Drop stringr dependency (#1841)

roxygen2 (and hence devtoosl) no longer depends on stringr (or transitively, stringi), making
it easier to install in constrained Linux environments.

Replacements:

- `str_detect()` → `grepl()`
- `str_trim()` → `trimws()`
- `str_replace()` / `str_remove()` → `sub()`
- `str_replace_all()` → `gsub()` and local `re_replace_all()` helper
- `str_split()` → `strsplit()`
- `str_extract()` → `regexpr()` + `regmatches()`
- `str_count()` → local `re_count()` helper
- `str_sub()` → `substr()` / `substring()`
- `str_match()` → `regmatches()` + `regexec()`
- `str_match_all()` → `regmatches()` + `gregexec()`
- `str_locate_all()` → `gregexpr()`
- `str_split_fixed()` → local `re_split_half()` helper (only ever needed n=2)

Base R quirks worked around:

- `strsplit("", pattern)` returns `character(0)`, not `""` like stringr, so `tag_words()` now guards against empty input before splitting.
- `substr<-` can't change string length, so `str_sub<-` assignments were replaced with manual `paste0(left, val, right)` concatenation.
- `regmatches()` + `gregexec()` returns a list-of-matrices that needs `t()` to match stringr's column-oriented `str_match_all()` layout.
- `gregexpr()` returns a list with a single `-1` element (not zero-length) when there's no match, requiring an explicit `> 0` filter in `re_count()`.
- `stringr::fixed()` wrappers in tests were replaced with `fixed = TRUE` arguments to base functions.
- The `(?x)` free-spacing flag is used inline in `paste0()`-assembled regexps (since base R has no `regex(comments = TRUE)` wrapper), with comments placed on each `paste0()` line instead.

Author: hadley

DESCRIPTION

@@ -31,7 +31,6 @@ Imports:
     pkgload (>= 1.0.2),
     R6 (>= 2.1.2),
     rlang (>= 1.1.0),
-    stringr (>= 1.0.0),
     utils,
     withr,
     xml2

NAMESPACE

@@ -274,7 +274,6 @@ export(update_collate)
 export(vignette_roclet)
 export(warn_roxy_tag)
 import(rlang)
-import(stringr)
 importFrom(R6,R6Class)
 importFrom(knitr,knit)
 importFrom(knitr,opts_chunk)

NEWS.md

@@ -1,4 +1,5 @@
 # roxygen2 (development version)
+* roxygen2 no longer depends on stringr/stringi. This means that no package in the devtools constellation depends on stringr, which in turn means you no longer need stringi, making it a bit easier to install in constrained Linux environments.
 * roxygen2 options can now be set using `Config/roxygen2/` fields in DESCRIPTION (e.g. `Config/roxygen2/markdown: TRUE`) instead of the `Roxygen` field. The old `Roxygen` field is still supported. Similarly, the roxygen2 version is now stored in `Config/roxygen2/version` instead of `RoxygenNote` (#1328).
 * Tags that expect single-line input now warn when they span multiple lines, catching common mistakes. Affected tags: `@aliases`, `@concept`, `@encoding`, `@exportClass`, `@exportMethod`, `@exportPattern`, `@exportS3Method`, `@importFrom`, `@importClassesFrom`, `@importMethodsFrom`, `@include`, `@inheritParams`, `@keywords`, `@method`, `@name`, `@order`, `@rdname`, `@S3method`, `@template`, and `@useDynLib` (#1642, #1688). This may break some existing usage, but it prevents a wide class of otherwise silent errors.
 * `@examplesIf` now warns when there is no example code after the condition (#1695).

R/block.R

@@ -240,16 +240,16 @@ parse_description <- function(tags) {
   }
 
   intro <- tags[[1]]
-  intro$val <- str_trim(intro$raw)
+  intro$val <- trimws(intro$raw)
   if (intro$val == "") {
     return(tags[-1])
   }
 
   tags <- tags[-1]
   tag_names <- tag_names[-1]
 
-  paragraphs <- str_split(intro$val, fixed('\n\n'))[[1]]
-  lines <- str_count(paragraphs, "\n") + rep(2, length(paragraphs))
+  paragraphs <- strsplit(intro$val, '\n\n', fixed = TRUE)[[1]]
+  lines <- re_count(paragraphs, "\n") + rep(2, length(paragraphs))
   offsets <- c(0, cumsum(lines))
 
   # 1st paragraph = title (unless has @title)

R/collate.R

@@ -103,5 +103,5 @@ base_path <- function(path, base) {
   path <- normalizePath(path, winslash = "/")
   base <- normalizePath(base, winslash = "/")
 
-  str_replace(path, fixed(paste0(base, "/")), "")
+  sub(paste0(base, "/"), "", path, fixed = TRUE)
 }

R/markdown-code.R

@@ -58,7 +58,7 @@ markdown_evaluate <- function(text) {
   rcode_pos <- parse_md_pos(map_chr(rcode_nodes, xml_attr, "sourcepos"))
   rcode_pos <- work_around_cmark_sourcepos_bug(text, rcode_pos)
   out <- eval_code_nodes(rcode_nodes)
-  str_set_all_pos(text, rcode_pos, out, rcode_nodes)
+  re_set_all_pos(text, rcode_pos, out, rcode_nodes)
 }
 
 # Work around commonmark sourcepos bug for inline R code
@@ -68,7 +68,7 @@ work_around_cmark_sourcepos_bug <- function(text, rcode_pos) {
     return(rcode_pos)
   }
 
-  lines <- str_split(text, fixed("\n"))[[1]]
+  lines <- strsplit(text, "\n", fixed = TRUE)[[1]]
 
   for (l in seq_len(nrow(rcode_pos))) {
     # Do not try to fix multi-line code, we error for that (below)
@@ -79,7 +79,7 @@ work_around_cmark_sourcepos_bug <- function(text, rcode_pos) {
     start <- rcode_pos$start_column[l]
 
     # Maybe correct? At some point this will be fixed upstream, hopefully.
-    if (str_sub(line, start - 1, start + 1) == "`r ") {
+    if (substr(line, start - 1, start + 1) == "`r ") {
       next
     }
 
@@ -91,10 +91,11 @@ work_around_cmark_sourcepos_bug <- function(text, rcode_pos) {
     # the real "`r " left by six characters, there happens to be another
     # "`r " there.
 
-    indent <- nchar(str_extract(line, "^[ ]+"))
+    m <- regexpr("^[ ]+", line)
+    indent <- attr(m, "match.length")
     if (
-      !is.na(indent) &&
-        str_sub(line, start - 1 + indent, start + 1 + indent) == "`r "
+      m > 0L &&
+        substr(line, start - 1 + indent, start + 1 + indent) == "`r "
     ) {
       rcode_pos$start_column[l] <- rcode_pos$start_column[l] + indent
       rcode_pos$end_column[l] <- rcode_pos$end_column[l] + indent
@@ -106,7 +107,7 @@ work_around_cmark_sourcepos_bug <- function(text, rcode_pos) {
 
 is_markdown_code_node <- function(x) {
   info <- xml_attr(x, "info")
-  str_sub(xml_text(x), 1, 2) == "r " ||
+  substr(xml_text(x), 1, 2) == "r " ||
     (!is.na(info) && grepl("^[{][a-zA-z]+[}, ]", info))
 }
 
@@ -158,7 +159,7 @@ knitr_chunk_defaults <- function() {
   )
 }
 
-str_set_all_pos <- function(text, pos, value, nodes) {
+re_set_all_pos <- function(text, pos, value, nodes) {
   # Cmark has a bug when reporting source positions for multi-line
   # code tags, and it does not count the indenting space in the
   # continuation lines: https://github.com/commonmark/cmark/issues/296
@@ -169,7 +170,7 @@ str_set_all_pos <- function(text, pos, value, nodes) {
 
   # Need to split the string, because of the potential multi-line
   # code tags, and then also recode the positions
-  lens <- nchar(str_split(text, fixed("\n"))[[1]])
+  lens <- nchar(strsplit(text, "\n", fixed = TRUE)[[1]])
   shifts <- c(0, cumsum(lens + 1L))
   shifts <- shifts[-length(shifts)]
   start <- shifts[pos$start_line] + pos$start_column

R/markdown-escaping.R

@@ -119,7 +119,7 @@ find_fragile_rd_tags <- function(text, fragile) {
   ftags <- ftags[keep, ]
 
   if (nrow(ftags)) {
-    ftags$text <- str_sub(text, ftags$start, ftags$argend)
+    ftags$text <- substring(text, ftags$start, ftags$argend)
   }
 
   ftags
@@ -142,7 +142,7 @@ find_all_rd_tags <- function(text) {
   ## Find the end of the argument list for each tag. Note that
   ## tags might be embedded into the arguments of other tags.
   tags$argend <- map_int(seq_len(nrow(tags)), function(i) {
-    tag_plus <- str_sub(text, tags$end[i], text_len)
+    tag_plus <- substr(text, tags$end[i], text_len)
     findEndOfTag(tag_plus, is_code = FALSE, start = 0L) + tags$end[i]
   })
 
@@ -163,12 +163,28 @@ find_all_rd_tags <- function(text) {
 
 find_all_tag_names <- function(text) {
   ## Find the tags without arguments first
-  tag_pos <- str_locate_all(text, r"(\\[a-zA-Z][a-zA-Z0-9]*)")[[1]]
+  m <- gregexpr(r"(\\[a-zA-Z][a-zA-Z0-9]*)", text)[[1]]
+  if (m[[1]] == -1L) {
+    tag_pos <- matrix(
+      integer(),
+      ncol = 2,
+      dimnames = list(NULL, c("start", "end"))
+    )
+  } else {
+    tag_pos <- cbind(
+      start = as.integer(m),
+      end = as.integer(m) + attr(m, "match.length") - 1L
+    )
+  }
 
-  data.frame(
-    tag = str_sub(text, tag_pos[, "start"], tag_pos[, "end"]),
-    as.data.frame(tag_pos)
-  )
+  if (nrow(tag_pos) == 0) {
+    data.frame(tag = character(), start = integer(), end = integer())
+  } else {
+    data.frame(
+      tag = substring(text, tag_pos[, "start"], tag_pos[, "end"]),
+      as.data.frame(tag_pos)
+    )
+  }
 }
 
 #' Replace fragile Rd tags with placeholders
@@ -184,7 +200,7 @@ find_all_tag_names <- function(text) {
 protect_rd_tags <- function(text, rd_tags) {
   id <- make_random_string()
 
-  text <- str_sub_same(text, rd_tags, id)
+  text <- re_sub_same(text, rd_tags, id)
 
   attr(text, "roxygen-markdown-subst") <-
     list(tags = rd_tags, id = id)
@@ -207,7 +223,7 @@ protect_rd_tags <- function(text, rd_tags) {
 #'
 #' @noRd
 
-str_sub_same <- function(str, repl, id) {
+re_sub_same <- function(str, repl, id) {
   repl <- repl[order(repl$start), ]
 
   if (is.unsorted(repl$end) || is.unsorted(repl$argend)) {
@@ -217,7 +233,11 @@ str_sub_same <- function(str, repl, id) {
   for (i in seq_len(nrow(repl))) {
     ## The trailing - is needed, to distinguish between -1 and -10
     new_text <- paste0(id, "-", i, "-")
-    str_sub(str, repl$start[i], repl$argend[i]) <- new_text
+    str <- paste0(
+      substr(str, 1, repl$start[i] - 1),
+      new_text,
+      substr(str, repl$argend[i] + 1, nchar(str))
+    )
 
     ## Need to shift other coordinates (we shift everything,
     ## it is just simpler).

R/markdown-link.R

@@ -64,26 +64,30 @@ add_linkrefs_to_md <- function(text) {
 }
 
 get_md_linkrefs <- function(text) {
-  refs <- str_match_all(
+  refs <- regmatches(
     text,
-    regex(
-      comments = TRUE,
-      "
-        (?<=[^\\]\\\\]|^)       # must not be preceded by ] or \
-        \\[([^\\]\\[]+)\\]      # match anything inside of []
-        (?:\\[([^\\]\\[]+)\\])? # match optional second pair of []
-        (?=[^\\[{]|$)           # must not be followed by [ or {
-      "
+    gregexec(
+      paste0(
+        "(?x)",
+        "(?<=[^\\]\\\\]|^)", # must not be preceded by ] or \
+        "\\[([^\\]\\[]+)\\]", # match anything inside of []
+        "(?:\\[([^\\]\\[]+)\\])?", # match optional second pair of []
+        "(?=[^\\[{]|$)" # must not be followed by [ or {
+      ),
+      text,
+      perl = TRUE
     )
   )[[1]]
+  if (length(refs) > 0) {
+    refs <- t(refs)
+  }
 
   if (length(refs) == 0) {
     return(character())
   }
 
   ## For the [fun] form the link text is the same as the destination.
-  # Need to check both NA and "" for different versions of stringr
-  refs[, 3] <- ifelse(is.na(refs[, 3]) | refs[, 3] == "", refs[, 2], refs[, 3])
+  refs[, 3] <- ifelse(refs[, 3] == "", refs[, 2], refs[, 3])
 
   refs3encoded <- map_chr(refs[, 3], URLencode)
   paste0("[", refs[, 3], "]: ", "R:", refs3encoded)
@@ -126,12 +130,12 @@ parse_link <- function(destination, contents, state) {
 
   is_code <- is_code || (grepl("[(][)]$", destination) && !has_link_text)
 
-  pkg <- str_match(destination, "^(.*)::")[1, 2]
+  pkg <- regmatches(destination, regexec("^(.*)::", destination))[[1]][2]
   explicit_pkg <- !is.na(pkg)
   fun <- utils::tail(strsplit(destination, "::", fixed = TRUE)[[1]], 1)
   topic <- sub("[(][)]$", "", fun)
-  if (!has_link_text && str_detect(destination, "-class$")) {
-    fun <- str_match(fun, "^(.*)-class$")[1, 2]
+  if (!has_link_text && grepl("-class$", destination)) {
+    fun <- regmatches(fun, regexec("^(.*)-class$", fun))[[1]][2]
   }
 
   # Standardise links: cross-packagae always get prefix;

R/markdown.R

@@ -78,7 +78,7 @@ markdown_pass1 <- function(text) {
   rcode_pos <- parse_md_pos(map_chr(rcode_nodes, xml_attr, "sourcepos"))
   rcode_pos <- work_around_cmark_sourcepos_bug(text, rcode_pos)
   out <- eval_code_nodes(rcode_nodes)
-  str_set_all_pos(text, rcode_pos, out, rcode_nodes)
+  re_set_all_pos(text, rcode_pos, out, rcode_nodes)
 }
 
 # Work around commonmark sourcepos bug for inline R code
@@ -88,7 +88,7 @@ work_around_cmark_sourcepos_bug <- function(text, rcode_pos) {
     return(rcode_pos)
   }
 
-  lines <- str_split(text, fixed("\n"))[[1]]
+  lines <- strsplit(text, "\n", fixed = TRUE)[[1]]
 
   for (l in seq_len(nrow(rcode_pos))) {
     # Do not try to fix multi-line code, we error for that (below)
@@ -99,7 +99,7 @@ work_around_cmark_sourcepos_bug <- function(text, rcode_pos) {
     start <- rcode_pos$start_column[l]
 
     # Maybe correct? At some point this will be fixed upstream, hopefully.
-    if (str_sub(line, start - 1, start + 1) == "`r ") {
+    if (substr(line, start - 1, start + 1) == "`r ") {
       next
     }
 
@@ -111,10 +111,11 @@ work_around_cmark_sourcepos_bug <- function(text, rcode_pos) {
     # the real "`r " left by six characters, there happens to be another
     # "`r " there.
 
-    indent <- nchar(str_extract(line, "^[ ]+"))
+    m <- regexpr("^[ ]+", line)
+    indent <- attr(m, "match.length")
     if (
-      !is.na(indent) &&
-        str_sub(line, start - 1 + indent, start + 1 + indent) == "`r "
+      m > 0L &&
+        substr(line, start - 1 + indent, start + 1 + indent) == "`r "
     ) {
       rcode_pos$start_column[l] <- rcode_pos$start_column[l] + indent
       rcode_pos$end_column[l] <- rcode_pos$end_column[l] + indent
@@ -126,7 +127,7 @@ work_around_cmark_sourcepos_bug <- function(text, rcode_pos) {
 
 is_markdown_code_node <- function(x) {
   info <- xml_attr(x, "info")
-  str_sub(xml_text(x), 1, 2) == "r " ||
+  substr(xml_text(x), 1, 2) == "r " ||
     (!is.na(info) && grepl("^[{][a-zA-z]+[}, ]", info))
 }
 
@@ -178,7 +179,7 @@ knitr_chunk_defaults <- function() {
   )
 }
 
-str_set_all_pos <- function(text, pos, value, nodes) {
+re_set_all_pos <- function(text, pos, value, nodes) {
   # Cmark has a bug when reporting source positions for multi-line
   # code tags, and it does not count the indenting space in the
   # continuation lines: https://github.com/commonmark/cmark/issues/296
@@ -189,7 +190,7 @@ str_set_all_pos <- function(text, pos, value, nodes) {
 
   # Need to split the string, because of the potential multi-line
   # code tags, and then also recode the positions
-  lens <- nchar(str_split(text, fixed("\n"))[[1]])
+  lens <- nchar(strsplit(text, "\n", fixed = TRUE)[[1]])
   shifts <- c(0, cumsum(lens + 1L))
   shifts <- shifts[-length(shifts)]
   start <- shifts[pos$start_line] + pos$start_column
@@ -235,13 +236,13 @@ mdxml_children_to_rd_top <- function(xml, state) {
   state$section_tag <- uuid()
   out <- map_chr(xml_children(xml), mdxml_node_to_rd, state)
   out <- c(out, mdxml_close_sections(state))
-  rd <- str_trim(paste0(out, collapse = ""))
+  rd <- trimws(paste0(out, collapse = ""))
   if (state$has_sections) {
     secs <- strsplit(rd, state$section_tag, fixed = TRUE)[[1]] %||% ""
     titles <- c("", state$titles)
     # strsplit drops trailing empty strings, so pad to match titles length
     secs <- c(secs, rep("", length(titles) - length(secs)))
-    rd <- structure(str_trim(secs), names = titles)
+    rd <- structure(trimws(secs), names = titles)
   }
   rd
 }

R/namespace.R

@@ -224,23 +224,23 @@ roxy_tag_ns.roxy_tag_exportS3Method <- function(x, block, env) {
       return()
     }
 
-    if (!str_detect(x$val, "::")) {
+    if (!grepl("::", x$val, fixed = TRUE)) {
       warn_roxy_tag(x, "must have form package::generic")
       return()
     }
 
-    generic <- str_split(x$val, "::")[[1]]
+    generic <- re_split_half(x$val, "::")
     generic_re <- paste0("^", generic[[2]], "\\.")
 
-    if (!str_detect(obj$alias, generic_re)) {
+    if (!grepl(generic_re, obj$alias)) {
       warn_roxy_tag(
         x,
         "generic ({.str {generic[[2]]}}) doesn't match function ({.str {obj$alias}})",
       )
       return()
     }
 
-    class <- str_remove(obj$alias, generic_re)
+    class <- sub(generic_re, "", obj$alias)
     method <- c(x$val, class)
   } else {
     method <- x$val