Code: Select all
# Script to grab titles from webpages
# Updated version by teel @ IRCnet
#
# https://github.com/teeli/urltitle
#
# Detects URL from IRC channels and prints out the title
#
# Version Log:
# 0.11 Updated regex parser to only parse titles inside <head> tags
# Added HTTP error status logging
# 0.10 Fixed XPath parsing error and added regex fallback if XPath fails
# 0.09 HTTPs redirects, case-insensitive HTTP header fix, other small bug fixes
# 0.08 Changed putserv to puthelp to queue the messages
# 0.07 Added Content-Type check (text/html only) and exceptino handling for tDom with a fallback to
# regexp if tDom fails.
# 0.06 Added XPATH support to title parsing (only if tdom package is available)
# 0.05 Added SNI support for TLS (with TLS version check)
# 0.04 HTML parsing for titles added
# 0.03c HTTPS support is now optional and will be automatically dropeed if TCL TSL package does not exist
# 0.03b Some formatting
# 0.03 HTTPS support
# 0.02 Updated version by teel. Added support for redirects, trimmed titles (remove extra whitespaces),
# some optimization
# 0.01a Original version by rosc
#
################################################################################################################
#
# Original script:
# Copyright C.Leonhardt (rosc2112 at yahoo com) Aug.11.2007
# http://members.dandy.net/~fbn/urltitle.tcl.txt
# Loosely based on the tinyurl script by Jer and other bits and pieces of my own..
#
################################################################################################################
#
# Usage:
#
# 1) Set the configs below
# 2) .chanset #channelname +urltitle ;# enable script
# 3) .chanset #channelname +logurltitle ;# enable logging
# Then just input a url in channel and the script will retrieve the title from the corresponding page.
#
################################################################################################################
namespace eval UrlTitle {
# CONFIG
variable ignore "bdkqr|dkqr" ;# User flags script will ignore input from
variable length 5 ;# minimum url length to trigger channel eggdrop use
variable delay 1 ;# minimum seconds to wait before another eggdrop use
variable timeout 5000 ;# geturl timeout (1/1000ths of a second)
variable fetchLimit 5 ;# How many times to process redirects before erroring
# BINDS
bind pubm "-|-" {*://*} UrlTitle::handler
setudef flag urltitle ;# Channel flag to enable script.
setudef flag logurltitle ;# Channel flag to enable logging of script.
# INTERNAL
variable last 1 ;# Internal variable, stores time of last eggdrop use, don't change..
variable scriptVersion 0.11
# PACKAGES
package require http ;# You need the http package..
variable httpsSupport false
variable htmlSupport false
variable tdomSupport false
if {![catch {variable tlsVersion [package require tls]}]} {
set httpsSupport true
if {[package vcompare $tlsVersion 1.6.4] < 0} {
putlog "UrlTitle: TCL TLS version 1.6.4 or newer is required for proper https support (SNI)"
}
}
if {![catch {package require htmlparse}]} {
set htmlSupport true
}
if {![catch {package require tdom}]} {
set tdomSupport true
}
# Enable SNI support for TLS if suitable TLS version is installed
proc socket {args} {
variable tlsVersion
set opts [lrange $args 0 end-2]
set host [lindex $args end-1]
set port [lindex $args end]
if {[package vcompare $tlsVersion 1.7.11] >= 0} {
# tls version 1.7.11 should support autoservername
::tls::socket -autoservername true {*}$opts $host $port
} elseif {[package vcompare $tlsVersion 1.6.4] >= 0} {
::tls::socket -ssl3 false -ssl2 false -tls1 true -servername $host {*}$opts $host $port
} else {
# default fallback without servername (SNI certs will not work)
::tls::socket -ssl3 false -ssl2 false -tls1 true {*}$opts $host $port
}
}
proc handler {nick host user chan text} {
variable httpsSupport
variable htmlSupport
variable delay
variable last
variable ignore
variable length
set unixtime [clock seconds]
if {[channel get $chan urltitle] && ($unixtime - $delay) > $last && (![matchattr $user $ignore])} {
foreach word [split $text] {
if {[string length $word] >= $length && [regexp {^(f|ht)tp(s|)://} $word] && \
![regexp {://([^/:]*:([^/]*@|\d+(/|$))|.*/\.)} $word]} {
set last $unixtime
# enable https if supported
if {$httpsSupport} {
::http::register https 443 [list UrlTitle::socket]
}
set urtitle [UrlTitle::parse $word]
if {$htmlSupport} {
set urtitle [::htmlparse::mapEscapes $urtitle]
}
# unregister https if supported
if {$httpsSupport} {
::http::unregister https
}
if {$urtitle eq ""} {
break
}
if {[string length $urtitle]} {
puthelp "PRIVMSG $chan :Title: $urtitle"
}
break
}
}
}
# change to return 0 if you want the pubm trigger logged additionally..
return 1
}
# General HTTP redirect handler
proc Fetch {url args} {
variable fetchLimit
for {set count 0} {$count < $fetchLimit} {incr count} {
set token [::http::geturl $url {*}$args]
if {[::http::status $token] ne "ok" || ![string match 3?? [::http::ncode $token]]} {
break
}
set meta [::http::meta $token]
if {[dict exists $meta Location]} {
set url [dict get $meta Location]
}
if {[dict exists $meta location]} {
set url [dict get $meta location]
}
::http::cleanup $token
}
return $token
}
proc parseTitleXPath {data} {
set title ""
if {[catch {set doc [dom parse -html -simple $data]} results]} {
# fallback to regex parsing if tdom fails
set title [parseTitleRegex $data]
} else {
# parse dom
set root [$doc documentElement]
set node [$root selectNodes {//head/title/text()}]
if {$node != ""} {
# return title if XPath was able to parse it
set title [$node data]
} else {
# Fallback to regex if XPath failed
set title [parseTitleRegex $data]
}
}
}
proc parseTitleRegex {data} {
set title ""
# fallback to regex parsing if tdom fails
regexp -nocase {<head>.*<title.*>(.*?)</title>.*</head>} $data match title
set title [regsub -all -nocase {\s+} $title " "]
return $title
}
proc parse {url} {
variable timeout
variable tdomSupport
set title ""
if {[info exists url] && [string length $url]} {
if {[catch {set http [Fetch $url -timeout $timeout]} results]} {
putlog "Connection to $url failed"
putlog "Error: $results"
} else {
if { [::http::status $http] == "ok" } {
set data [::http::data $http]
set status [::http::code $http]
set meta [::http::meta $http]
# only parse html files for titles
if {
([dict exists $meta Content-Type] && [string first "text/html" [dict get $meta Content-Type]] >= 0) ||
([dict exists $meta content-type] && [string first "text/html" [dict get $meta content-type]] >= 0)
} {
switch -regexp -- $status {
"HTTP.*200.*" {
if {$tdomSupport} {
# use XPATH if tdom is supported
set title [parseTitleXPath $data]
} else {
# fallback to regex parsing if tdom is not enabled
set title [parseTitleRegex $data]
}
}
"HTTP\/[0-1]\.[0-1].3.*" {
if {[dict exists $meta Location]} {
set title [UrlTitle::parse [dict get $meta Location]]
}
if {[dict exists $meta location]} {
set title [UrlTitle::parse [dict get $meta location]]
}
}
default {
putlog "Error: $status ($url)"
}
}
}
} else {
putlog "Connection to $url failed"
}
::http::cleanup $http
}
}
return $title
}
putlog "Initialized Url Title Grabber v$scriptVersion"
}
gettin this error:
got the code here:16:15:55 <TCL-Tester> [16:15:38] Connection to https://github.com/teeli/urltitle failed
16:15:55 <TCL-Tester> [16:15:38] Error: error reading "sock5cd82c0": software caused connection abort
https://github.com/teeli/urltitle