relative port address not updated

[kkvilekval]

furl ('http://domain.com:9999').join ('http://:9021/').url
 'http://:9021/

expected `http://domain.com:9021/'

In [17]: furl ('http://domain.com:9999').join ('https//:9021/').url
Out[17]: 'http://:9021/'

expected `https://domain.com:9021/'

[gruns]

Hi kkvilekval,

While admittedly a bit of a head scratcher initially, this is expected behavior.

The host of 'http://:9021/' is empty. As such, join() adopts the empty
string as the new host.

For context, this is the behavior of Python's standard urllib.parse.urljoin(),
too.

>>> from urllib.parse import urljoin
>>> urljoin('http://domain.com:9999', 'http://:9021/')
'http://:9021/'

And other language's standard libraries, too. Like Node's.

$ node
> url.resolve('http://domain.com:9999/', 'http://:9021/')
'http://:9021/'

Back to your original example. If you want to set the port of an existing URL
to, the easiest way to do that is with set().

>>> from furl import furl
>>> furl('http://domain.com:9999').set(port=furl('http://:9021/').port).url
'http://domain.com:9021'

Or, to preserve the original host, join on a URL with that same host.

>>> from furl import furl
>>> furl('http://domain.com:9999').join('http://domain.com:9021/').url
'http://domain.com:9021'

Does that answer your question?

[kkvilekval]

Furl seemed like a good place to actually have a join that would work properly (well at least similar to how the browser does joins). The other libraries don't really mimic browser behavior well and so I was looking for an alternative.. I ended up coding it externally but thought it might make a good addition:

def urljoin(base, partial, **kw):
    """ Join all components for a url overriding any components in base from partial

    @param base: The base url
    @param partial: The overriding url
    @param kw :  extra url parameters
    @return a new url string
    """
    updated = furl.furl (base)
    partial = furl.furl (partial)

    updated.scheme = partial.scheme or updated.scheme
    updated.username = partial.username or updated.username
    updated.password = partial.username or updated.username
    updated.host     = partial.host or updated.host
    updated.port     = partial.port or updated.port
    updated.path = urlparse.urljoin (str(updated.path), str(partial.path))
    updated.query.params  = partial.query.params or updated.query.params
    updated.query.params.update (kw)
    return updated.url

[gruns]

(well at least similar to how the browser does joins)

Browsers don't join http://domain.com:9999/ and http://:9021/ into
http://domain.com:9021/ either, though. For example, a quick test:

• https://jsbin.com/lufefupota/edit?html,output

In both Chrome and Firefox, the click me anchor doesn't navigate to

• https://jsbin.com:9021/lufefupota/edit?html,output

It doesn't navigate anywhere.

Do you have an example piece of software (application, library, tool, etc) that
joins http://domain.com:9999/ and http://:9021/ into
http://domain.com:9021/?

[gruns]

Little bump -- did you find an example piece of software (application, library, tool, etc) that
joins http://domain.com:9999/ and http://:9021/ into http://domain.com:9021/?

[kkvilekval]

I will close .. after checking the spec on relative urls (https://www.ietf.org/rfc/rfc1808.txt) it appears that netloc is not parsed at all and is replaced completely see (step 3 in section 4).

That said.. I do find the functionality useful especially when dealing with multi-port services such as websockets and others.