747 lines
20 KiB
Perl
747 lines
20 KiB
Perl
#!/usr/bin/perl
|
|
# Page inlining and blogging.
|
|
package IkiWiki::Plugin::inline;
|
|
|
|
use warnings;
|
|
use strict;
|
|
use Encode;
|
|
use IkiWiki 3.00;
|
|
use URI;
|
|
|
|
my %knownfeeds;
|
|
my %page_numfeeds;
|
|
my @inline;
|
|
my $nested=0;
|
|
|
|
sub import {
|
|
hook(type => "getopt", id => "inline", call => \&getopt);
|
|
hook(type => "getsetup", id => "inline", call => \&getsetup);
|
|
hook(type => "checkconfig", id => "inline", call => \&checkconfig);
|
|
hook(type => "sessioncgi", id => "inline", call => \&sessioncgi);
|
|
hook(type => "preprocess", id => "inline",
|
|
call => \&IkiWiki::preprocess_inline);
|
|
hook(type => "pagetemplate", id => "inline",
|
|
call => \&IkiWiki::pagetemplate_inline);
|
|
hook(type => "format", id => "inline", call => \&format, first => 1);
|
|
# Hook to change to do pinging since it's called late.
|
|
# This ensures each page only pings once and prevents slow
|
|
# pings interrupting page builds.
|
|
hook(type => "change", id => "inline", call => \&IkiWiki::pingurl);
|
|
}
|
|
|
|
sub getopt () {
|
|
eval q{use Getopt::Long};
|
|
error($@) if $@;
|
|
Getopt::Long::Configure('pass_through');
|
|
GetOptions(
|
|
"rss!" => \$config{rss},
|
|
"atom!" => \$config{atom},
|
|
"allowrss!" => \$config{allowrss},
|
|
"allowatom!" => \$config{allowatom},
|
|
"pingurl=s" => sub {
|
|
push @{$config{pingurl}}, $_[1];
|
|
},
|
|
);
|
|
}
|
|
|
|
sub getsetup () {
|
|
return
|
|
plugin => {
|
|
safe => 1,
|
|
rebuild => undef,
|
|
section => "core",
|
|
},
|
|
rss => {
|
|
type => "boolean",
|
|
example => 0,
|
|
description => "enable rss feeds by default?",
|
|
safe => 1,
|
|
rebuild => 1,
|
|
},
|
|
atom => {
|
|
type => "boolean",
|
|
example => 0,
|
|
description => "enable atom feeds by default?",
|
|
safe => 1,
|
|
rebuild => 1,
|
|
},
|
|
allowrss => {
|
|
type => "boolean",
|
|
example => 0,
|
|
description => "allow rss feeds to be used?",
|
|
safe => 1,
|
|
rebuild => 1,
|
|
},
|
|
allowatom => {
|
|
type => "boolean",
|
|
example => 0,
|
|
description => "allow atom feeds to be used?",
|
|
safe => 1,
|
|
rebuild => 1,
|
|
},
|
|
pingurl => {
|
|
type => "string",
|
|
example => "http://rpc.technorati.com/rpc/ping",
|
|
description => "urls to ping (using XML-RPC) on feed update",
|
|
safe => 1,
|
|
rebuild => 0,
|
|
},
|
|
}
|
|
|
|
sub checkconfig () {
|
|
if (($config{rss} || $config{atom}) && ! length $config{url}) {
|
|
error(gettext("Must specify url to wiki with --url when using --rss or --atom"));
|
|
}
|
|
if ($config{rss}) {
|
|
push @{$config{wiki_file_prune_regexps}}, qr/\.rss$/;
|
|
}
|
|
if ($config{atom}) {
|
|
push @{$config{wiki_file_prune_regexps}}, qr/\.atom$/;
|
|
}
|
|
if (! exists $config{pingurl}) {
|
|
$config{pingurl}=[];
|
|
}
|
|
}
|
|
|
|
sub format (@) {
|
|
my %params=@_;
|
|
|
|
# Fill in the inline content generated earlier. This is actually an
|
|
# optimisation.
|
|
$params{content}=~s{<div class="inline" id="([^"]+)"></div>}{
|
|
delete @inline[$1,]
|
|
}eg;
|
|
return $params{content};
|
|
}
|
|
|
|
sub sessioncgi ($$) {
|
|
my $q=shift;
|
|
my $session=shift;
|
|
|
|
if ($q->param('do') eq 'blog') {
|
|
my $page=titlepage(decode_utf8($q->param('title')));
|
|
$page=~s/(\/)/"__".ord($1)."__"/eg; # don't create subdirs
|
|
# if the page already exists, munge it to be unique
|
|
my $from=$q->param('from');
|
|
my $add="";
|
|
while (exists $IkiWiki::pagecase{lc($from."/".$page.$add)}) {
|
|
$add=1 unless length $add;
|
|
$add++;
|
|
}
|
|
$q->param('page', "/$from/$page$add");
|
|
# now go create the page
|
|
$q->param('do', 'create');
|
|
# make sure the editpage plugin is loaded
|
|
if (IkiWiki->can("cgi_editpage")) {
|
|
IkiWiki::cgi_editpage($q, $session);
|
|
}
|
|
else {
|
|
error(gettext("page editing not allowed"));
|
|
}
|
|
exit;
|
|
}
|
|
}
|
|
|
|
# Back to ikiwiki namespace for the rest, this code is very much
|
|
# internal to ikiwiki even though it's separated into a plugin.
|
|
package IkiWiki;
|
|
|
|
my %toping;
|
|
my %feedlinks;
|
|
|
|
sub preprocess_inline (@) {
|
|
my %params=@_;
|
|
|
|
if (! exists $params{pages} && ! exists $params{pagenames}) {
|
|
error gettext("missing pages parameter");
|
|
}
|
|
my $raw=yesno($params{raw});
|
|
my $archive=yesno($params{archive});
|
|
my $rss=(($config{rss} || $config{allowrss}) && exists $params{rss}) ? yesno($params{rss}) : $config{rss};
|
|
my $atom=(($config{atom} || $config{allowatom}) && exists $params{atom}) ? yesno($params{atom}) : $config{atom};
|
|
my $quick=exists $params{quick} ? yesno($params{quick}) : 0;
|
|
my $feeds=exists $params{feeds} ? yesno($params{feeds}) : !$quick && ! $raw;
|
|
my $emptyfeeds=exists $params{emptyfeeds} ? yesno($params{emptyfeeds}) : 1;
|
|
my $feedonly=yesno($params{feedonly});
|
|
if (! exists $params{show} && ! $archive) {
|
|
$params{show}=10;
|
|
}
|
|
if (! exists $params{feedshow} && exists $params{show}) {
|
|
$params{feedshow}=$params{show};
|
|
}
|
|
my $desc;
|
|
if (exists $params{description}) {
|
|
$desc = $params{description}
|
|
}
|
|
else {
|
|
$desc = $config{wikiname};
|
|
}
|
|
my $actions=yesno($params{actions});
|
|
if (exists $params{template}) {
|
|
$params{template}=~s/[^-_a-zA-Z0-9]+//g;
|
|
}
|
|
else {
|
|
$params{template} = $archive ? "archivepage" : "inlinepage";
|
|
}
|
|
|
|
my @list;
|
|
|
|
if (exists $params{pagenames}) {
|
|
foreach my $p (qw(sort pages)) {
|
|
if (exists $params{$p}) {
|
|
error sprintf(gettext("the %s and %s parameters cannot be used together"),
|
|
"pagenames", $p);
|
|
}
|
|
}
|
|
|
|
@list = map { bestlink($params{page}, $_) }
|
|
split ' ', $params{pagenames};
|
|
|
|
if (yesno($params{reverse})) {
|
|
@list=reverse(@list);
|
|
}
|
|
|
|
foreach my $p (@list) {
|
|
add_depends($params{page}, $p, deptype($quick ? "presence" : "content"));
|
|
}
|
|
}
|
|
else {
|
|
my $num=0;
|
|
if ($params{show}) {
|
|
$num=$params{show};
|
|
}
|
|
if ($params{feedshow} && $num < $params{feedshow} && $num > 0) {
|
|
$num=$params{feedshow};
|
|
}
|
|
if ($params{skip} && $num) {
|
|
$num+=$params{skip};
|
|
}
|
|
|
|
@list = pagespec_match_list($params{page}, $params{pages},
|
|
deptype => deptype($quick ? "presence" : "content"),
|
|
filter => sub { $_[0] eq $params{page} },
|
|
sort => exists $params{sort} ? $params{sort} : "age",
|
|
reverse => yesno($params{reverse}),
|
|
($num ? (num => $num) : ()),
|
|
);
|
|
}
|
|
|
|
if (exists $params{skip}) {
|
|
@list=@list[$params{skip} .. $#list];
|
|
}
|
|
|
|
my @feedlist;
|
|
if ($feeds) {
|
|
if (exists $params{feedshow} &&
|
|
$params{feedshow} && @list > $params{feedshow}) {
|
|
@feedlist=@list[0..$params{feedshow} - 1];
|
|
}
|
|
else {
|
|
@feedlist=@list;
|
|
}
|
|
}
|
|
|
|
if ($params{show} && @list > $params{show}) {
|
|
@list=@list[0..$params{show} - 1];
|
|
}
|
|
|
|
if ($feeds && exists $params{feedpages}) {
|
|
@feedlist = pagespec_match_list(
|
|
$params{page}, "($params{pages}) and ($params{feedpages})",
|
|
deptype => deptype($quick ? "presence" : "content"),
|
|
list => \@feedlist,
|
|
);
|
|
}
|
|
|
|
my ($feedbase, $feednum);
|
|
if ($feeds) {
|
|
# Ensure that multiple feeds on a page go to unique files.
|
|
|
|
# Feedfile can lead to conflicts if usedirs is not enabled,
|
|
# so avoid supporting it in that case.
|
|
delete $params{feedfile} if ! $config{usedirs};
|
|
# Tight limits on legal feedfiles, to avoid security issues
|
|
# and conflicts.
|
|
if (defined $params{feedfile}) {
|
|
if ($params{feedfile} =~ /\// ||
|
|
$params{feedfile} !~ /$config{wiki_file_regexp}/) {
|
|
error("illegal feedfile");
|
|
}
|
|
$params{feedfile}=possibly_foolish_untaint($params{feedfile});
|
|
}
|
|
$feedbase=targetpage($params{page}, "", $params{feedfile});
|
|
|
|
my $feedid=join("\0", $feedbase, map { $_."\0".$params{$_} } sort keys %params);
|
|
if (exists $knownfeeds{$feedid}) {
|
|
$feednum=$knownfeeds{$feedid};
|
|
}
|
|
else {
|
|
if (exists $page_numfeeds{$params{destpage}}{$feedbase}) {
|
|
if ($feeds) {
|
|
$feednum=$knownfeeds{$feedid}=++$page_numfeeds{$params{destpage}}{$feedbase};
|
|
}
|
|
}
|
|
else {
|
|
$feednum=$knownfeeds{$feedid}="";
|
|
if ($feeds) {
|
|
$page_numfeeds{$params{destpage}}{$feedbase}=1;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
my ($rssurl, $atomurl, $rssdesc, $atomdesc);
|
|
if ($feeds) {
|
|
if ($rss) {
|
|
$rssurl=abs2rel($feedbase."rss".$feednum, dirname(htmlpage($params{destpage})));
|
|
$rssdesc = sprintf(gettext("%s (RSS feed)"), $desc);
|
|
}
|
|
if ($atom) {
|
|
$atomurl=abs2rel($feedbase."atom".$feednum, dirname(htmlpage($params{destpage})));
|
|
$atomdesc = sprintf(gettext("%s (Atom feed)"), $desc);
|
|
}
|
|
}
|
|
|
|
my $ret="";
|
|
|
|
if (length $config{cgiurl} && ! $params{preview} && (exists $params{rootpage} ||
|
|
(exists $params{postform} && yesno($params{postform}))) &&
|
|
IkiWiki->can("cgi_editpage")) {
|
|
# Add a blog post form, with feed buttons.
|
|
my $formtemplate=template_depends("blogpost.tmpl", $params{page}, blind_cache => 1);
|
|
$formtemplate->param(cgiurl => IkiWiki::cgiurl());
|
|
$formtemplate->param(rootpage => rootpage(%params));
|
|
if ($feeds) {
|
|
if ($rss) {
|
|
$formtemplate->param(rssurl => $rssurl);
|
|
$formtemplate->param(rssdesc => $rssdesc);
|
|
}
|
|
if ($atom) {
|
|
$formtemplate->param(atomurl => $atomurl);
|
|
$formtemplate->param(atomdesc => $atomdesc);
|
|
}
|
|
}
|
|
if (exists $params{postformtext}) {
|
|
$formtemplate->param(postformtext =>
|
|
$params{postformtext});
|
|
}
|
|
else {
|
|
$formtemplate->param(postformtext =>
|
|
gettext("Add a new post titled:"));
|
|
}
|
|
$ret.=$formtemplate->output;
|
|
|
|
# The post form includes the feed buttons, so
|
|
# emptyfeeds cannot be hidden.
|
|
$emptyfeeds=1;
|
|
}
|
|
elsif ($feeds && !$params{preview} && ($emptyfeeds || @feedlist)) {
|
|
# Add feed buttons.
|
|
my $linktemplate=template_depends("feedlink.tmpl", $params{page}, blind_cache => 1);
|
|
if ($rss) {
|
|
$linktemplate->param(rssurl => $rssurl);
|
|
$linktemplate->param(rssdesc => $rssdesc);
|
|
}
|
|
if ($atom) {
|
|
$linktemplate->param(atomurl => $atomurl);
|
|
$linktemplate->param(atomdesc => $atomdesc);
|
|
}
|
|
$ret.=$linktemplate->output;
|
|
}
|
|
|
|
if (! $feedonly) {
|
|
my $template;
|
|
if (! $raw) {
|
|
# cannot use wiki pages as templates; template not sanitized due to
|
|
# format hook hack
|
|
eval {
|
|
$template=template_depends($params{template}.".tmpl", $params{page},
|
|
blind_cache => 1);
|
|
};
|
|
if ($@) {
|
|
error sprintf(gettext("failed to process template %s"), $params{template}.".tmpl").": $@";
|
|
}
|
|
}
|
|
my $needcontent=$raw || (!($archive && $quick) && $template->query(name => 'content'));
|
|
|
|
foreach my $page (@list) {
|
|
my $file = $pagesources{$page};
|
|
my $type = pagetype($file);
|
|
if (! $raw) {
|
|
if ($needcontent) {
|
|
# Get the content before populating the
|
|
# template, since getting the content uses
|
|
# the same template if inlines are nested.
|
|
my $content=get_inline_content($page, $params{destpage});
|
|
$template->param(content => $content);
|
|
}
|
|
$template->param(pageurl => urlto($page, $params{destpage}));
|
|
$template->param(inlinepage => $page);
|
|
$template->param(title => pagetitle(basename($page)));
|
|
$template->param(ctime => displaytime($pagectime{$page}, $params{timeformat}, 1));
|
|
$template->param(mtime => displaytime($pagemtime{$page}, $params{timeformat}));
|
|
$template->param(first => 1) if $page eq $list[0];
|
|
$template->param(last => 1) if $page eq $list[$#list];
|
|
$template->param(html5 => $config{html5});
|
|
|
|
if ($actions) {
|
|
my $file = $pagesources{$page};
|
|
my $type = pagetype($file);
|
|
if ($config{discussion}) {
|
|
if ($page !~ /.*\/\Q$config{discussionpage}\E$/i &&
|
|
(length $config{cgiurl} ||
|
|
exists $pagesources{$page."/".lc($config{discussionpage})})) {
|
|
$template->param(have_actions => 1);
|
|
$template->param(discussionlink =>
|
|
htmllink($page,
|
|
$params{destpage},
|
|
$config{discussionpage},
|
|
noimageinline => 1,
|
|
forcesubpage => 1));
|
|
}
|
|
}
|
|
if (length $config{cgiurl} &&
|
|
defined $type &&
|
|
IkiWiki->can("cgi_editpage")) {
|
|
$template->param(have_actions => 1);
|
|
$template->param(editurl => cgiurl(do => "edit", page => $page));
|
|
|
|
}
|
|
}
|
|
|
|
run_hooks(pagetemplate => sub {
|
|
shift->(page => $page, destpage => $params{destpage},
|
|
template => $template,);
|
|
});
|
|
|
|
$ret.=$template->output;
|
|
$template->clear_params;
|
|
}
|
|
else {
|
|
if (defined $type) {
|
|
$ret.="\n".
|
|
linkify($page, $params{destpage},
|
|
preprocess($page, $params{destpage},
|
|
filter($page, $params{destpage},
|
|
readfile(srcfile($file)))));
|
|
}
|
|
else {
|
|
$ret.="\n".
|
|
readfile(srcfile($file));
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
if ($feeds && ($emptyfeeds || @feedlist)) {
|
|
if ($rss) {
|
|
my $rssp=$feedbase."rss".$feednum;
|
|
will_render($params{destpage}, $rssp);
|
|
if (! $params{preview}) {
|
|
writefile($rssp, $config{destdir},
|
|
genfeed("rss",
|
|
$config{url}."/".$rssp, $desc, $params{guid}, $params{destpage}, @feedlist));
|
|
$toping{$params{destpage}}=1 unless $config{rebuild};
|
|
$feedlinks{$params{destpage}}.=qq{<link rel="alternate" type="application/rss+xml" title="$rssdesc" href="$rssurl" />};
|
|
}
|
|
}
|
|
if ($atom) {
|
|
my $atomp=$feedbase."atom".$feednum;
|
|
will_render($params{destpage}, $atomp);
|
|
if (! $params{preview}) {
|
|
writefile($atomp, $config{destdir},
|
|
genfeed("atom", $config{url}."/".$atomp, $desc, $params{guid}, $params{destpage}, @feedlist));
|
|
$toping{$params{destpage}}=1 unless $config{rebuild};
|
|
$feedlinks{$params{destpage}}.=qq{<link rel="alternate" type="application/atom+xml" title="$atomdesc" href="$atomurl" />};
|
|
}
|
|
}
|
|
}
|
|
|
|
clear_inline_content_cache();
|
|
|
|
return $ret if $raw || $nested;
|
|
push @inline, $ret;
|
|
return "<div class=\"inline\" id=\"$#inline\"></div>\n\n";
|
|
}
|
|
|
|
sub pagetemplate_inline (@) {
|
|
my %params=@_;
|
|
my $page=$params{page};
|
|
my $template=$params{template};
|
|
|
|
$template->param(feedlinks => $feedlinks{$page})
|
|
if exists $feedlinks{$page} && $template->query(name => "feedlinks");
|
|
}
|
|
|
|
{
|
|
my %inline_content;
|
|
my $cached_destpage="";
|
|
|
|
sub get_inline_content ($$) {
|
|
my $page=shift;
|
|
my $destpage=shift;
|
|
|
|
if (exists $inline_content{$page} && $cached_destpage eq $destpage) {
|
|
return $inline_content{$page};
|
|
}
|
|
|
|
my $file=$pagesources{$page};
|
|
my $type=pagetype($file);
|
|
my $ret="";
|
|
if (defined $type) {
|
|
$nested++;
|
|
$ret=htmlize($page, $destpage, $type,
|
|
linkify($page, $destpage,
|
|
preprocess($page, $destpage,
|
|
filter($page, $destpage,
|
|
readfile(srcfile($file))))));
|
|
$nested--;
|
|
if (isinternal($page)) {
|
|
# make inlined text of internal pages searchable
|
|
run_hooks(indexhtml => sub {
|
|
shift->(page => $page, destpage => $page,
|
|
content => $ret);
|
|
});
|
|
}
|
|
}
|
|
|
|
if ($cached_destpage ne $destpage) {
|
|
clear_inline_content_cache();
|
|
$cached_destpage=$destpage;
|
|
}
|
|
return $inline_content{$page}=$ret;
|
|
}
|
|
|
|
sub clear_inline_content_cache () {
|
|
%inline_content=();
|
|
}
|
|
|
|
}
|
|
|
|
sub date_822 ($) {
|
|
my $time=shift;
|
|
|
|
my $lc_time=POSIX::setlocale(&POSIX::LC_TIME);
|
|
POSIX::setlocale(&POSIX::LC_TIME, "C");
|
|
my $ret=POSIX::strftime("%a, %d %b %Y %H:%M:%S %z", localtime($time));
|
|
POSIX::setlocale(&POSIX::LC_TIME, $lc_time);
|
|
return $ret;
|
|
}
|
|
|
|
sub absolute_urls ($$) {
|
|
# needed because rss sucks
|
|
my $html=shift;
|
|
my $baseurl=shift;
|
|
|
|
my $url=$baseurl;
|
|
$url=~s/[^\/]+$//;
|
|
my $urltop; # calculated if needed
|
|
|
|
my $ret="";
|
|
|
|
eval q{use HTML::Parser; use HTML::Tagset};
|
|
die $@ if $@;
|
|
my $p = HTML::Parser->new(api_version => 3);
|
|
$p->handler(default => sub { $ret.=join("", @_) }, "text");
|
|
$p->handler(start => sub {
|
|
my ($tagname, $pos, $text) = @_;
|
|
if (ref $HTML::Tagset::linkElements{$tagname}) {
|
|
while (4 <= @$pos) {
|
|
# use attribute sets from right to left
|
|
# to avoid invalidating the offsets
|
|
# when replacing the values
|
|
my ($k_offset, $k_len, $v_offset, $v_len) =
|
|
splice(@$pos, -4);
|
|
my $attrname = lc(substr($text, $k_offset, $k_len));
|
|
next unless grep { $_ eq $attrname } @{$HTML::Tagset::linkElements{$tagname}};
|
|
next unless $v_offset; # 0 v_offset means no value
|
|
my $v = substr($text, $v_offset, $v_len);
|
|
$v =~ s/^([\'\"])(.*)\1$/$2/;
|
|
if ($v=~/^#/) {
|
|
$v=$baseurl.$v; # anchor
|
|
}
|
|
elsif ($v=~/^(?!\w+:)[^\/]/) {
|
|
$v=$url.$v; # relative url
|
|
}
|
|
elsif ($v=~/^\//) {
|
|
if (! defined $urltop) {
|
|
# what is the non path part of the url?
|
|
my $top_uri = URI->new($url);
|
|
$top_uri->path_query(""); # reset the path
|
|
$urltop = $top_uri->as_string;
|
|
}
|
|
$v=$urltop.$v; # url relative to top of site
|
|
}
|
|
$v =~ s/\"/"/g; # since we quote with ""
|
|
substr($text, $v_offset, $v_len) = qq("$v");
|
|
}
|
|
}
|
|
$ret.=$text;
|
|
}, "tagname, tokenpos, text");
|
|
$p->parse($html);
|
|
$p->eof;
|
|
|
|
return $ret;
|
|
}
|
|
|
|
sub genfeed ($$$$$@) {
|
|
my $feedtype=shift;
|
|
my $feedurl=shift;
|
|
my $feeddesc=shift;
|
|
my $guid=shift;
|
|
my $page=shift;
|
|
my @pages=@_;
|
|
|
|
my $url=URI->new(encode_utf8(urlto($page,"",1)));
|
|
|
|
my $itemtemplate=template_depends($feedtype."item.tmpl", $page, blind_cache => 1);
|
|
my $content="";
|
|
my $lasttime = 0;
|
|
foreach my $p (@pages) {
|
|
my $u=URI->new(encode_utf8(urlto($p, "", 1)));
|
|
my $pcontent = absolute_urls(get_inline_content($p, $page), $url);
|
|
|
|
$itemtemplate->param(
|
|
title => pagetitle(basename($p)),
|
|
url => $u,
|
|
permalink => $u,
|
|
cdate_822 => date_822($pagectime{$p}),
|
|
mdate_822 => date_822($pagemtime{$p}),
|
|
cdate_3339 => date_3339($pagectime{$p}),
|
|
mdate_3339 => date_3339($pagemtime{$p}),
|
|
);
|
|
|
|
if (exists $pagestate{$p}) {
|
|
if (exists $pagestate{$p}{meta}{guid}) {
|
|
eval q{use HTML::Entities};
|
|
$itemtemplate->param(guid => HTML::Entities::encode_numeric($pagestate{$p}{meta}{guid}));
|
|
}
|
|
|
|
if (exists $pagestate{$p}{meta}{updated}) {
|
|
$itemtemplate->param(mdate_822 => date_822($pagestate{$p}{meta}{updated}));
|
|
$itemtemplate->param(mdate_3339 => date_3339($pagestate{$p}{meta}{updated}));
|
|
}
|
|
}
|
|
|
|
if ($itemtemplate->query(name => "enclosure")) {
|
|
my $file=$pagesources{$p};
|
|
my $type=pagetype($file);
|
|
if (defined $type) {
|
|
$itemtemplate->param(content => $pcontent);
|
|
}
|
|
else {
|
|
my $size=(srcfile_stat($file))[8];
|
|
my $mime="unknown";
|
|
eval q{use File::MimeInfo};
|
|
if (! $@) {
|
|
$mime = mimetype($file);
|
|
}
|
|
$itemtemplate->param(
|
|
enclosure => $u,
|
|
type => $mime,
|
|
length => $size,
|
|
);
|
|
}
|
|
}
|
|
else {
|
|
$itemtemplate->param(content => $pcontent);
|
|
}
|
|
|
|
run_hooks(pagetemplate => sub {
|
|
shift->(page => $p, destpage => $page,
|
|
template => $itemtemplate);
|
|
});
|
|
|
|
$content.=$itemtemplate->output;
|
|
$itemtemplate->clear_params;
|
|
|
|
$lasttime = $pagemtime{$p} if $pagemtime{$p} > $lasttime;
|
|
}
|
|
|
|
my $template=template_depends($feedtype."page.tmpl", $page, blind_cache => 1);
|
|
$template->param(
|
|
title => $page ne "index" ? pagetitle($page) : $config{wikiname},
|
|
wikiname => $config{wikiname},
|
|
pageurl => $url,
|
|
content => $content,
|
|
feeddesc => $feeddesc,
|
|
guid => $guid,
|
|
feeddate => date_3339($lasttime),
|
|
feedurl => $feedurl,
|
|
version => $IkiWiki::version,
|
|
);
|
|
run_hooks(pagetemplate => sub {
|
|
shift->(page => $page, destpage => $page,
|
|
template => $template);
|
|
});
|
|
|
|
return $template->output;
|
|
}
|
|
|
|
sub pingurl (@) {
|
|
return unless @{$config{pingurl}} && %toping;
|
|
|
|
eval q{require RPC::XML::Client};
|
|
if ($@) {
|
|
debug(gettext("RPC::XML::Client not found, not pinging"));
|
|
return;
|
|
}
|
|
|
|
# daemonize here so slow pings don't slow down wiki updates
|
|
defined(my $pid = fork) or error("Can't fork: $!");
|
|
return if $pid;
|
|
chdir '/';
|
|
POSIX::setsid() or error("Can't start a new session: $!");
|
|
open STDIN, '/dev/null';
|
|
open STDOUT, '>/dev/null';
|
|
open STDERR, '>&STDOUT' or error("Can't dup stdout: $!");
|
|
|
|
# Don't need to keep a lock on the wiki as a daemon.
|
|
IkiWiki::unlockwiki();
|
|
|
|
foreach my $page (keys %toping) {
|
|
my $title=pagetitle(basename($page), 0);
|
|
my $url=urlto($page, "", 1);
|
|
foreach my $pingurl (@{$config{pingurl}}) {
|
|
debug("Pinging $pingurl for $page");
|
|
eval {
|
|
my $client = RPC::XML::Client->new($pingurl);
|
|
my $req = RPC::XML::request->new('weblogUpdates.ping',
|
|
$title, $url);
|
|
my $res = $client->send_request($req);
|
|
if (! ref $res) {
|
|
error("Did not receive response to ping");
|
|
}
|
|
my $r=$res->value;
|
|
if (! exists $r->{flerror} || $r->{flerror}) {
|
|
error("Ping rejected: ".(exists $r->{message} ? $r->{message} : "[unknown reason]"));
|
|
}
|
|
};
|
|
if ($@) {
|
|
error "Ping failed: $@";
|
|
}
|
|
}
|
|
}
|
|
|
|
exit 0; # daemon done
|
|
}
|
|
|
|
|
|
sub rootpage (@) {
|
|
my %params=@_;
|
|
|
|
my $rootpage;
|
|
if (exists $params{rootpage}) {
|
|
$rootpage=bestlink($params{page}, $params{rootpage});
|
|
if (!length $rootpage) {
|
|
$rootpage=$params{rootpage};
|
|
}
|
|
}
|
|
else {
|
|
$rootpage=$params{page};
|
|
}
|
|
return $rootpage;
|
|
}
|
|
|
|
1
|