Swish-E Logo


spider.pl - Example Perl program to spider web servers


Table of Contents:

[ TOC ]

SYNOPSIS

 
    spider.pl [<spider config file>] [<URL> ...]

 
    # Spider using some common defaults and capture the output
    # into a file

 
    ./spider.pl default http://myserver.com/ > output.txt

 
    # or using a config file

 
    spider.config:
    @servers = (
        {
            base_url    => 'http://myserver.com/',
            email       => 'me@myself.com',
            # other spider settings described below
        },
    );

 
    ./spider.pl spider.config > output.txt

 
    # or using the default config file SwishSpiderConfig.pl
    ./spider.pl > output.txt

 
    # using with swish-e

 
    ./spider.pl spider.config | swish-e -c swish.config -S prog -i stdin

 
    # or in two steps
    ./spider.pl spider.config > output.txt
    swish-e -c swish.config -S prog -i stdin < output.txt

 
    # or with compression
    ./spider.pl spider.config | gzip > output.gz
    gzip -dc output.gz | swish-e -c swish.config -S prog -i stdin

 
    # or having swish-e call the spider directly using the
    # spider config file SwishSpiderConfig.pl:
    swish-e -c swish.config -S prog -i spider.pl

 
    # or the above but passing passing a parameter to the spider:
    echo "SwishProgParameters  spider.config" >> swish.config
    echo "IndexDir spider.pl" >> swish.config
    swish-e -c swish.config -S prog

 
    Note: When running on some versions of Windows (e.g. Win ME and Win 98 SE)
    you may need to tell Perl to run the spider directly:

 
        perl spider.pl | swish-e -S prog -c swish.conf -i stdin

 
    This pipes the output of the spider directly into swish.

[ TOC ]


DESCRIPTION

spider.pl is a program for fetching documnts from a web server, and outputs the documents to STDOUT in a special format designed to be read by Swish-e.

The spider can index non-text documents such as PDF and MS Word by use of filter (helper) programs. These programs are not part of the Swish-e distribution and must be installed separately. See the section on filtering below.

A configuration file is noramlly used to control what documents are fetched from the web server(s). The configuration file and its options are described below. The is also a "default" config suitable for spidering.

The spider is designed to spider web pages and fetch documents from one host at a time -- offsite links are not followed. But, you can configure the spider to spider multiple sites in a single run.

spider.pl is distributed with Swish-e and is installed in the swish-e library directory at installation time. This directory (libexedir) can be seen by running the command:

 
    swish-e -h

Typically on unix-type systems the spider is installed at:

 
    /usr/local/lib/swish-e/spider.pl

This spider stores all links in memory while processing and does not do parallel requests.

[ TOC ]


Running the spider

The output from spider.pl can be captured to a temporary file which is then fed into swish-e:

 
    ./spider.pl > docs.txt
    swish-e -c config -S prog -i stdin < docs.txt

or the output can be passed to swish-e via a pipe:

 
   ./spider.pl | swish-e -c config -S prog -i stdin

or the swish-e can run the spider directly:

 
   swish-e -c config -S prog -i spider.pl

One advantage of having Swish-e run spider.pl is that Swish-e knows where to locate the program (based on libexecdir compiled into swish-e).

When running the spider without any parameters it looks for a configuration file called SwishSpiderConfig.pl in the current directory. The spider will abort with an error if this file is not found.

A configuration file can be specified as the first parameter to the spider:

 
    ./spider.pl spider.config > output.txt

If running the spider via Swish-e (i.e. Swish-e runs the spider) then use the Swish-e config option SwishProgParameters to specify the config file:

In swish.config:

 
    # Use spider.pl as the external program:
    IndexDir spider.pl
    # And pass the name of the spider config file to the spider:
    SwishProgParameters spider.config

And then run Swish-e like this:

 
    swish-e -c swish.config -S prog

Finally, by using the special word "default" on the command line the spider will use a default configuration that is useful for indexing most sites. It's a good way to get started with the spider:

 
    ./spider.pl default http://my_server.com/index.html > output.txt

There's no "best" way to run the spider. I like to capture to a file and then feed that into Swish-e.

The spider does require Perl's LWP library and a few other reasonably common modules. Most well maintained systems should have these modules installed. See REQUIREMENTS below for more information. It's a good idea to check that you are running a current version of these modules.

Note: the "prog" document source in Swish-e bypasses many Swish-e configuration settings. For example, you cannot use the IndexOnly directive with the "prog" document source. This is by design to limit the overhead when using an external program for providing documents to swish; after all, with "prog", if you don't want to index a file, then don't give it to swish to index in the first place.

So, for spidering, if you do not wish to index images, for example, you will need to either filter by the URL or by the content-type returned from the web server. See CALLBACK FUNCTIONS below for more information.

[ TOC ]


Robots Exclusion Rules and being nice

By default, this script will not spider files blocked by robots.txt. In addition, The script will check for <meta name="robots"..> tags, which allows finer control over what files are indexed and/or spidered. See http://www.robotstxt.org/wc/exclusion.html for details.

This spider provides an extension to the <meta> tag exclusion, by adding a NOCONTENTS attribute. This attribute turns on the no_contents setting, which asks swish-e to only index the document's title (or file name if not title is found).

For example:

 
      <META NAME="ROBOTS" CONTENT="NOCONTENTS, NOFOLLOW">

says to just index the document's title, but don't index its contents, and don't follow any links within the document. Granted, it's unlikely that this feature will ever be used...

If you are indexing your own site, and know what you are doing, you can disable robot exclusion by the ignore_robots_file configuration parameter, described below. This disables both robots.txt and the meta tag parsing. You may disable just the meta tag parsing by using ignore_robots_headers.

This script only spiders one file at a time, so load on the web server is not that great. And with libwww-perl-5.53_91 HTTP/1.1 keep alive requests can reduce the load on the server even more (and potentially reduce spidering time considerably).

Still, discuss spidering with a site's administrator before beginning. Use the delay_sec to adjust how fast the spider fetches documents. Consider running a second web server with a limited number of children if you really want to fine tune the resources used by spidering.

[ TOC ]


Duplicate Documents

The spider program keeps track of URLs visited, so a document is only indexed one time.

The Digest::MD5 module can be used to create a "fingerprint" of every page indexed and this fingerprint is used in a hash to find duplicate pages. For example, MD5 will prevent indexing these as two different documents:

 
    http://localhost/path/to/some/index.html
    http://localhost/path/to/some/

But note that this may have side effects you don't want. If you want this file indexed under this URL:

 
    http://localhost/important.html

But the spider happens to find the exact content in this file first:

 
    http://localhost/developement/test/todo/maybeimportant.html

Then only that URL will be indexed.

[ TOC ]


Broken relative links

Sometimes web page authors use too many /../ segments in relative URLs which reference documents above the document root. Some web servers such as Apache will return a 400 Bad Request when requesting a document above the root. Other web servers such as Micorsoft IIS/5.0 will try and "correct" these errors. This correction will lead to loops when spidering.

The spider can fix these above-root links by placing the following in your spider config:

 
    remove_leading_dots => 1,

It is not on by default so that the spider can report the broken links (as 400 errors on sane webservers).

[ TOC ]


Compression

If The Perl module Compress::Zlib is installed the spider will send the

 
   Accept-Encoding: gzip

header and uncompress the document if the server returns the header

 
   Content-Encoding: gzip

MD5 checksomes are done on the compressed data.

MD5 may slow down indexing a tiny bit, so test with and without if speed is an issue (which it probably isn't since you are spidering in the first place). This feature will also use more memory.

[ TOC ]


REQUIREMENTS

Perl 5 (hopefully at least 5.00503) or later.

You must have the LWP Bundle on your computer. Load the LWP::Bundle via the CPAN.pm shell, or download libwww-perl-x.xx from CPAN (or via ActiveState's ppm utility). Also required is the the HTML-Parser-x.xx bundle of modules also from CPAN (and from ActiveState for Windows).

 
    http://search.cpan.org/search?dist=libwww-perl
    http://search.cpan.org/search?dist=HTML-Parser

You will also need Digest::MD5 if you wish to use the MD5 feature. HTML::Tagset is also required. Other modules may be required (for example, the pod2xml.pm module has its own requirementes -- see perldoc pod2xml for info).

The spider.pl script, like everyone else, expects perl to live in /usr/local/bin. If this is not the case then either add a symlink at /usr/local/bin/perl to point to where perl is installed or modify the shebang (#!) line at the top of the spider.pl program.

Note that the libwww-perl package does not support SSL (Secure Sockets Layer) (https) by default. See README.SSL included in the libwww-perl package for information on installing SSL support.

[ TOC ]


CONFIGURATION FILE

The spider configuration file is a read by the script as Perl code. This makes the configuration a bit more complex than simple text config files, but allows the spider to be configured programmatically.

For example, the config file can contain logic for testing URLs against regular expressions or even against a database lookup while running.

The configuration file sets an array called @servers. This array can contain one or more hash structures of parameters. Each hash structure is a configuration for a single server.

Here's an example:

 
    my %main_site = (
        base_url   => 'http://example.com',
        same_hosts => 'www.example.com',
        email      => 'admin@example.com',
    );

 
    my %news_site = (
        base_url   => 'http://news.example.com',
        email      => 'admin@example.com',
    );

 
    @servers = ( \%main_site, \%news_site );
    1;

The above defines two Perl hashes (%main_site and %news_site) and then places a *reference* (the backslash before the name of the hash) to each of those hashes in the @servers array. The "1;" at the end is required at the end of the file (Perl must see a true value at the end of the file).

The config file path is the first parameter passed to the spider script.

 
    ./spider.pl F<config>

If you do not specify a config file then the spider will look for the file SwishSpiderConfig.pl in the current directory.

The Swish-e distribution includes a SwishSpiderConfig.pl file with a few example configurations. This example file is installed in the prog-bin/ documentation directory (on unix often this is /usr/local/share/swish-e/prog-bin).

When the special config file name "default" is used:

 
    SwishProgParameters default http://www.mysite/index.html [<URL>] [...]

Then a default set of parameters are used with the spider. This is a good way to start using the spider before attempting to create a configuration file.

The default settings skip any urls that look like images (well, .gif .jpeg .png), and attempts to filter PDF and MS Word documents IF you have the required filter programs installed (which are not part of the Swish-e distribution). The spider will follow "a" and "frame" type of links only.

Note that if you do use a spider configuration file that the default configuration will NOT be used (unless you set the "use_default_config" option in your config file).

[ TOC ]


CONFIGURATION OPTIONS

This describes the required and optional keys in the server configuration hash, in random order...

base_url

This required setting is the starting URL for spidering.

This sets the first URL the spider will fetch. It does NOT limit spidering to URLs at or below the level of the directory specified in this setting. For that feature you need to use the test_url callback function.

Typically, you will just list one URL for the base_url. You may specify more than one URL as a reference to a list and each will be spidered:

 
    base_url => [qw! http://swish-e.org/ http://othersite.org/other/index.html !],

but each site will use the same config opions. If you want to index two separate sites you will likely rather add an additional configuration to the @servers array.

You may specify a username and password:

 
    base_url => 'http://user:pass@swish-e.org/index.html',

If a URL is protected by Basic Authentication you will be prompted for a username and password. The parameter max_wait_time controls how long to wait for user entry before skipping the current URL. See also credentials below.

same_hosts

This optional key sets equivalent authority name(s) for the site you are spidering. For example, if your site is www.mysite.edu but also can be reached by mysite.edu (with or without www) and also web.mysite.edu then:

Example:

 
    $serverA{base_url} = 'http://www.mysite.edu/index.html';
    $serverA{same_hosts} = ['mysite.edu', 'web.mysite.edu'];

Now, if a link is found while spidering of:

 
    http://web.mysite.edu/path/to/file.html

it will be considered on the same site, and will actually spidered and indexed as:

 
    http://www.mysite.edu/path/to/file.html

Note: This should probably be called same_host_port because it compares the URI host:port against the list of host names in same_hosts. So, if you specify a port name in you will want to specify the port name in the the list of hosts in same_hosts:

 
    my %serverA = (
        base_url    => 'http://sunsite.berkeley.edu:4444/',
        same_hosts  => [ qw/www.sunsite.berkeley.edu:4444/ ],
        email       => 'my@email.address',
    );

email

This required key sets the email address for the spider. Set this to your email address.

agent

This optional key sets the name of the spider.

link_tags

This optional tag is a reference to an array of tags. Only links found in these tags will be extracted. The default is to only extract links from >a< tags.

For example, to extract tags from a tags and from frame tags:

 
    my %serverA = (
        base_url    => 'http://sunsite.berkeley.edu:4444/',
        same_hosts  => [ qw/www.sunsite.berkeley.edu:4444/ ],
        email       => 'my@email.address',
        link_tags   => [qw/ a frame /],
    );

use_default_config

This option is new for Swish-e 2.4.3.

The spider has a hard-coded default configuration that's available when the spider is run with the configuration file listed as "default":

 
    ./spider.pl default <url>

This default configuration skips urls that match the regular expression:

 
    /\.(?:gif|jpeg|png)$/i

and the spider will attempt to use the SWISH::Filter module for filtering non-text documents. (You still need to install programs to do the actual filtering, though).

Here's the basic config for the "default" mode:

 
    @servers = (
    {
        email               => 'swish@user.failed.to.set.email.invalid',
        link_tags           => [qw/ a frame /],
        keep_alive          => 1,
        test_url            => sub {  $_[0]->path !~ /\.(?:gif|jpeg|png)$/i },
        test_response       => $response_sub,
        use_head_requests   => 1,  # Due to the response sub
        filter_content      => $filter_sub,
    } );

The filter_content callback will be used if SWISH::Filter was loaded and ready to use. This doesn't mean that filtering will work automatically -- you will likely need to install aditional programs for filtering (like Xpdf or Catdoc).

The test_response callback will be set to test if a given content type can be filtered by SWISH::Filter (if SWISH::Filter was loaded), otherwise, it will check for content-type of text/* -- any text type of document.

Normally, if you specify your own config file:

 
    ./spider.pl my_own_spider.config

then you must setup those features available in the default setting in your own config file. But, if you wish to build upon the "default" config file then set this option.

For example, to use the default config but specify your own email address:

 
    @servers = (
        {
            email               => my@email.address,
            use_default_config  => 1,
            delay_sec           => 0,
        },
    );
    1;

What this does is "merge" your config file with the default config file.

delay_sec

This optional key sets the delay in seconds to wait between requests. See the LWP::RobotUA man page for more information. The default is 5 seconds. Set to zero for no delay.

When using the keep_alive feature (recommended) the delay will be used only where the previous request returned a "Connection: closed" header.

delay_min (deprecated)

Set the delay to wait between requests in minutes. If both delay_sec and delay_min are defined, delay_sec will be used.

max_wait_time

This setting is the number of seconds to wait for data to be returned from the request. Data is returned in chunks to the spider, and the timer is reset each time a new chunk is reported. Therefore, documents (requests) that take longer than this setting should not be aborted as long as some data is received every max_wait_time seconds. The default it 30 seconds.

NOTE: This option has no effect on Windows.

max_time

This optional key will set the max minutes to spider. Spidering for this host will stop after max_time minutes, and move on to the next server, if any. The default is to not limit by time.

max_files

This optional key sets the max number of files to spider before aborting. The default is to not limit by number of files. This is the number of requests made to the remote server, not the total number of files to index (see max_indexed). This count is displayted at the end of indexing as Unique URLs.

This feature can (and perhaps should) be use when spidering a web site where dynamic content may generate unique URLs to prevent run-away spidering.

max_indexed

This optional key sets the max number of files that will be indexed. The default is to not limit. This is the number of files sent to swish for indexing (and is reported by Total Docs when spidering ends).

max_size

This optional key sets the max size of a file read from the web server. This defaults to 5,000,000 bytes. If the size is exceeded the resource is skipped and a message is written to STDERR if the DEBUG_SKIPPED debug flag is set.

Set max_size to zero for unlimited size. If the server returns a Content-Length header then that will be used. Otherwise, the document will be checked for size limitation as it arrives. That's a good reason to have your server send Content-Length headers.

See also use_head_requests below.

keep_alive

This optional parameter will enable keep alive requests. This can dramatically speed up spidering and reduce the load on server being spidered. The default is to not use keep alives, although enabling it will probably be the right thing to do.

To get the most out of keep alives, you may want to set up your web server to allow a lot of requests per single connection (i.e MaxKeepAliveRequests on Apache). Apache's default is 100, which should be good.

When a connection is not closed the spider does not wait the "delay_sec" time when making the next request. In other words, there is no delay in requesting documents while the connection is open.

Note: try to filter as many documents as possible before making the request to the server. In other words, use test_url to look for files ending in .html instead of using test_response to look for a content type of text/html if possible. Do note that aborting a request from test_response will break the current keep alive connection.

Note: you must have at least libwww-perl-5.53_90 installed to use this feature.

use_head_requests

This option is new as of swish-e 2.4.3 and can effect the speed of spidering and the load of the web server.

To understand this you will likely need to read about the CALLBACK FUNCTIONS below -- specifically about the test_response callback function. This option is also only used when keep_alive is also enabled (although it could be debated that it's useful without keep alives).

This option tells the spider to use http HEAD requests before each request.

Normally, the spider simply does a GET request and after receiving the first chunk of data back from the web server calls the test_response callback function (if one is defined in your config file). The test_response callback function is a good place to test the content-type header returned from the server and reject types that you do not want to index.

Now, *if* you are using the keep_alive feature then rejecting a document will often (always?) break the keep alive connection.

So, what the use_head_requests option does is issue a HEAD request for every document, checks for a Content-Length header (to check if the document is larger than max_size, and then calls your test_response callback function. If your callback function returns true then a GET request is used to fetch the document.

The idea is that by using HEAD requests instead of GET request a false return from your test_response callback function (i.e. rejecting the document) will not break the keep alive connection.

Now, don't get too excited about this. Before using this think about the ratio of rejected documents to accepted documents. If you reject no documents then using this feature will double the number of requests to the web server -- which will also double the number of connections to the web server. But, if you reject a large percentage of documents then this feature will help maximize the number of keep alive requests to the server (i.e. reduce the number of separate connections needed).

There's also another problem with using HEAD requests. Some broken servers may not respond correctly to HEAD requests (some issues a 500 error), but respond fine to a normal GET request. This is something to watch out for.

Finally, if you do not have a test_response callback AND max_size is set to zero then setting use_head_requests will have no effect.

And, with all other factors involved you might find this option has no effect at all.

skip

This optional key can be used to skip the current server. It's only purpose is to make it easy to disable a specific server hash in a configuration file.

debug

Set this item to a comma-separated list of debugging options.

Options are currently:

 
    errors, failed, headers, info, links, redirect, skipped, url

Here are basically the levels:

 
    errors      =>   general program errors (not used at this time)
    url         =>   print out every URL processes
    headers     =>   prints the response headers
    failed      =>   failed to return a 200
    skipped     =>   didn't index for some reason
    info        =>   a little more verbose
    links       =>   prints links as they are extracted
    redirect    =>   prints out redirected URLs

Debugging can be also be set by an environment variable SPIDER_DEBUG when running spider.pl. You can specify any of the above debugging options, separated by a comma.

For example with Bourne type shell:

 
    SPIDER_DEBUG=url,links spider.pl [....]

Before Swish-e 2.4.3 you had to use the internal debugging constants or'ed together like so:

 
    debug => DEBUG_URL | DEBUG_FAILED | DEBUG_SKIPPED,

You can still do this, but the string version is easier. In fact, if you want to turn on debugging dynamically (for example in a test_url() callback function) then you currently *must* use the DEBUG_* constants. The string is converted to a number only at the start of spiderig -- after that the debug parameter is converted to a number.

quiet

If this is true then normal, non-error messages will be supressed. Quiet mode can also be set by setting the environment variable SPIDER_QUIET to any true value.

 
    SPIDER_QUIET=1

max_depth

The max_depth parameter can be used to limit how deeply to recurse a web site. The depth is just a count of levels of web pages decended, and not related to the number of path elements in a URL.

A max_depth of zero says to only spider the page listed as the base_url. A max_depth of one will spider the base_url page, plus all links on that page, and no more. The default is to spider all pages.

ignore_robots_file

If this is set to true then the robots.txt file will not be checked when spidering this server. Don't use this option unless you know what you are doing.

use_cookies

If this is set then a "cookie jar" will be maintained while spidering. Some (poorly written ;) sites require cookies to be enabled on clients.

This requires the HTTP::Cookies module.

use_md5

If this setting is true, then a MD5 digest "fingerprint" will be made from the content of every spidered document. This digest number will be used as a hash key to prevent indexing the same content more than once. This is helpful if different URLs generate the same content.

Obvious example is these two documents will only be indexed one time:

 
    http://localhost/path/to/index.html
    http://localhost/path/to/

This option requires the Digest::MD5 module. Spidering with this option might be a tiny bit slower.

validate_links

Just a hack. If you set this true the spider will do HEAD requests all links (e.g. off-site links), just to make sure that all your links work.

credentials

You may specify a username and password to be used automatically when spidering:

 
    credentials => 'username:password',

A username and password supplied in a URL will override this setting. This username and password will be used for every request.

See also the get_password callback function below. get_password, if defined, will be called when a page requires authorization.

credential_timeout

Sets the number of seconds to wait for user input when prompted for a username or password. The default is 30 seconds.

Set this to zero to wait forever. Probably not a good idea.

Set to undef to disable asking for a password.

 
    credential_timeout => undef,

remove_leading_dots

Removes leading dots from URLs that might reference documents above the document root. The default is to not remove the dots.

[ TOC ]


CALLBACK FUNCTIONS

Callback functions can be defined in your parameter hash. These optional settings are callback subroutines that are called while processing URLs.

A little perl discussion is in order:

In perl, a scalar variable can contain a reference to a subroutine. The config example above shows that the configuration parameters are stored in a perl hash.

 
    my %serverA = (
        base_url    => 'http://sunsite.berkeley.edu:4444/',
        same_hosts  => [ qw/www.sunsite.berkeley.edu:4444/ ],
        email       => 'my@email.address',
        link_tags   => [qw/ a frame /],
    );

There's two ways to add a reference to a subroutine to this hash:

sub foo { return 1; }

 
    my %serverA = (
        base_url    => 'http://sunsite.berkeley.edu:4444/',
        same_hosts  => [ qw/www.sunsite.berkeley.edu:4444/ ],
        email       => 'my@email.address',
        link_tags   => [qw/ a frame /],
        test_url    => \&foo,  # a reference to a named subroutine
    );

Or the subroutine can be coded right in place:

 
    my %serverA = (
        base_url    => 'http://sunsite.berkeley.edu:4444/',
        same_hosts  => [ qw/www.sunsite.berkeley.edu:4444/ ],
        email       => 'my@email.address',
        link_tags   => [qw/ a frame /],
        test_url    => sub { reutrn 1; },
    );

The above example is not very useful as it just creates a user callback function that always returns a true value (the number 1). But, it's just an example.

The function calls are wrapped in an eval, so calling die (or doing something that dies) will just cause that URL to be skipped. If you really want to stop processing you need to set $server->{abort} in your subroutine (or send a kill -HUP to the spider).

The first two parameters passed are a URI object (to have access to the current URL), and a reference to the current server hash. The server hash is just a global hash for holding data, and useful for setting flags as described below.

Other parameters may be also passed in depending the the callback function, as described below. In perl parameters are passed in an array called "@_". The first element (first parameter) of that array is $_[0], and the second is $_[1], and so on. Depending on how complicated your function is you may wish to shift your parameters off of the @_ list to make working with them easier. See the examples below.

To make use of these routines you need to understand when they are called, and what changes you can make in your routines. Each routine deals with a given step, and returning false from your routine will stop processing for the current URL.

test_url

test_url allows you to skip processing of urls based on the url before the request to the server is made. This function is called for the base_url links (links you define in the spider configuration file) and for every link extracted from a fetched web page.

This function is a good place to skip links that you are not interested in following. For example, if you know there's no point in requesting images then you can exclude them like:

 
    test_url => sub {
        my $uri = shift;
        return 0 if $uri->path =~ /\.(gif|jpeg|png)$/;
        return 1;
    },

Or to write it another way:

 
    test_url => sub { $_[0]->path !~ /\.(gif|jpeg|png)$/ },

Another feature would be if you were using a web server where path names are NOT case sensitive (e.g. Windows). You can normalize all links in this situation using something like

 
    test_url => sub {
        my $uri = shift;
        return 0 if $uri->path =~ /\.(gif|jpeg|png)$/;

 
        $uri->path( lc $uri->path ); # make all path names lowercase
        return 1;
    },

The important thing about test_url (compared to the other callback functions) is that it is called while extracting links, not while actually fetching that page from the web server. Returning false from test_url simple says to not add the URL to the list of links to spider.

You may set a flag in the server hash (second parameter) to tell the spider to abort processing.

 
    test_url => sub {
        my $server = $_[1];
        $server->{abort}++ if $_[0]->path =~ /foo\.html/;
        return 1;
    },

You cannot use the server flags:

 
    no_contents
    no_index
    no_spider

This is discussed below.

test_response

This function allows you to filter based on the response from the remote server (such as by content-type).

Web servers use a Content-Type: header to define the type of data returned from the server. On a web server you could have a .jpeg file be a web page -- file extensions may not always indicate the type of the file.

If you enable use_head_requests then this function is called after the spider makes a HEAD request. Otherwise, this function is called while the web pages is being fetched from the remote server, typically after just enought data has been returned to read the response from the web server.

The test_response callback function is called with the following parameters:

 
    ( $uri, $server, $response, $content_chunk )

The $response variable is a HTTP::Response object and provies methods of examining the server's response. The $content_chunk is the first chunk of data returned from the server (if not a HEAD request).

When not using use_head_requests the spider requests a document in "chunks" of 4096 bytes. 4096 is only a suggestion of how many bytes to return in each chunk. The test_response routine is called when the first chunk is received only. This allows ignoring (aborting) reading of a very large file, for example, without having to read the entire file. Although not much use, a reference to this chunk is passed as the forth parameter.

If you are spidering a site with many different types of content that you do not wish to index (and cannot use a test_url callback to determine what docs to skip) then you will see better performance using both the use_head_requests and keep_alive features. (Aborting a GET request kills the keep-alive session.)

For example, to only index true HTML (text/html) pages:

 
    test_response => sub {
        my $content_type = $_[2]->content_type;
        return $content_type =~ m!text/html!;
    },

You can also set flags in the server hash (the second parameter) to control indexing:

 
    no_contents -- index only the title (or file name), and not the contents
    no_index    -- do not index this file, but continue to spider if HTML
    no_spider   -- index, but do not spider this file for links to follow
    abort       -- stop spidering any more files

For example, to avoid index the contents of "private.html", yet still follow any links in that file:

 
    test_response => sub {
        my $server = $_[1];
        $server->{no_index}++ if $_[0]->path =~ /private\.html$/;
        return 1;
    },

Note: Do not modify the URI object in this call back function.

filter_content

This callback function is called right before sending the content to swish. Like the other callback function, returning false will cause the URL to be skipped. Setting the abort server flag and returning false will abort spidering.

You can also set the no_contents flag.

This callback function is passed four parameters. The URI object, server hash, the HTTP::Response object, and a reference to the content.

You can modify the content as needed. For example you might not like upper case:

 
    filter_content => sub {
        my $content_ref = $_[3];

 
        $$content_ref = lc $$content_ref;
        return 1;
    },

I more reasonable example would be converting PDF or MS Word documents for parsing by swish. Examples of this are provided in the prog-bin directory of the swish-e distribution.

You may also modify the URI object to change the path name passed to swish for indexing.

 
    filter_content => sub {
        my $uri = $_[0];
        $uri->host('www.other.host') ;
        return 1;
    },

Swish-e's ReplaceRules feature can also be used for modifying the path name indexed.

Note: Swish-e now includes a method of filtering based on the SWISH::Filter Perl modules. See the SwishSpiderConfig.pl file for an example how to use SWISH::Filter in a filter_content callback function.

If you use the "default" configuration (i.e. pass "default" as the first parameter to the spider) then SWISH::Filter is used automatically. This only adds code for calling the programs to filter your content -- you still need to install applications that do the hard work (like xpdf for pdf conversion and catdoc for MS Word conversion).

The a function included in the spider.pl for calling SWISH::Filter when using the "default" config can also be used in your config file. There's a function called swish_filter() that returns a list of two subroutines. So in your config you could do:

 
    my ($filter_sub, $response_sub ) = swish_filter();

 
    @server = ( {
        test_response   => $response_sub,
        filter_content  => $filter_sub,
        [...],
    } );

The $response_sub is not required, but is useful if using HEAD requests (use_head_requests): It tests the content type from the server to see if there's any filters that can handle the document. The $filter_sub does all the work of filtering a document.

Make sense? If not, then that's what the Swish-e list is for.

spider_done

This callback is called after processing a server (after each server listed in the @servers array if more than one).

This allows your config file to do any cleanup work after processing. For example, if you were keeping counts during, say, a test_response() callback function you could use the spider_done() callback to print the results.

output_function

If defined, this callback function is called instead of printing the content and header to STDOUT. This can be used if you want to store the output of the spider before indexing.

The output_function is called with the following parameters:

 
   ($server, $content, $uri, $response, $bytecount, $path);

Here is an example that simply shows two of the params passed:

 
    output_function => sub {
        my ($server, $content, $uri, $response, $bytecount, $path) = @_;
        print STDERR  "passed: uri $uri, bytecount $bytecount...\n";
        # no output to STDOUT for swish-e
    }

You can do almost the same thing with a filter_content callback.

get_password

This callback is called when a HTTP password is needed (i.e. after the server returns a 401 error). The function can test the URI and Realm and then return a username and password separated by a colon:

 
    get_password => sub {
        my ( $uri, $server, $response, $realm ) = @_;
        if ( $uri->path =~ m!^/path/to/protected! && $realm eq 'private' ) {
            return 'joe:secret931password';
        }
        return;  # sorry, I don't know the password.
    },

Use the credentials setting if you know the username and password and they will be the same for every request. That is, for a site-wide password.

Note that you can create your own counters to display in the summary list when spidering is finished by adding a value to the hash pointed to by $server-{counts}>.

 
    test_url => sub {
        my $server = $_[1];
        $server->{no_index}++ if $_[0]->path =~ /private\.html$/;
        $server->{counts}{'Private Files'}++;
        return 1;
    },

Each callback function must return true to continue processing the URL. Returning false will cause processing of the current URL to be skipped.

[ TOC ]


More on setting flags

Swish (not this spider) has a configuration directive NoContents that will instruct swish to index only the title (or file name), and not the contents. This is often used when indexing binary files such as image files, but can also be used with html files to index only the document titles.

As shown above, you can turn this feature on for specific documents by setting a flag in the server hash passed into the test_response or filter_content subroutines. For example, in your configuration file you might have the test_response callback set as:

 
    test_response => sub {
        my ( $uri, $server, $response ) = @_;
        # tell swish not to index the contents if this is of type image
        $server->{no_contents} = $response->content_type =~ m[^image/];
        return 1;  # ok to index and spider this document
    }

The entire contents of the resource is still read from the web server, and passed on to swish, but swish will also be passed a No-Contents header which tells swish to enable the NoContents feature for this document only.

Note: Swish will index the path name only when NoContents is set, unless the document's type (as set by the swish configuration settings IndexContents or DefaultContents) is HTML and a title is found in the html document.

Note: In most cases you probably would not want to send a large binary file to swish, just to be ignored. Therefore, it would be smart to use a filter_content callback routine to replace the contents with single character (you cannot use the empty string at this time).

A similar flag may be set to prevent indexing a document at all, but still allow spidering. In general, if you want completely skip spidering a file you return false from one of the callback routines (test_url, test_response, or filter_content). Returning false from any of those three callbacks will stop processing of that file, and the file will not be spidered.

But there may be some cases where you still want to spider (extract links) yet, not index the file. An example might be where you wish to index only PDF files, but you still need to spider all HTML files to find the links to the PDF files.

 
    $server{test_response} = sub {
        my ( $uri, $server, $response ) = @_;
        $server->{no_index} = $response->content_type ne 'application/pdf';
        return 1;  # ok to spider, but don't index
    }

So, the difference between no_contents and no_index is that no_contents will still index the file name, just not the contents. no_index will still spider the file (if it's text/html) but the file will not be processed by swish at all.

Note: If no_index is set in a test_response callback function then the document will not be filtered. That is, your filter_content callback function will not be called.

The no_spider flag can be set to avoid spiderering an HTML file. The file will still be indexed unless no_index is also set. But if you do not want to index and spider, then simply return false from one of the three callback funtions.

[ TOC ]


SIGNALS

Sending a SIGHUP to the running spider will cause it to stop spidering. This is a good way to abort spidering, but let swish index the documents retrieved so far.

[ TOC ]


CHANGES

List of some of the changes

[ TOC ]


Thu Sep 30 2004 - changes for Swish-e 2.4.3

Code reorganization and a few new featues. Updated docs a little tiny bit. Introduced a few spelling mistakes.

Config opiton: use_default_config

It used to be that you could run the spider like:

 
    spider.pl default <some url>

and the spider would use its own internal config. But if you used your own config file then the defaults were not used. This options allows you to merge your config with the default config. Makes making small changes to the default easy.

Config option: use_head_requests

Tells the spider to make a HEAD request before GET'ing the document from the web server. Useful if you use keep_alive and have a test_response() callback that rejects many documents (which breaks the connection).

Config option: spider_done

Callback to tell you (or tell your config as it may be) that the spider is done. Useful if you need to do some extra processing when done spidering -- like record counts to a file.

Config option: get_password

This callback is called when a document returns a 401 error needing a username and password. Useful if spidering a site proteced with multiple passwords.

Config option: output_function

If defined spider.pl calls this instead of sending ouptut to STDOUT.

Config option: debug

Now you can use the words instead of or'ing the DEBUG_* constants together.

[ TOC ]


TODO

Add a "get_document" callback that is called right before making the "GET" request. This would make it easier to use cached documents. You can do that now in a test_url callback or in a test_response when using HEAD request.

Save state of the spider on SIGHUP so spidering could be restored at a later date.

[ TOC ]


COPYRIGHT

Copyright 2001 Bill Moseley

This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.

[ TOC ]


SUPPORT

Send all questions to the The SWISH-E discussion list.

See http://sunsite.berkeley.edu/SWISH-E. [ TOC ]