URLs Are The Uniform Way to Locate Resources

http://adam.blog.heroku.com/past/2010/3/30/urls_are_the_uniform_way_to_locate_resources/

by Titanous • 16 years ago

54 20 16 years ago

adam.blog.heroku.com

Except URI's aren't uniform. They're used in so many ways as to be confusing. For instance:

"git://github.com/thoughtbot/paperclip.git" vs "smtp://user:pass@hostname/domain" vs "http://news.ycombinator.com/user?id=krainboltgreene vs "chrome://history/"

The user name appears in the path, the username section, and the query params. All of these are pretty "standard" uses. So while they might be called "uniform" the reality is absolutely different.

m0th872• 16 years ago

All of your examples still portray uniformity:

> git://github.com/thoughtbot/paperclip.git

You are accessing a resource that is partially identified by the username.

> smtp://user:pass@hostname/domain

The username and password is used for authentication to access the resource, but is part of its unique identification.

> http://news.ycombinator.com/user?id=krainboltgreene

The mistake actually lies with HN, because part of what should be used to identify the resource is instead a param.

> chrome://history/

This still fits because you are accessing a uniquely identified resource. Just because it is user-dependent does not mean it breaks REST.

blasdel• 16 years ago

URL Parameters ≠ RPC

The resource is the whole string between the domain and the fragment identifier. '/' is no different from '?' or '&' — it's all the same once you drop your false-REST preconceptions.

The HTML spec does define a standard for concatenating the name-value pairs from form elements onto target attributes for use in GET requests, but that doesn't reify it into the definition of HTTP, much less the definition of REST.

jerf2• 16 years ago

"Except URI's aren't uniform."

Sure they are! They're structured data serialized to a binary blob that claims to be a string but contains no encoding indication with guidance given by an internet standard for the contents of the string before the first colon but no guidance whatsoever after that! What's not uniform about that?

Normally I wouldn't be so pedantic but at the point where you're talking about "mysql://myuser:mypass@db8.myhost.com:3306/mydatabase" as if it's some sort of solution to a problem you've really dropped the ball.

URLs are only meaningful in a given semantic context. "http" and "https" are meaningful because we all agree what they mean (see RFCs below). "git" is meaningful-ish because there's only one thing that plausibly can be said to give a definition, but your application does not magically gain any understanding of the subsequent what-might-as-well-be-a-binary-blob merely by virtue of sticking "git:" in front of it. "mysql" is simply meaningless. I use Perl and therefore DBI and I observe that it too has a sort of "mysql URL" but it looks nothing like Ruby's. Uniformity is relatively to a universal agreement about what it means, and mysql lacks that. For that matter, git may very well lack it in the future, if someone implements other gits (and I've seen attempts). Without that universal agreement, you don't have a URL. You can't make a URL by fiat.

For non-standard URLs, what follows the colon... and for that matter what precedes it... is nothing more and nothing less than a binary blob in a constrained character set. And this article should be treated exactly as if it were an article about how all your resource location problems go away if you just express them in terms of opaque binary blobs, because once you leave http and https behind, that is what you are doing. Not "like" what you are doing, it is what you are doing, full stop. It may work for you and it may work for your friends, but that is not by the magic of calling your binary blob a URL, it is by the magic of agreeing to a way to interpret bits, and that's hardly any sort of breakthrough.

(Just to be clear, this is vehement agreement that acknowledges that you got to this point first.)

By the way, since I can already guess that someone will reply with something like m0th87's point, I invite you to read the URL RFC: http://www.ietf.org/rfc/rfc1738.txt But read it carefully, for what it actually mandates. Section 2.2: "Many URL schemes reserve certain characters for a special meaning"... none of them are universal to URLs, they are all scheme-specific, which means you can't trust their meaning in undefined schemes. Section 2.3: "Some URL schemes ... contain names that can be considered hierarchial"... / doesn't have a universal meaning, it's relative to the scheme. 3.1 describes the double-slash, which I can now say I've seen used incorrectly in both directions. Section 3.5 defining "mailto:" observes that URLs aren't even necessarily resources. (Section 3.10, file URLs defined in a way that violates the earlier discussion of double-slash. I understand why, but it's still a violation.)

And if you want to talk URI (http://www.ietf.org/rfc/rfc2396.txt ), section 3 starts right off with "The URI syntax is dependent upon the scheme."

derefr2• 16 years ago

So, what you're saying, basically, is that URLs confer no special advantage over URNs—which are specified to just be binary blobs with a schema[/namespace] identifier attached.

I don't agree with this. URLs, in practice, are a standardized format, predicated mostly on how HTTP has handled them. Any active, well-known URL schema will use

    schema://username:password@host:port/resource/path?query=parameters&more=with%20percent%20encoding#and-fragment-identifier

as that is what we consider to be a URL, no matter what the RFC says. And that format is useful for encoding a great many things. Just because some libraries have chosen to create things that resemble URLs (such as MySQL, as you mentioned), does not mean that they are URLs as the term is descriptively, not prescriptively, defined.

sjs3• 16 years ago

Of course it depends on the protocol and/or application. That's the way it is meant to be and why they will be useful for a long time.

However, a popular convention is to use protocol://[username[:password]@]host[:port]/path/components/separated/by/slashes?with=query&params=like&this and in practice many URL schemes use this format or a subset of it.

sjs3• 16 years ago

They are all protocol://host/path?params. Looks pretty uniform to me.

mlLK• 16 years ago

I agree for the most part. Even Tim Berners-Lee has his regrets formulating the semantics of the URL, http://en.wikipedia.org/wiki/Uniform_Resource_Locator#Histor...

protocol:TLD/domain/www/path/file.html just makes more sense.

richcollins• 16 years ago

I don't see why the json format that he mentioned couldn't easily be made uniform through standard keys (protocol, port, host, path ... etc)

sjs3• 16 years ago

We could do that but I can't think of any benefits. Not to say there aren't any, but the obvious negatives would likely outweigh them 100:1.

andrewtj• 16 years ago

It could — I'm doing something kind of similar with a DNS service I'm building. It has an HTTP interface which amongst other things exposes DNS-SD services (which include host, protocol, port and service-specific key-value pairs).

earle• 16 years ago

NO SHIT! This is the top link on Hacker News? Is this a fucking joke?

Why are -incorrect- cursory coverage of baseline RFCs Hacker News worthy?

psadauskas• 16 years ago

Because its become very apparent how few web-developers understand "web". The more articles like this that point them in the right direction, the better.

dutchflyboy• 16 years ago

I agree, this article is really useless. I mean, just look up what the acrynom means: http://en.wikipedia.org/wiki/Uniform_Resource_Locator

It's the definition!

pak• 16 years ago

I never liked the idea of putting passwords into URLs... it just gives people the wrong idea about how they should handle their password.

To me, URLs and passwords are orthogonal. One says "this is where it is", the other says "let me in please".

benkant• 16 years ago

Surprised you didn't know.

URLs Are The Uniform Way to Locate Resources

Related Stories