With some Web services URL canonicalization has a downside. What works great for major search engines like Google can fire back when a Web service like Yahoo thinks circumcising URLs is cool. Proper URL canonicalization might, for example, screw your blog’s reputation at Technorati.
In fact the problem is not your URL canonicalization, e.g. 301 redirects from http://example.com to http://example.com/ respectively http://example.com/directory to http://example.com/directory/, but crappy software that removes trailing forward slashes from your URLs.
Dear Web developers, if you really think that home page locations respectively directory URLs look way cooler without the trailing slash, then by all means manipulate the anchor text, but do not manipulate HREF values, and do not store truncated URLs in your databases (not that “http://example.com” as anchor text makes any sense when the URL in HREF points to “http://example.com/”). Spreading invalid URLs is not funny. People as well as Web robots take invalid URLs from your pages for various purposes. Many usages of invalid URLs are capable to damage the search engine rankings of the link destinations. You can’t control that, hence don’t screw our URLs. Never. Period.
Folks who don’t agree with the above said read on.
TOC:
- What is a trailing slash? About URLs, directory URIs, default documents, directory indexes, …
- How to rescue stolen trailing slashes About Apache’s handling of directory requests, and rewriting respectively redirecting invalid directory URIs in .htaccess as well as in PHP scripts.
- Why stealing trailing slashes is not cool Truncating slashes is not only plain robbery (bandwidth theft), it often causes malfunctions at the destination server and 3rd party services as well.
- How URL canonicalization irritates Technorati 301 redirects that “add” a trailing slash to directory URLs, respectively virtual URIs that mimic directories, seem to irritate Technorati so much that it can’t compute reputation, recent post lists, and so on.
What is a trailing slash?
The Web’s standards say (links and full quotes): The trailing path segment delimiter “/” represents an empty last path segment. Normalization should not remove delimiters when their associated component is empty. (Read the polite “should” as “must”.)
To understand that, lets look at the most common URL components:
scheme:// server-name.tld /path ?query-string #fragment
The (red) path part begins with a forward slash “/” and must consist of at least one byte (the trailing slash itself in case of the home page URL http://example.com/).
If an URL ends with a slash, it points to a directory’s default document, or, if there’s no default document, to a list of objects stored in a directory. The home page link lacks a directory name, because “/” after the TLD (.com|net|org|…) stands for the root directory.
Automated directory indexes (a list of links to all files) should be forbidden, use Options -Indexes in .htaccess to send such requests to your 403-Forbidden page.
In order to set default file names and their search sequence for your directories use DirectoryIndex index.html index.htm index.php /error_handler/missing_directory_index_doc.php. In this example: on request of http://example.com/directory/ Apache will first look for /directory/index.html, then if that doesn’t exist for /directory/index.htm, then /directory/index.php, and if all that fails, it will serve an error page (that should log such requests so that the Webmaster can upload the missing default document to /directory/).
The URL http://example.com (without the trailing slash) is invalid, and there’s no specification telling a reason why a Web server should respond to it with meaningful contents. Actually, the location http://example.com points to Null (nil, zilch, nada, zip, nothing), hence the correct response is “404 - we haven’t got ‘nothing to serve’ yet”.
The same goes for sub-directories. If there’s no file named “/dir”, the URL http://example.com/dir points to Null too. If you’ve a directory named “/dir”, the canonical URL http://example.com/dir/ either points to a directory index page (an autogenerated list of all files) or the directory’s default document “index.(html|htm|shtml|php|…)”. A request of http://example.com/dir –without the trailing slash that tells the Web server that the request is for a directory’s index– resolves to “not found”.
You must not reference a default document by its name! If you’ve links like http://example.com/index.html you can’t change the underlying technology without serious hassles. Say you’ve a static site with a file structure like /index.html, /contact/index.html, /about/index.html and so on. Tomorrow you’ll realize that static stuff sucks, hence you’ll develop a dynamic site with PHP. You’ll end up with new files: /index.php, /contact/index.php, /about/index.php and so on. If you’ve coded your internal links as http://example.com/contact/ etc. they’ll still work, without redirects from .html to .php. Just change the DirectoryIndex directive from “… index.html … index.php …” to “… index.php … index.html …”. (Of course you can configure Apache to parse .html files for PHP code, but that’s another story.)
It seems that truncating default document names can make sense for services that deal with URLs, but watch out for sites that serve different contents under various extensions of “index” files (intentionally or not). I’d say that folks submitting their ugly index.html files to directories, search engines, top lists and whatnot deserve all the hassles that come with later changes.
How to rescue stolen trailing slashes
Since Web servers know that users are faulty by design, they jump through a couple of resource burning hoops in order to either add the trailing slash so that relative references inside HTML documents (CSS/JS/feed links, image locations, HREF values …) work correctly, or apply voodoo to accomplish that without (visibly) changing the address bar.
With Apache, DirectorySlash On enables this behavior (check whether your Apache version does 301 or 302 redirects, in case of 302s find another solution). You can also rewrite invalid requests in .htaccess when you need special rules:
RewriteEngine on
RewriteBase /content/
RewriteRule ^dir1$ http://example.com/content/dir1/ [R=301,L]
RewriteRule ^dir2$ http://example.com/content/dir2/ [R=301,L]
With content management systems (CMS) that generate virtual URLs on the fly, often there’s no other chance than hacking the software to canonicalize invalid requests. To prevent search engines from indexing invalid URLs that are in fact duplicates of canonical URLs, you’ll perform permanent redirects (301).
Here is a WordPress (header.php) example:
$requestUri = $_SERVER["REQUEST_URI"];
$queryString = $_SERVER["QUERY_STRING"];
$doRedirect = FALSE;
$fileExtensions = array(".html", ".htm", ".php");
$serverName = $_SERVER["SERVER_NAME"];
$canonicalServerName = $serverName;
// if you prefer http://example.com/* URLs remove the "www.":
$srvArr = explode(".", $serverName);
$canonicalServerName = $srvArr[count($srvArr) - 2] ."." .$srvArr[count($srvArr) - 1];
$url = parse_url ("http://" .$canonicalServerName .$requestUri);
$requestUriPath = $url["path"];
if (substr($requestUriPath, -1, 1) != "/") {
$isFile = FALSE;
foreach($fileExtensions as $fileExtension) {
if ( strtolower(substr($requestUriPath, strlen($fileExtension) * -1, strlen($fileExtension))) == strtolower($fileExtension) ) {
$isFile = TRUE;
}
}
if (!$isFile) {
$requestUriPath .= "/";
$doRedirect = TRUE;
}
}
$canonicalUrl = "http://" .$canonicalServerName .$requestUriPath;
if ($queryString) {
$canonicalUrl .= "?" . $queryString;
}
if ($url["fragment"]) {
$canonicalUrl .= "#" . $url["fragment"];
}
if ($doRedirect) {
@header("HTTP/1.1 301 Moved Permanently", TRUE, 301);
@header("Location: $canonicalUrl");
exit;
}
Check your permalink settings and edit the values of $fileExtensions and $canonicalServerName accordingly. For other CMSs adapt the code, perhaps you need to change the handling of query strings and fragments. The code above will not run under IIS, because it has no REQUEST_URI variable.
Why stealing trailing slashes is not cool
This section expressed in one sentence: Cool URLs don’t change, hence changing other people’s URLs is not cool.
Folks should understand the “U” in URL as unique. Each URL addresses one and only one particular resource. Technically spoken, if you change one single character of an URL, the altered URL points to a different resource, or nowhere.
Think of URLs as phone numbers. When you call 555-0100 you reach the switchboard, 555-0101 is the fax, and 555-0109 is the phone extension of somebody. When you steal the last digit, dialing 555-010, you get nowhere.
Only a fool would assert that a phone number shortened by one digit is way cooler than the complete phone number that actually connects somewhere. Well, the last digit of a phone number and the trailing slash of a directory link aren’t much different. If somebody hands out an URL (with trailing slash), then use it as is, or don’t use it at all. Don’t “prettify” it, because any change destroys its serviceability.
If one requests a directory without the trailing slash, most Web servers will just reply to the user agent (brower, screen reader, bot) with a redirect header telling that one must use a trailing slash, then the user agent has to re-issue the request in the formally correct way. From a Webmaster’s perspective, burning resources that thoughtlessly is plain theft. From a user’s perspective, things will often work without the slash, but they’ll be quicker with it. “Often” doesn’t equal “always”:
- Some Web servers will serve the 404 page.
- Some Web servers will serve the wrong content, because /dir is a valid script, virtual URI, or page that has nothing to do with the index of /dir/.
- Many Web servers will respond with a 302 HTTP response code (Found) instead of a correct 301-redirect, so that most search engines discovering the sneakily circumcised URL will index the contents of the canonical URL under the invalid URL. Now all search engine users will request the incomplete URL too, running into unnecessary redirects.
- Some Web servers will serve identical contents for /dir and /dir/, that leads to duplicate content issues with search engines that index both URLs from links. Most Web services that rank URLs will assign different scorings to all known URL variants, instead of accumulated rankings to both URLs (which would be the right thing to do, but is technically, well, challenging).
- Some user agents can’t handle (301) redirects properly. Exotic user agents might serve the user an empty page or the redirect’s “error message”, and Web robots like the crawlers sent out by Technorati or MSN-LiveSearch hang up respectively process garbage.
Does it really make sense to maliciously manipulate URLs just because some clueless developers say “dude, without the slash it looks way cooler”? Nope. Stealing trailing slashes in general as well as storing amputated URLs is a brain dead approach.
KISS (keep it simple, stupid) is a great principle. “Cosmetic corrections” like trimming URLs add unnecessary complexity that leads to erroneous behavior and requires even more code tweaks. GIGO (garbage in, garbage out) is another great principle that applies here. Smart algos don’t change their inputs. As long as the input is processible, they accept it, otherwise they skip it.
Exceptions
URLs in print, radio, and offline in general, should be truncated in a way that browsers can figure out the location - “domain.co.uk” in print and “domain dot co dot uk” on radio is enough. The necessary redirect is cheaper than a visitor who doesn’t type in the canonical URL including scheme, www-prefix, and trailing slash.
How URL canonicalization seems to irritate Technorati
Due to the not exactly responsively (respectively swamped) Technorati user support parts of this section should be interpreted as educated speculation. Also, I didn’t research enough cases to come to a working theory. So here is just the story “how Technorati fails to deal with my blog”.
When I moved my blog from blogspot to this domain, I’ve enhanced the faulty WordPress URL canonicalization. If any user agent requests http://sebastians-pamphlets.com it gets redirected to http://sebastians-pamphlets.com/. Invalid post/page URLs like http://sebastians-pamphlets.com/about redirect to http://sebastians-pamphlets.com/about/. All redirects are permanent, returning the HTTP response code “301″.
I’ve claimed my blog as http://sebastians-pamphlets.com/, but Technorati shows its URL without the trailing slash.
…<div class="url"><a href="http://sebastians-pamphlets.com">http://sebastians-pamphlets.com</a> </div> <a class="image-link" href="/blogs/sebastians-pamphlets.com"><img …
By the way, they forgot dozens of fans (folks who “fave’d” either my old blogspot outlet or this site) too.
I’ve added a description and tons of tags, that both don’t show up on public pages. It seems my tags were deleted, at least they aren’t visible in edit mode any more.
Shortly after the submission, Technorati stopped to adjust the reputation score from newly discovered inbound links. Furthermore, the list of my recent posts became stale, although I’ve pinged Technorati with every update, and technorati received my update notifications via ping services too. And yes, I’ve tried manual pings to no avail.
I’ve gained lots of fresh inbound links, but the authority score didn’t change. So I’ve asked Technorati’s support for help. A few weeks later, in December/2007, I’ve got an answer:
I’ve taken a look at the issue regarding picking up your pings for “sebastians-pamphlets.com”. After making a small adjustment, I’ve sent our spiders to revisit your page and your blog should be indexed successfully from now on.
Please let us know if you experience any problems in the future. Do not hesitate to contact us if you have any other questions.
Indeed, Technorati updated the reputation score from “56″ to “191″, and refreshed the list of posts including the most recent one.
Of course the “small adjustment” didn’t persist (I assume that a batch process stole the trailing slash that the friendly support person has added). I’ve sent a follow-up email asking whether that’s a slash issue or not, but didn’t receive a reply yet. I’m quite sure that Technorati doesn’t follow 301-redirects, so that’s a plausible cause for this bug at least.
Since December 2007 Technorati didn’t update my authority score (just the rank goes up and down depending on the number of inbound links Technorati shows on the reactions page - by the way these numbers are often unreal and change in the range of hundreds from day to day).
It seems Technorati didn’t index my posts since then (December/18/2007), so probably my outgoing links don’t count for their destinations.
(All screenshots were taken on February/05/2008. When you click the Technorati links today, it could hopefully will look differently.)
I’m not amused. I’m curious what would happen when I add
if (!preg_match("/Technorati/i", "$userAgent")) {/* redirect code */}
to my canonicalization routine, but I can resist to handle particular Web robots. My URL canonicalization should be identical both for visitors and crawlers. Technorati should be able to fix this bug without code changes at my end or weeky support requests. Wishful thinking? Maybe.
Update 2008-03-06: Technorati crawls my blog again. The 301 redirects weren’t the issue. I’ll explain that in a follow-up post soon.
Seems I’ve made it on the short-list at the SEMMY 2009 Awards in the Search Tech category. Great ego food. I’m honored. Thanks for nominating me that often! And thanks to John Andrews and Todd Mintz for the kind judgement!
It could help them to tidy up their flawed REP support.
Wow, I won! Thank you all!
Entries 
… you’ve boosted your business Web site’s rankings with shitloads of crappy links. The 11th SEO commandment: Don’t promote your white hat sites with black hat link building methods! It may work for a while, but once you find your butt in Google-jail, there’s no way out. Not even a reconsideration request can help because you can’t provide its prerequisites.
Matt Cutts asks us 
What was meant as a quick test of 
Fabrice Canel from 
My somewhat sarcastic post “
After the great 