Bug #6057: URI - Nonsensical Behavior

• Status: Closed
• Assignee: akira (akira yamada)
• Target version: -
• ruby -v: 1.9.3-p125
• Backport: 2.3: UNKNOWN, 2.4: UNKNOWN, 2.5: UNKNOWN

[sgonyea (Scott Gonyea)]

The behavior of Ruby's URI is confusing, to say the least.

URI("http://foo.com/bar") + "baz"
# => #<URI::HTTP:0x710713de URL:http://foo.com/baz> 

URI.join("http://foo.com/baz", "bar")
# => #<URI::HTTP:0x29fd5dc4 URL:http://foo.com/bar>

I could at least understand this behavior being the case if we were adding "/bar" (although it'd still be non-intuitive). That URI throws away its current path makes it a tedious library. Net::HTTP expects it as an input, yet as you interact with URI, you are constantly massaging it into and out of string form.

In my opinion, URI should model the behavior of Pathname.

[sgonyea (Scott Gonyea)]

Edit: Moronic comment redacted.

[sgonyea (Scott Gonyea)]

I'd be happy to write this improved functionality. But would it be welcome?

[sgonyea (Scott Gonyea)]

Another example:

URI.join("http://foo.com/baz", "bar", "bim")

=> #<URI::HTTP:0x1005412e URL:http://foo.com/bim>

Also, if we did model the behavior of Pathname, then joining of "/baz" would reset the path which I do think is acceptable and good behavior.

[duerst (Martin Dürst)]

I have only done a few quick tests, but it seems that the '+' operator implements relative resolution (see http://tools.ietf.org/html/rfc3986#section-5.2). For URIs, this makes a lot of sense.

[sgonyea (Scott Gonyea)]

Ah. That's the issue.

1.9.3-p125 :016 > uri = URI("http://google.com/foo/bar/")
=> #<URI::HTTP:0x007fd97a891038 URL:http://google.com/foo/bar/>
1.9.3-p125 :017 > uri + "baz"
=> #<URI::HTTP:0x007fd97a887ab0 URL:http://google.com/foo/bar/baz>

If the URI's path itself has a trailing "/" then it will append the new path.

Likewise:

1.9.3-p125 :023 > URI.join("http://google.com/foo/", "bar", "baz")
=> #<URI::HTTP:0x007fd97b8c7308 URL:http://google.com/foo/baz>
1.9.3-p125 :024 > URI.join("http://google.com/foo/", "bar/", "baz")
=> #<URI::HTTP:0x007fd97b8ba928 URL:http://google.com/foo/bar/baz>
1.9.3-p125 :025 >

How very unintuitive :(. I guess the next question would be if URI.join actually needs to follow the merge rules to the letter, but I imagine that argument is moot :).

In my opinion, URI should model the behavior of Pathname.

I totally agree.

Additionally I absolutely dislike using global methods instead of class
factory methods. I think it isn't at all obvious and it poisons the
global namespace.

URI.parse(args) over Kernel#URI(args)

[drbrain (Eric Hodel)]

Changing the behavior of URI#+ from the current will break many libraries (including mechanize) that depend upon the current behavior of relative resolution.

The current behavior is frequently used to combine, for example, an anchor href with the current location to generate a new, correct URI if the href is a relative URI like "foo" or "../foo", a complete path like "/foo" or a complete URI like "http://other.example/bar".

Scott Gonyea [e-mail] writes:

Issue #6057 has been updated by Scott Gonyea.

I'd be happy to write this improved functionality. But would it be welcome?

Is there anyone who actually likes this behavior?

[ko1 (Koichi Sasada)]

• Assignee set to akira (akira yamada)

[shyouhei (Shyouhei Urabe)]

• Status changed from Open to Assigned

[mame (Yusuke Endoh)]

• Description updated (diff)

[mame (Yusuke Endoh)]

• Has duplicate Bug #10075: URI#join needs documentation of its behavior added

[mame (Yusuke Endoh)]

• Related to Bug #15276: URI.join behaviour is unintuitive added

[mame (Yusuke Endoh)]

• Status changed from Assigned to Closed

This ticket was duplicated by Bug #10075: URI#join needs documentation of its behavior, and that ticket has been already closed with documentation improvement. So I'm closing this.