In urltools: Vectorised Tools for URL Handling and Parsing

Elegant URL handling with urltools

URLs are treated, by base R, as nothing more than components of a data retrieval process: they exist to create connections to retrieve datasets. This is an essential feature for the language to have, but it also means that URL handlers are designed for situations where URLs get you to the data - not situations where URLs are the data.

There is no support for encoding or decoding URLs en-masse, and no support for parsing and interpreting them. urltools provides this support!

URL encoding and decoding

Base R provides two functions - URLdecode and URLencode - for taking percentage-encoded URLs and turning them into regular strings, or vice versa. As discussed, these are primarily designed to enable connections, and so they have several inherent limitations, including a lack of vectorisation, that make them unsuitable for large datasets.

Not only are they not vectorised, they also have several particularly idiosyncratic bugs and limitations: URLdecode, for example, breaks if the decoded value is out of range:

URLdecode("test%gIL")
Error in rawToChar(out) : embedded nul in string: '\0L'
In addition: Warning message:
In URLdecode("%gIL") : out-of-range values treated as 0 in coercion to raw

URLencode, on the other hand, encodes slashes on its most strict setting - without paying attention to where those slashes are: if we attempt to URLencode an entire URL, we get:

URLencode("https://en.wikipedia.org/wiki/Article", reserved = TRUE)
[1] "https%3a%2f%2fen.wikipedia.org%2fwiki%2fArticle"

That's a completely unusable URL (or ewRL, if you will).

urltools replaces both functions with url_decode and url_encode respectively:

library(urltools)
url_decode("test%gIL")
[1] "test"
url_encode("https://en.wikipedia.org/wiki/Article")
[1] "https://en.wikipedia.org%2fwiki%2fArticle"

As you can see, url_decode simply excludes out-of-range characters from consideration, while url_encode detects characters that make up part of the URLs scheme, and leaves them unencoded. Both are extremely fast; with urltools, you can decode a vector of 1,000,000 URLs in 0.9 seconds.

Alongside these, we have functions for encoding and decoding the 'punycode' format of URLs - ones that are designed to be internationalised and have unicode characters in them. These also take one argument, a vector of URLs, and can be found at puny_encode and puny_decode respectively.

URL parsing

Once you've got your nicely decoded (or encoded) URLs, it's time to do something with them - and, most of the time, you won't actually care about most of the URL. You'll want to look at the scheme, or the domain, or the path, but not the entire thing as one string.

The solution is url_parse, which takes a URL and breaks it out into its RfC 3986 components: scheme, domain, port, path, query string and fragment identifier. This is, again, fully vectorised, and can happily be run over hundreds of thousands of URLs, rapidly processing them. The results are provided as a data.frame, since most people use data.frames to store data.

> parsed_address <- url_parse("https://en.wikipedia.org/wiki/Article")
> str(parsed_address)
'data.frame':   1 obs. of  6 variables:
 $ scheme   : chr "https"
 $ domain   : chr "en.wikipedia.org"
 $ port     : chr NA
 $ path     : chr "wiki/Article"
 $ parameter: chr NA
 $ fragment : chr NA

We can also perform the opposite of this operation with url_compose:

> url_compose(parsed_address)
[1] "https://en.wikipedia.org/wiki/article"

Getting/setting URL components

With the inclusion of a URL parser, we suddenly have the opportunity for lubridate-style component getting and setting. Syntax is identical to that of lubridate, but uses URL components as function names.

url <- "https://en.wikipedia.org/wiki/Article"
scheme(url)
"https"
scheme(url) <- "ftp"
url
"ftp://en.wikipedia.org/wiki/Article"

Fields that can be extracted or set are scheme, domain, port, path, parameters and fragment.

Suffix and TLD extraction

Once we've extracted a domain from a URL with domain or url_parse, we can identify which bit is the domain name, and which bit is the suffix:

> url <- "https://en.wikipedia.org/wiki/Article"
> domain_name <- domain(url)
> domain_name
[1] "en.wikipedia.org"
> str(suffix_extract(domain_name))
'data.frame':   1 obs. of  4 variables:
 $ host     : chr "en.wikipedia.org"
 $ subdomain: chr "en"
 $ domain   : chr "wikipedia"
 $ suffix      : chr "org"

This relies on an internal database of public suffixes, accessible at suffix_dataset - we recognise, though, that this dataset may get a bit out of date, so you can also pass the results of the suffix_refresh function, which retrieves an updated dataset, to suffix_extract:

domain_name <- domain("https://en.wikipedia.org/wiki/Article")
updated_suffixes <- suffix_refresh()
suffix_extract(domain_name, updated_suffixes)

We can do the same thing with top-level domains, with precisely the same setup, except the functions and datasets are tld_refresh, tld_extract and tld_dataset.

In the other direction we have host_extract, which retrieves, well, the host! If the URL has subdomains, it'll be the lowest-level subdomain. If it doesn't, it'll be the actual domain name, without the suffixes:

domain_name <- domain("https://en.wikipedia.org/wiki/Article")
host_extract(domain_name)

Query manipulation

Once a URL is parsed, it's sometimes useful to get the value associated with a particular query parameter. As an example, take the URL http://en.wikipedia.org/wiki/api.php?action=parse&pageid=1023&export=json. What pageID is being used? What is the export format? We can find out with param_get.

> str(param_get(urls = "http://en.wikipedia.org/wiki/api.php?action=parse&pageid=1023&export=json",
                     parameter_names = c("pageid","export")))
'data.frame':   1 obs. of  2 variables:
 $ pageid: chr "1023"
 $ export: chr "json"

This isn't the only function for query manipulation; we can also dynamically modify the values a particular parameter might have, or strip them out entirely.

To modify the values, we use param_set:

url <- "http://en.wikipedia.org/wiki/api.php?action=parse&pageid=1023&export=json"
url <- param_set(url, key = "pageid", value = "12")
url
# [1] "http://en.wikipedia.org/wiki/api.php?action=parse&pageid=12&export=json"

As you can see this works pretty well; it even works in situations where the URL doesn't have a query yet:

url <- "http://en.wikipedia.org/wiki/api.php"
url <- param_set(url, key = "pageid", value = "12")
url
# [1] "http://en.wikipedia.org/wiki/api.php?pageid=12"

On the other hand we might have a parameter we just don't want any more - that can be handled with param_remove, which can take multiple parameters as well as multiple URLs:

url <- "http://en.wikipedia.org/wiki/api.php?action=parse&pageid=1023&export=json"
url <- param_remove(url, keys = c("action","export"))
url
# [1] "http://en.wikipedia.org/wiki/api.php?pageid=1023"