zuregex

Base R already matches regular expressions with PCRE2 when you ask for perl = TRUE. But it compiles the pattern again on every call, matches on one thread, hands back capture groups as attributes you have to take apart, and knows only \\1 to \\9 in replacements. A package that needs PCRE2 from C has to bundle its own copy, as vectorscan does.

zuregex bundles one PCRE2, the same version on every platform, and builds on it twice: as re_*() functions for R, and as a C API other packages reach through LinkingTo.

• Captures as data. re_match() returns a data frame with a column per group, built in C, and can type the columns as utils::strcapture() does.
• Compiled once. Patterns go through a cache, so a pattern used in a loop is compiled once, JIT included.
• Many cores. Matching runs on several threads with OpenMP on large inputs, with the same results on any number of threads.
• PCRE2’s full replacement language. Groups by name, past nine, with defaults and conditionals, and case forcing.
• Bounded work. Match, depth and heap limits per call, so one pathological subject gives an error or an NA instead of a stall.

Installation

Install zuregex from CRAN with:

install.packages("zuregex")

Or install the development version from GitHub with:

# install.packages("pak")
pak::pak("pedrobtz/zuregex")

Usage

library(zuregex)

Take a few lines from a web server’s log.

log <- c(
  "2026-10-21 08:15:02 GET /index.html 200 512",
  "2026-10-21 08:15:09 POST /login 302 0",
  "2026-10-21 08:16:44 GET /missing 404 128",
  "not a log line"
)

re_detect() and re_count() work as grepl() and stringr::str_count() do, element by element.

re_detect(log, "\\s(?:GET|POST)\\s")
#> [1]  TRUE  TRUE  TRUE FALSE
re_count(log, "\\d+")
#> [1] 8 8 8 0

re_match() turns named groups into columns. A line that does not match is a row of NA. proto gives the columns types, here an integer status.

pattern <- "(?<time>\\S+ \\S+) (?<method>[A-Z]+) (?<path>\\S+) (?<status>\\d+)"
re_match(log, pattern)
#>                  time method        path status
#> 1 2026-10-21 08:15:02    GET /index.html    200
#> 2 2026-10-21 08:15:09   POST      /login    302
#> 3 2026-10-21 08:16:44    GET    /missing    404
#> 4                <NA>   <NA>        <NA>   <NA>
typed <- re_match(
  log,
  pattern,
  proto = data.frame(time = "", method = "", path = "", status = 0L)
)
str(typed)
#> 'data.frame':    4 obs. of  4 variables:
#>  $ time  : chr  "2026-10-21 08:15:02" "2026-10-21 08:15:09" "2026-10-21 08:16:44" NA
#>  $ method: chr  "GET" "POST" "GET" NA
#>  $ path  : chr  "/index.html" "/login" "/missing" NA
#>  $ status: int  200 302 404 NA

Replacements use PCRE2’s templates: ${name} for a named group, and ${n:-default} for a group that may not take part.

re_replace(log, "^(?<y>\\d+)-(?<m>\\d+)-(?<d>\\d+)", "${d}/${m}/${y}")
#> [1] "21/10/2026 08:15:02 GET /index.html 200 512"
#> [2] "21/10/2026 08:15:09 POST /login 302 0"      
#> [3] "21/10/2026 08:16:44 GET /missing 404 128"   
#> [4] "not a log line"
re_replace(c("v1.2", "v3"), "v(\\d+)(?:\\.(\\d+))?", "major $1, minor ${2:-0}")
#> [1] "major 1, minor 2" "major 3, minor 0"

A function works too. It is called once, with a data frame of every match.

re_replace(log, "/\\w+", function(m) toupper(m[["0"]]))
#> [1] "2026-10-21 08:15:02 GET /INDEX.html 200 512"
#> [2] "2026-10-21 08:15:09 POST /LOGIN 302 0"      
#> [3] "2026-10-21 08:16:44 GET /MISSING 404 128"   
#> [4] "not a log line"

re_compile() sets flags once:

re_detect(c("Hello", "hello"), re_compile("HELLO", flags = "i"))
#> [1] TRUE TRUE

and re_limits() bounds the work. ^(a+)+$ backtracks exponentially on a long run of a that ends in something else. With a limit, that element stops early and gives NA, with one warning for the call.

re_detect(
  c(paste0(strrep("a", 30), "!"), "aaa"),
  "^(a+)+$",
  limits = re_limits(match = 1e4, on_limit = "na")
)
#> Warning in re_detect(c(paste0(strrep("a", 30), "!"), "aaa"), "^(a+)+$", : 1
#> element of `x` hit a match limit and is `NA`.
#> [1]   NA TRUE

Every function follows the same pattern: x first, then pattern, which is one pattern for every element or one per element, never recycled.

From C

Packages that need PCRE2 from C add zuregex to Imports and LinkingTo and include <zuregex-r.h>, instead of bundling their own copy. Patterns and match data are opaque handles, so an upgrade of the bundled PCRE2 never changes a consumer’s build. The header documents the API.

Learn more

• The reference index lists every function, grouped as the API is.
• ?re_compile has the flag letters, ?re_replace the template syntax, and ?re_limits what limits can and cannot promise.

Related work

zuregex bundles PCRE2 and checks itself against PCRE2’s own test results, base R’s perl = TRUE and stringr. Its API borrows from Python’s pcre2.py, JavaScript’s node-pcre2 and Go’s go-re2. For matching in linear time, which PCRE2 cannot promise, use re2.