you put your Perl man pages in a separate tree, like -F. By default, section 1 will be used -unless the file ends in F<.pm> in which case section 3 will be selected. - -=item lax - -Don't complain when required sections aren't present. - -=back - -=head1 Anatomy of a Proper Man Page - -For those not sure of the proper layout of a man page, here's -an example of the skeleton of a proper man page. Head of the -major headers should be setout as a C<=head1> directive, and -are historically written in the rather startling ALL UPPER CASE -format, although this is not mandatory. -Minor headers may be included using C<=head2>, and are -typically in mixed case. - -=over 10 - -=item NAME - -Mandatory section; should be a comma-separated list of programs or -functions documented by this podpage, such as: - - foo, bar - programs to do something - -=item SYNOPSIS - -A short usage summary for programs and functions, which -may someday be deemed mandatory. - -=item DESCRIPTION - -Long drawn out discussion of the program. It's a good idea to break this -up into subsections using the C<=head2> directives, like - - =head2 A Sample Subection - - =head2 Yet Another Sample Subection - -=item OPTIONS - -Some people make this separate from the description. - -=item RETURN VALUE - -What the program or function returns if successful. - -=item ERRORS - -Exceptions, return codes, exit stati, and errno settings. - -=item EXAMPLES - -Give some example uses of the program. - -=item ENVIRONMENT - -Envariables this program might care about. - -=item FILES - -All files used by the program. You should probably use the FEE -for these. - -=item SEE ALSO - -Other man pages to check out, like man(1), man(7), makewhatis(8), or catman(8). - -=item NOTES - -Miscellaneous commentary. - -=item CAVEATS - -Things to take special care with; sometimes called WARNINGS. - -=item DIAGNOSTICS - -All possible messages the program can print out--and -what they mean. - -=item BUGS - -Things that are broken or just don't work quite right. - -=item RESTRICTIONS - -Bugs you don't plan to fix :-) - -=item AUTHOR - -Who wrote it (or AUTHORS if multiple). - -=item HISTORY - -Programs derived from other sources sometimes have this, or -you might keep a modification log here. - -=back - -=head1 EXAMPLES - - pod2man program > program.1 - pod2man some_module.pm > /usr/perl/man/man3/some_module.3 - pod2man --section=7 note.pod > note.7 - -=head1 DIAGNOSTICS - -The following diagnostics are generated by B. Items -marked "(W)" are non-fatal, whereas the "(F)" errors will cause -B to immediately exit with a non-zero status. - -=over 4 - -=item bad option in paragraph %d of %s: ``%s'' should be [%s]<%s> - -(W) If you start include an option, you should set it off -as bold, italic, or code. - -=item can't open %s: %s - -(F) The input file wasn't available for the given reason. - -=item Improper man page - no dash in NAME header in paragraph %d of %s - -(W) The NAME header did not have an isolated dash in it. This is -considered important. - -=item Invalid man page - no NAME line in %s - -(F) You did not include a NAME header, which is essential. - -=item roff font should be 1 or 2 chars, not `%s' (F) - -(F) The font specified with the C<--fixed> option was not -a one- or two-digit roff font. - -=item %s is missing required section: %s - -(W) Required sections include NAME, DESCRIPTION, and if you're -using a section starting with a 3, also a SYNOPSIS. Actually, -not having a NAME is a fatal. - -=item Unknown escape: %s in %s - -(W) An unknown HTML entity (probably for an 8-bit character) was given via -a CE> directive. Besides amp, lt, gt, and quot, recognized -entities are Aacute, aacute, Acirc, acirc, AElig, aelig, Agrave, agrave, -Aring, aring, Atilde, atilde, Auml, auml, Ccedil, ccedil, Eacute, eacute, -Ecirc, ecirc, Egrave, egrave, ETH, eth, Euml, euml, Iacute, iacute, Icirc, -icirc, Igrave, igrave, Iuml, iuml, Ntilde, ntilde, Oacute, oacute, Ocirc, -ocirc, Ograve, ograve, Oslash, oslash, Otilde, otilde, Ouml, ouml, szlig, -THORN, thorn, Uacute, uacute, Ucirc, ucirc, Ugrave, ugrave, Uuml, uuml, -Yacute, yacute, and yuml. - -=item Unmatched =back - -(W) You have a C<=back> without a corresponding C<=over>. - -=item Unrecognized pod directive: %s - -(W) You specified a pod directive that isn't in the known list of -C<=head1>, C<=head2>, C<=item>, C<=over>, C<=back>, or C<=cut>. - - -=back - -=head1 NOTES - -If you would like to print out a lot of man page continuously, you -probably want to set the C and D registers to set contiguous page -numbering and even/odd paging, at least on some versions of man(7). -Settting the F register will get you some additional experimental -indexing: - - troff -man -rC1 -rD1 -rF1 perl.1 perldata.1 perlsyn.1 ... - -The indexing merely outputs messages via C<.tm> for each -major page, section, subsection, item, and any CE> -directives. - - -=head1 RESTRICTIONS - -None at this time. - -=head1 BUGS - -The =over and =back directives don't really work right. They -take absolute positions instead of offsets, don't nest well, and -making people count is suboptimal in any event. - -=head1 AUTHORS - -Original prototype by Larry Wall, but so massively hacked over by -Tom Christiansen such that Larry probably doesn't recognize it anymore. - -=cut - -$/ = ""; -$cutting = 1; -@Indices = (); - -# We try first to get the version number from a local binary, in case we're -# running an installed version of Perl to produce documentation from an -# uninstalled newer version's pod files. -if ($^O ne 'plan9' and $^O ne 'dos' and $^O ne 'os2' and $^O ne 'MSWin32') { - my $perl = (-x './perl' && -f './perl' ) ? - './perl' : - ((-x '../perl' && -f '../perl') ? - '../perl' : - ''); - ($version,$patch) = `$perl -e 'print $]'` =~ /^(\d\.\d{3})(\d{2})?/ if $perl; -} -# No luck; we'll just go with the running Perl's version -($version,$patch) = $] =~ /^(.{5})(\d{2})?/ unless $version; -$DEF_RELEASE = "perl $version"; -$DEF_RELEASE .= ", patch $patch" if $patch; - - -sub makedate { - my $secs = shift; - my ($sec,$min,$hour,$mday,$mon,$year,$wday,$yday,$isdst) = localtime($secs); - my $mname = (qw{Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec})[$mon]; - $year += 1900; - return "$mday/$mname/$year"; -} - -use Getopt::Long; - -$DEF_SECTION = 1; -$DEF_CENTER = "User Contributed Perl Documentation"; -$STD_CENTER = "Perl Programmers Reference Guide"; -$DEF_FIXED = 'CW'; -$DEF_LAX = 0; - -sub usage { - warn "$0: @_\n" if @_; - die <"; -$Filename = $name; -if ($section =~ /^1/) { - require File::Basename; - $name = uc File::Basename::basename($name); -} -$name =~ s/\.(pod|p[lm])$//i; - -# Lose everything up to the first of -# */lib/*perl* standard or site_perl module -# */*perl*/lib from -D prefix=/opt/perl -# */*perl*/ random module hierarchy -# which works. -$name =~ s-//+-/-g; -if ($name =~ s-^.*?/lib/[^/]*perl[^/]*/--i - or $name =~ s-^.*?/[^/]*perl[^/]*/lib/--i - or $name =~ s-^.*?/[^/]*perl[^/]*/--i) { - # Lose ^site(_perl)?/. - $name =~ s-^site(_perl)?/--; - # Lose ^arch/. (XXX should we use Config? Just for archname?) - $name =~ s~^(.*-$^O|$^O-.*)/~~o; - # Lose ^version/. - $name =~ s-^\d+\.\d+/--; -} - -# Translate Getopt/Long to Getopt::Long, etc. -$name =~ s(/)(::)g; - -if ($name ne 'something') { - FCHECK: { - open(F, "< $ARGV[0]") || die "can't open $ARGV[0]: $!"; - while () { - next unless /^=\b/; - if (/^=head1\s+NAME\s*$/) { # an /m would forgive mistakes - $_ = ; - unless (/\s*-+\s+/) { - $oops++; - warn "$0: Improper man page - no dash in NAME header in paragraph $. of $ARGV[0]\n" - } else { - my @n = split /\s+-+\s+/; - if (@n != 2) { - $oops++; - warn "$0: Improper man page - malformed NAME header in paragraph $. of $ARGV[0]\n" - } - else { - $n[0] =~ s/\n/ /g; - $n[1] =~ s/\n/ /g; - %namedesc = @n; - } - } - last FCHECK; - } - next if /^=cut\b/; # DB_File and Net::Ping have =cut before NAME - next if /^=pod\b/; # It is OK to have =pod before NAME - next if /^=(for|begin|end)\s+comment\b/; # It is OK to have =for =begin or =end comment before NAME - die "$0: Invalid man page - 1st pod line is not NAME in $ARGV[0]\n" unless $lax; - } - die "$0: Invalid man page - no documentation in $ARGV[0]\n" unless $lax; - } - close F; -} - -print <<"END"; -.rn '' }` -''' \$RCSfile\$\$Revision\$\$Date\$ -''' -''' \$Log\$ -''' -.de Sh -.br -.if t .Sp -.ne 5 -.PP -\\fB\\\\\$1\\fR -.PP -.. -.de Sp -.if t .sp .5v -.if n .sp -.. -.de Ip -.br -.ie \\\\n(.\$>=3 .ne \\\\\$3 -.el .ne 3 -.IP "\\\\\$1" \\\\\$2 -.. -.de Vb -.ft $CFont -.nf -.ne \\\\\$1 -.. -.de Ve -.ft R - -.fi -.. -''' -''' -''' Set up \\*(-- to give an unbreakable dash; -''' string Tr holds user defined translation string. -''' Bell System Logo is used as a dummy character. -''' -.tr \\(*W-|\\(bv\\*(Tr -.ie n \\{\\ -.ds -- \\(*W- -.ds PI pi -.if (\\n(.H=4u)&(1m=24u) .ds -- \\(*W\\h'-12u'\\(*W\\h'-12u'-\\" diablo 10 pitch -.if (\\n(.H=4u)&(1m=20u) .ds -- \\(*W\\h'-12u'\\(*W\\h'-8u'-\\" diablo 12 pitch -.ds L" "" -.ds R" "" -''' \\*(M", \\*(S", \\*(N" and \\*(T" are the equivalent of -''' \\*(L" and \\*(R", except that they are used on ".xx" lines, -''' such as .IP and .SH, which do another additional levels of -''' double-quote interpretation -.ds M" """ -.ds S" """ -.ds N" """"" -.ds T" """"" -.ds L' ' -.ds R' ' -.ds M' ' -.ds S' ' -.ds N' ' -.ds T' ' -'br\\} -.el\\{\\ -.ds -- \\(em\\| -.tr \\*(Tr -.ds L" `` -.ds R" '' -.ds M" `` -.ds S" '' -.ds N" `` -.ds T" '' -.ds L' ` -.ds R' ' -.ds M' ` -.ds S' ' -.ds N' ` -.ds T' ' -.ds PI \\(*p -'br\\} -END - -print <<'END'; -.\" If the F register is turned on, we'll generate -.\" index entries out stderr for the following things: -.\" TH Title -.\" SH Header -.\" Sh Subsection -.\" Ip Item -.\" X<> Xref (embedded -.\" Of course, you have to process the output yourself -.\" in some meaninful fashion. -.if \nF \{ -.de IX -.tm Index:\\$1\t\\n%\t"\\$2" -.. -.nr % 0 -.rr F -.\} -END - -print <<"END"; -.TH $name $section "$RP" "$date" "$center" -.UC -END - -push(@Indices, qq{.IX Title "$name $section"}); - -while (($name, $desc) = each %namedesc) { - for ($name, $desc) { s/^\s+//; s/\s+$//; } - push(@Indices, qq(.IX Name "$name - $desc"\n)); -} - -print <<'END'; -.if n .hy 0 -.if n .na -.ds C+ C\v'-.1v'\h'-1p'\s-2+\h'-1p'+\s0\v'.1v'\h'-1p' -.de CQ \" put $1 in typewriter font -END -print ".ft $CFont\n"; -print <<'END'; -'if n "\c -'if t \\&\\$1\c -'if n \\&\\$1\c -'if n \&" -\\&\\$2 \\$3 \\$4 \\$5 \\$6 \\$7 -'.ft R -.. -.\" @(#)ms.acc 1.5 88/02/08 SMI; from UCB 4.2 -. \" AM - accent mark definitions -.bd B 3 -. \" fudge factors for nroff and troff -.if n \{\ -. ds #H 0 -. ds #V .8m -. ds #F .3m -. ds #[ \f1 -. ds #] \fP -.\} -.if t \{\ -. ds #H ((1u-(\\\\n(.fu%2u))*.13m) -. ds #V .6m -. ds #F 0 -. ds #[ \& -. ds #] \& -.\} -. \" simple accents for nroff and troff -.if n \{\ -. ds ' \& -. ds ` \& -. ds ^ \& -. ds , \& -. ds ~ ~ -. ds ? ? -. ds ! ! -. ds / -. ds q -.\} -.if t \{\ -. ds ' \\k:\h'-(\\n(.wu*8/10-\*(#H)'\'\h"|\\n:u" -. ds ` \\k:\h'-(\\n(.wu*8/10-\*(#H)'\`\h'|\\n:u' -. ds ^ \\k:\h'-(\\n(.wu*10/11-\*(#H)'^\h'|\\n:u' -. ds , \\k:\h'-(\\n(.wu*8/10)',\h'|\\n:u' -. ds ~ \\k:\h'-(\\n(.wu-\*(#H-.1m)'~\h'|\\n:u' -. ds ? \s-2c\h'-\w'c'u*7/10'\u\h'\*(#H'\zi\d\s+2\h'\w'c'u*8/10' -. ds ! \s-2\(or\s+2\h'-\w'\(or'u'\v'-.8m'.\v'.8m' -. ds / \\k:\h'-(\\n(.wu*8/10-\*(#H)'\z\(sl\h'|\\n:u' -. ds q o\h'-\w'o'u*8/10'\s-4\v'.4m'\z\(*i\v'-.4m'\s+4\h'\w'o'u*8/10' -.\} -. \" troff and (daisy-wheel) nroff accents -.ds : \\k:\h'-(\\n(.wu*8/10-\*(#H+.1m+\*(#F)'\v'-\*(#V'\z.\h'.2m+\*(#F'.\h'|\\n:u'\v'\*(#V' -.ds 8 \h'\*(#H'\(*b\h'-\*(#H' -.ds v \\k:\h'-(\\n(.wu*9/10-\*(#H)'\v'-\*(#V'\*(#[\s-4v\s0\v'\*(#V'\h'|\\n:u'\*(#] -.ds _ \\k:\h'-(\\n(.wu*9/10-\*(#H+(\*(#F*2/3))'\v'-.4m'\z\(hy\v'.4m'\h'|\\n:u' -.ds . \\k:\h'-(\\n(.wu*8/10)'\v'\*(#V*4/10'\z.\v'-\*(#V*4/10'\h'|\\n:u' -.ds 3 \*(#[\v'.2m'\s-2\&3\s0\v'-.2m'\*(#] -.ds o \\k:\h'-(\\n(.wu+\w'\(de'u-\*(#H)/2u'\v'-.3n'\*(#[\z\(de\v'.3n'\h'|\\n:u'\*(#] -.ds d- \h'\*(#H'\(pd\h'-\w'~'u'\v'-.25m'\f2\(hy\fP\v'.25m'\h'-\*(#H' -.ds D- D\\k:\h'-\w'D'u'\v'-.11m'\z\(hy\v'.11m'\h'|\\n:u' -.ds th \*(#[\v'.3m'\s+1I\s-1\v'-.3m'\h'-(\w'I'u*2/3)'\s-1o\s+1\*(#] -.ds Th \*(#[\s+2I\s-2\h'-\w'I'u*3/5'\v'-.3m'o\v'.3m'\*(#] -.ds ae a\h'-(\w'a'u*4/10)'e -.ds Ae A\h'-(\w'A'u*4/10)'E -.ds oe o\h'-(\w'o'u*4/10)'e -.ds Oe O\h'-(\w'O'u*4/10)'E -. \" corrections for vroff -.if v .ds ~ \\k:\h'-(\\n(.wu*9/10-\*(#H)'\s-2\u~\d\s+2\h'|\\n:u' -.if v .ds ^ \\k:\h'-(\\n(.wu*10/11-\*(#H)'\v'-.4m'^\v'.4m'\h'|\\n:u' -. \" for low resolution devices (crt and lpr) -.if \n(.H>23 .if \n(.V>19 \ -\{\ -. ds : e -. ds 8 ss -. ds v \h'-1'\o'\(aa\(ga' -. ds _ \h'-1'^ -. ds . \h'-1'. -. ds 3 3 -. ds o a -. ds d- d\h'-1'\(ga -. ds D- D\h'-1'\(hy -. ds th \o'bp' -. ds Th \o'LP' -. ds ae ae -. ds Ae AE -. ds oe oe -. ds Oe OE -.\} -.rm #[ #] #H #V #F C -END - -$indent = 0; - -$begun = ""; - -# Unrolling [^A-Z>]|[A-Z](?!<) gives: // MRE pp 165. -my $nonest = '(?:[^A-Z>]*(?:[A-Z](?!<)[^A-Z>]*)*)'; - -while (<>) { - if ($cutting) { - next unless /^=/; - $cutting = 0; - } - if ($begun) { - if (/^=end\s+$begun/) { - $begun = ""; - } - elsif ($begun =~ /^(roff|man)$/) { - print STDOUT $_; - } - next; - } - chomp; - - # Translate verbatim paragraph - - if (/^\s/) { - @lines = split(/\n/); - for (@lines) { - 1 while s - {^( [^\t]* ) \t ( \t* ) } - { $1 . ' ' x (8 - (length($1)%8) + 8 * (length($2))) }ex; - s/\\/\\e/g; - s/\A/\\&/s; - } - $lines = @lines; - makespace() unless $verbatim++; - print ".Vb $lines\n"; - print join("\n", @lines), "\n"; - print ".Ve\n"; - $needspace = 0; - next; - } - - $verbatim = 0; - - if (/^=for\s+(\S+)\s*/s) { - if ($1 eq "man" or $1 eq "roff") { - print STDOUT $',"\n\n"; - } else { - # ignore unknown for - } - next; - } - elsif (/^=begin\s+(\S+)\s*/s) { - $begun = $1; - if ($1 eq "man" or $1 eq "roff") { - print STDOUT $'."\n\n"; - } - next; - } - - # check for things that'll hosed our noremap scheme; affects $_ - init_noremap(); - - if (!/^=item/) { - - # trofficate backslashes; must do it before what happens below - s/\\/noremap('\\e')/ge; - - # protect leading periods and quotes against *roff - # mistaking them for directives - s/^(?:[A-Z]<)?[.']/\\&$&/gm; - - # first hide the escapes in case we need to - # intuit something and get it wrong due to fmting - - 1 while s/([A-Z]<$nonest>)/noremap($1)/ge; - - # func() is a reference to a perl function - s{ - \b - ( - [:\w]+ \(\) - ) - } {I<$1>}gx; - - # func(n) is a reference to a perl function or a man page - s{ - ([:\w]+) - ( - \( [^\051]+ \) - ) - } {I<$1>\\|$2}gx; - - # convert simple variable references - s/(\s+)([\$\@%][\w:]+)(?!\()/${1}C<$2>/g; - - if (m{ ( - [\-\w]+ - \( - [^\051]*? - [\@\$,] - [^\051]*? - \) - ) - }x && $` !~ /([LCI]<[^<>]*|-)$/ && !/^=\w/) - { - warn "$0: bad option in paragraph $. of $ARGV: ``$1'' should be [LCI]<$1>\n"; - $oops++; - } - - while (/(-[a-zA-Z])\b/g && $` !~ /[\w\-]$/) { - warn "$0: bad option in paragraph $. of $ARGV: ``$1'' should be [CB]<$1>\n"; - $oops++; - } - - # put it back so we get the <> processed again; - clear_noremap(0); # 0 means leave the E's - - } else { - # trofficate backslashes - s/\\/noremap('\\e')/ge; - - } - - # need to hide E<> first; they're processed in clear_noremap - s/(E<[^<>]+>)/noremap($1)/ge; - - - $maxnest = 10; - while ($maxnest-- && /[A-Z]/font($1) . $2 . font('R')/eg; - - # files and filelike refs in italics - s/F<($nonest)>/I<$1>/g; - - # no break -- usually we want C<> for this - s/S<($nonest)>/nobreak($1)/eg; - - # LREF: a la HREF L - s:L<([^|>]+)\|[^>]+>:$1:g; - - # LREF: a manpage(3f) - s:L<([a-zA-Z][^\s\/]+)(\([^\)]+\))?>:the I<$1>$2 manpage:g; - - # LREF: an =item on another manpage - s{ - L< - ([^/]+) - / - ( - [:\w]+ - (\(\))? - ) - > - } {the C<$2> entry in the I<$1> manpage}gx; - - # LREF: an =item on this manpage - s{ - ((?: - L< - / - ( - [:\w]+ - (\(\))? - ) - > - (,?\s+(and\s+)?)? - )+) - } { internal_lrefs($1) }gex; - - # LREF: a =head2 (head1?), maybe on a manpage, maybe right here - # the "func" can disambiguate - s{ - L< - (?: - ([a-zA-Z]\S+?) / - )? - "?(.*?)"? - > - }{ - do { - $1 # if no $1, assume it means on this page. - ? "the section on I<$2> in the I<$1> manpage" - : "the section on I<$2>" - } - }gesx; # s in case it goes over multiple lines, so . matches \n - - s/Z<>/\\&/g; - - # comes last because not subject to reprocessing - s/C<($nonest)>/noremap("${CFont_embed}${1}\\fR")/eg; - } - - if (s/^=//) { - $needspace = 0; # Assume this. - - s/\n/ /g; - - ($Cmd, $_) = split(' ', $_, 2); - - $dotlevel = 1; - if ($Cmd eq 'head1') { - $dotlevel = 1; - } - elsif ($Cmd eq 'head2') { - $dotlevel = 1; - } - elsif ($Cmd eq 'item') { - $dotlevel = 2; - } - - if (defined $_) { - &escapes($dotlevel); - s/"/""/g; - } - - clear_noremap(1); - - if ($Cmd eq 'cut') { - $cutting = 1; - } - elsif ($Cmd eq 'head1') { - s/\s+$//; - delete $wanna_see{$_} if exists $wanna_see{$_}; - print qq{.SH "$_"\n}; - push(@Indices, qq{.IX Header "$_"\n}); - } - elsif ($Cmd eq 'head2') { - print qq{.Sh "$_"\n}; - push(@Indices, qq{.IX Subsection "$_"\n}); - } - elsif ($Cmd eq 'over') { - push(@indent,$indent); - $indent += ($_ + 0) || 5; - } - elsif ($Cmd eq 'back') { - $indent = pop(@indent); - warn "$0: Unmatched =back in paragraph $. of $ARGV\n" unless defined $indent; - $needspace = 1; - } - elsif ($Cmd eq 'item') { - s/^\*( |$)/\\(bu$1/g; - # if you know how to get ":s please do - s/\\\*\(L"([^"]+?)\\\*\(R"/'$1'/g; - s/\\\*\(L"([^"]+?)""/'$1'/g; - s/[^"]""([^"]+?)""[^"]/'$1'/g; - # here do something about the $" in perlvar? - print STDOUT qq{.Ip "$_" $indent\n}; - push(@Indices, qq{.IX Item "$_"\n}); - } - elsif ($Cmd eq 'pod') { - # this is just a comment - } - else { - warn "$0: Unrecognized pod directive in paragraph $. of $ARGV: $Cmd\n"; - } - } - else { - if ($needspace) { - &makespace; - } - &escapes(0); - clear_noremap(1); - print $_, "\n"; - $needspace = 1; - } -} - -print <<"END"; - -.rn }` '' -END - -if (%wanna_see && !$lax) { - @missing = keys %wanna_see; - warn "$0: $Filename is missing required section" - . (@missing > 1 && "s") - . ": @missing\n"; - $oops++; -} - -foreach (@Indices) { print "$_\n"; } - -exit; -#exit ($oops != 0); - -######################################################################### - -sub nobreak { - my $string = shift; - $string =~ s/ /\\ /g; - $string; -} - -sub escapes { - my $indot = shift; - - s/X<(.*?)>/mkindex($1)/ge; - - # translate the minus in foo-bar into foo\-bar for roff - s/([^0-9a-z-])-([^-])/$1\\-$2/g; - - # make -- into the string version \*(-- (defined above) - s/\b--\b/\\*(--/g; - s/"--([^"])/"\\*(--$1/g; # should be a better way - s/([^"])--"/$1\\*(--"/g; - - # fix up quotes; this is somewhat tricky - my $dotmacroL = 'L'; - my $dotmacroR = 'R'; - if ( $indot == 1 ) { - $dotmacroL = 'M'; - $dotmacroR = 'S'; - } - elsif ( $indot >= 2 ) { - $dotmacroL = 'N'; - $dotmacroR = 'T'; - } - if (!/""/) { - s/(^|\s)(['"])/noremap("$1\\*($dotmacroL$2")/ge; - s/(['"])($|[\-\s,;\\!?.])/noremap("\\*($dotmacroR$1$2")/ge; - } - - #s/(?!")(?:.)--(?!")(?:.)/\\*(--/g; - #s/(?:(?!")(?:.)--(?:"))|(?:(?:")--(?!")(?:.))/\\*(--/g; - - - # make sure that func() keeps a bit a space tween the parens - ### s/\b\(\)/\\|()/g; - ### s/\b\(\)/(\\|)/g; - - # make C++ into \*C+, which is a squinched version (defined above) - s/\bC\+\+/\\*(C+/g; - - # make double underbars have a little tiny space between them - s/__/_\\|_/g; - - # PI goes to \*(PI (defined above) - s/\bPI\b/noremap('\\*(PI')/ge; - - # make all caps a teeny bit smaller, but don't muck with embedded code literals - my $hidCFont = font('C'); - if ($Cmd !~ /^head1/) { # SH already makes smaller - # /g isn't enough; 1 while or we'll be off - -# 1 while s{ -# (?!$hidCFont)(..|^.|^) -# \b -# ( -# [A-Z][\/A-Z+:\-\d_$.]+ -# ) -# (s?) -# \b -# } {$1\\s-1$2\\s0}gmox; - - 1 while s{ - (?!$hidCFont)(..|^.|^) - ( - \b[A-Z]{2,}[\/A-Z+:\-\d_\$]*\b - ) - } { - $1 . noremap( '\\s-1' . $2 . '\\s0' ) - }egmox; - - } -} - -# make troff just be normal, but make small nroff get quoted -# decided to just put the quotes in the text; sigh; -sub ccvt { - local($_,$prev) = @_; - noremap(qq{.CQ "$_" \n\\&}); -} - -sub makespace { - if ($indent) { - print ".Sp\n"; - } - else { - print ".PP\n"; - } -} - -sub mkindex { - my ($entry) = @_; - my @entries = split m:\s*/\s*:, $entry; - push @Indices, ".IX Xref " . join ' ', map {qq("$_")} @entries; - return ''; -} - -sub font { - local($font) = shift; - return '\\f' . noremap($font); -} - -sub noremap { - local($thing_to_hide) = shift; - $thing_to_hide =~ tr/\000-\177/\200-\377/; - return $thing_to_hide; -} - -sub init_noremap { - # escape high bit characters in input stream - s/([\200-\377])/"E<".ord($1).">"/ge; -} - -sub clear_noremap { - my $ready_to_print = $_[0]; - - tr/\200-\377/\000-\177/; - - # trofficate backslashes - # s/(?!\\e)(?:..|^.|^)\\/\\e/g; - - # now for the E<>s, which have been hidden until now - # otherwise the interative \w<> processing would have - # been hosed by the E - s { - E< - ( - ( \d + ) - | ( [A-Za-z]+ ) - ) - > - } { - do { - defined $2 - ? chr($2) - : - exists $HTML_Escapes{$3} - ? do { $HTML_Escapes{$3} } - : do { - warn "$0: Unknown escape in paragraph $. of $ARGV: ``$&''\n"; - "E<$1>"; - } - } - }egx if $ready_to_print; -} - -sub internal_lrefs { - local($_) = shift; - local $trailing_and = s/and\s+$// ? "and " : ""; - - s{L]+)>}{$1}g; - my(@items) = split( /(?:,?\s+(?:and\s+)?)/ ); - my $retstr = "the "; - my $i; - for ($i = 0; $i <= $#items; $i++) { - $retstr .= "C<$items[$i]>"; - $retstr .= ", " if @items > 2 && $i != $#items; - $retstr .= " and " if $i+2 == @items; - } - - $retstr .= " entr" . ( @items > 1 ? "ies" : "y" ) - . " elsewhere in this document"; - # terminal space to avoid words running together (pattern used - # strips terminal spaces) - $retstr .= " " if length $trailing_and; - $retstr .= $trailing_and; - - return $retstr; - -} - -BEGIN { -%HTML_Escapes = ( - 'amp' => '&', # ampersand - 'lt' => '<', # left chevron, less-than - 'gt' => '>', # right chevron, greater-than - 'quot' => '"', # double quote - - "Aacute" => "A\\*'", # capital A, acute accent - "aacute" => "a\\*'", # small a, acute accent - "Acirc" => "A\\*^", # capital A, circumflex accent - "acirc" => "a\\*^", # small a, circumflex accent - "AElig" => '\*(AE', # capital AE diphthong (ligature) - "aelig" => '\*(ae', # small ae diphthong (ligature) - "Agrave" => "A\\*`", # capital A, grave accent - "agrave" => "A\\*`", # small a, grave accent - "Aring" => 'A\\*o', # capital A, ring - "aring" => 'a\\*o', # small a, ring - "Atilde" => 'A\\*~', # capital A, tilde - "atilde" => 'a\\*~', # small a, tilde - "Auml" => 'A\\*:', # capital A, dieresis or umlaut mark - "auml" => 'a\\*:', # small a, dieresis or umlaut mark - "Ccedil" => 'C\\*,', # capital C, cedilla - "ccedil" => 'c\\*,', # small c, cedilla - "Eacute" => "E\\*'", # capital E, acute accent - "eacute" => "e\\*'", # small e, acute accent - "Ecirc" => "E\\*^", # capital E, circumflex accent - "ecirc" => "e\\*^", # small e, circumflex accent - "Egrave" => "E\\*`", # capital E, grave accent - "egrave" => "e\\*`", # small e, grave accent - "ETH" => '\\*(D-', # capital Eth, Icelandic - "eth" => '\\*(d-', # small eth, Icelandic - "Euml" => "E\\*:", # capital E, dieresis or umlaut mark - "euml" => "e\\*:", # small e, dieresis or umlaut mark - "Iacute" => "I\\*'", # capital I, acute accent - "iacute" => "i\\*'", # small i, acute accent - "Icirc" => "I\\*^", # capital I, circumflex accent - "icirc" => "i\\*^", # small i, circumflex accent - "Igrave" => "I\\*`", # capital I, grave accent - "igrave" => "i\\*`", # small i, grave accent - "Iuml" => "I\\*:", # capital I, dieresis or umlaut mark - "iuml" => "i\\*:", # small i, dieresis or umlaut mark - "Ntilde" => 'N\*~', # capital N, tilde - "ntilde" => 'n\*~', # small n, tilde - "Oacute" => "O\\*'", # capital O, acute accent - "oacute" => "o\\*'", # small o, acute accent - "Ocirc" => "O\\*^", # capital O, circumflex accent - "ocirc" => "o\\*^", # small o, circumflex accent - "Ograve" => "O\\*`", # capital O, grave accent - "ograve" => "o\\*`", # small o, grave accent - "Oslash" => "O\\*/", # capital O, slash - "oslash" => "o\\*/", # small o, slash - "Otilde" => "O\\*~", # capital O, tilde - "otilde" => "o\\*~", # small o, tilde - "Ouml" => "O\\*:", # capital O, dieresis or umlaut mark - "ouml" => "o\\*:", # small o, dieresis or umlaut mark - "szlig" => '\*8', # small sharp s, German (sz ligature) - "THORN" => '\\*(Th', # capital THORN, Icelandic - "thorn" => '\\*(th',, # small thorn, Icelandic - "Uacute" => "U\\*'", # capital U, acute accent - "uacute" => "u\\*'", # small u, acute accent - "Ucirc" => "U\\*^", # capital U, circumflex accent - "ucirc" => "u\\*^", # small u, circumflex accent - "Ugrave" => "U\\*`", # capital U, grave accent - "ugrave" => "u\\*`", # small u, grave accent - "Uuml" => "U\\*:", # capital U, dieresis or umlaut mark - "uuml" => "u\\*:", # small u, dieresis or umlaut mark - "Yacute" => "Y\\*'", # capital Y, acute accent - "yacute" => "y\\*'", # small y, acute accent - "yuml" => "y\\*:", # small y, dieresis or umlaut mark -); -} - diff --git a/util/pod2mantest b/util/pod2mantest deleted file mode 100755 index 384e683df4cc7636fbf98251d421e62654cca08f..0000000000000000000000000000000000000000 --- a/util/pod2mantest +++ /dev/null @@ -1,58 +0,0 @@ -#!/bin/sh - -# This script is used by test/Makefile to check whether a sane 'pod2man' -# is installed. -# ('make install' should not try to run 'pod2man' if it does not exist or if -# it is a broken 'pod2man' version that is known to cause trouble. if we find -# the system 'pod2man' to be broken, we use our own copy instead) -# -# In any case, output an appropriate command line for running (or not -# running) pod2man. - - -IFS=: -if test "$OSTYPE" = "msdosdjgpp"; then IFS=";"; fi - -try_without_dir=true -# First we try "pod2man", then "$dir/pod2man" for each item in $PATH. -for dir in dummy${IFS}$PATH; do - if [ "$try_without_dir" = true ]; then - # first iteration - pod2man=pod2man - try_without_dir=false - else - # second and later iterations - pod2man="$dir/pod2man" - if [ ! -f "$pod2man" ]; then # '-x' is not available on Ultrix - pod2man='' - fi - fi - - if [ ! "$pod2man" = '' ]; then - failure=none - - if "$pod2man" --section=1 --center=OpenSSL --release=dev pod2mantest.pod | fgrep OpenSSL >/dev/null; then - : - else - failure=BasicTest - fi - - if [ "$failure" = none ]; then - if "$pod2man" --section=1 --center=OpenSSL --release=dev pod2mantest.pod | grep '^MARKER - ' >/dev/null; then - failure=MultilineTest - fi - fi - - - if [ "$failure" = none ]; then - echo "$pod2man" - exit 0 - fi - - echo "$pod2man does not work properly ('$failure' failed). Looking for another pod2man ..." >&2 - fi -done - -echo "No working pod2man found. Consider installing a new version." >&2 -echo "As a workaround, we'll use a bundled old copy of pod2man.pl." >&2 -echo "$1 ../../util/pod2man.pl" diff --git a/util/pod2mantest.pod b/util/pod2mantest.pod deleted file mode 100644 index 5d2539a17fb143e32021877ff018660030e6c85f..0000000000000000000000000000000000000000 --- a/util/pod2mantest.pod +++ /dev/null @@ -1,15 +0,0 @@ -=pod - -=head1 NAME - -foo, bar, -MARKER - test of multiline name section - -=head1 DESCRIPTION - -This is a test .pod file to see if we have a buggy pod2man or not. -If we have a buggy implementation, we will get a line matching the -regular expression "^ +MARKER - test of multiline name section *$" -at the end of the resulting document. - -=cut