The package unglue features functions such as unglue(), unglue_data() and unglue_unnest() which provide in many cases a more readable alternative to base regex functions. Simple cases indeed don’t require regex knowledge at all.

It uses a syntax inspired from the functions of Jim Hester’s glue package to extract matched substrings using a pattern, but is not endorsed by the authors of glue nor tidyverse packages.

It is completely dependency free, though formula notation of functions is supported if rlang is installed.

using an example from ?glue::glue backwards

glued_data <- head(mtcars) %>% glue_data("{rownames(.)} has {hp} hp")
#> Mazda RX4 has 110 hp
#> Mazda RX4 Wag has 110 hp
#> Datsun 710 has 93 hp
#> Hornet 4 Drive has 110 hp
#> Hornet Sportabout has 175 hp
#> Valiant has 105 hp
unglue_data(glued_data, "{rownames(.)} has {hp} hp")
#>         rownames...  hp
#> 1         Mazda RX4 110
#> 2     Mazda RX4 Wag 110
#> 3        Datsun 710  93
#> 4    Hornet 4 Drive 110
#> 5 Hornet Sportabout 175
#> 6           Valiant 105

use several patterns, the first that matches will be used

facts <- c("Antarctica is the largest desert in the world!",
"The largest country in Europe is Russia!",
"The smallest country in Europe is Vatican!",
"Disneyland is the most visited place in Europe! Disneyland is in Paris!",
"The largest island in the world is Green Land!")
facts_df <- data.frame(id = 1:5, facts)

patterns <- c("The {adjective} {place_type} in {bigger_place} is {place}!",
            "{place} is the {adjective} {place_type=[^ ]+} in {bigger_place}!{=.*}")
unglue_data(facts, patterns)
#>        place    adjective place_type bigger_place
#> 1 Antarctica      largest     desert    the world
#> 2     Russia      largest    country       Europe
#> 3    Vatican     smallest    country       Europe
#> 4 Disneyland most visited      place       Europe
#> 5 Green Land      largest     island    the world

Note that the second pattern uses some regex, regex needs to be typed after an = sign, if its has no left hand side then the expression won’t be attributed to a variable. in fact the pattern "{foo}" is a shorthand for "{foo=.*?}".

escaping characters

Special characters outside of the curly braces should not be escaped.

sentences <- c("666 is [a number]", "foo is [a word]", "42 is [the answer]", "Area 51 is [unmatched]")
patterns <- c("{number=\\d+} is [{what}]", "{word=\\D+} is [{what}]")
unglue_data(sentences, patterns)
#>   number       what word
#> 1    666   a number <NA>
#> 2   <NA>     a word  foo
#> 3     42 the answer <NA>
#> 4   <NA>       <NA> <NA>

type conversion

In order to convert types automatically we can set convert = TRUE, in the example above the column number will be converted to numeric.

unglue_data(sentences, patterns, convert = TRUE)
#>   number       what word
#> 1    666   a number <NA>
#> 2     NA     a word  foo
#> 3     42 the answer <NA>
#> 4     NA       <NA> <NA>

convert = TRUE triggers the use of utils::type.convert with parameter as.is = TRUE. We can also set convert to another conversion function such as readr::type_convert, or to a formula is rlang is installed.


unglue_unnest() is named as a tribute to tidyr::unnest() as it’s equivalent to using sucessively unglue() and unnest() on a data frame column. It is similar to tidyr::extract() in its syntax and efforts were made to make it as consistent as possible.

unglue_unnest(facts_df, facts, patterns)
#>   id
#> 1  1
#> 2  2
#> 3  3
#> 4  4
#> 5  5
unglue_unnest(facts_df, facts, patterns, remove = FALSE)
#>   id
#> 1  1
#> 2  2
#> 3  3
#> 4  4
#> 5  5
#>                                                                     facts
#> 1                          Antarctica is the largest desert in the world!
#> 2                                The largest country in Europe is Russia!
#> 3                              The smallest country in Europe is Vatican!
#> 4 Disneyland is the most visited place in Europe! Disneyland is in Paris!
#> 5                          The largest island in the world is Green Land!


While unglue() returns a list of data frames, unglue_vec() returns a character vector (unless convert = TRUE), if several matches are found in a string the extracted match will be chosen by name or by position.

unglue_vec(sentences, patterns, "number")
#> [1] "666" NA    "42"  NA
unglue_vec(sentences, patterns, 1)
#> [1] "666" "foo" "42"  NA


unglue_detect() returns a logical vector, it’s convenient to check that the input was matched by a pattern, or to subset the input to take a look at unmatched elements.

unglue_detect(sentences, patterns)
subset(sentences, !unglue_detect(sentences, patterns))
#> [1] "Area 51 is [unmatched]"


unglue_regex() returns a character vector of regex patterns, all over functions are wrapped around it and it can be used to leverage the unglue in other functions.

#> {number=\\d+} is [{what}]   {word=\\D+} is [{what}] 
#> "^(\\d+) is \\[(.*?)\\]$" "^(\\D+) is \\[(.*?)\\]$"
unglue_regex(patterns, named_capture = TRUE)
#>                 {number=\\d+} is [{what}] 
#> "^(?<number>\\d+) is \\[(?<what>.*?)\\]$" 
#>                   {word=\\D+} is [{what}] 
#>   "^(?<word>\\D+) is \\[(?<what>.*?)\\]$"
unglue_regex(patterns, attributes = TRUE)
#> {number=\\d+} is [{what}]   {word=\\D+} is [{what}] 
#> "^(\\d+) is \\[(.*?)\\]$" "^(\\D+) is \\[(.*?)\\]$" 
#> attr(,"groups")
#> attr(,"groups")$`{number=\\d+} is [{what}]`
#> number   what 
#>      1      2 
#> attr(,"groups")$`{word=\\D+} is [{what}]`
#> word what 
#>    1    2

duplicated labels

We can ensure that a pattern is repeated by repeating its label

unglue_data(c("black is black","black is dark"), "{color} is {color}")
#>   color
#> 1 black
#> 2  <NA>

We can change this behavior by feeding a function to the multiple parameter, in that case this function will be applies on the matches.

unglue_data(c("black is black","black is dark"), "{color} is {color}", multiple = paste)
#>         color
#> 1 black black
#> 2  black dark