Figure out why roxygen2 tests fail with dev stringr

[hadley]

It's something to do with new str_replace behaviour

[hadley]

Minimal reprex:

str_replace_all('"', '(["\\\\])', "\\\\\\1")
# [1] "\\1"

I'm reasonably certain the regular expression is correct, and it should return '\\"' (i.e. the goal is to insert an additional backslash in front of a double quote or backslash)

@gagolews I think this was caused by 4905b0d - I suspect this might not be a problem that regular expressions are well suited to solve, and it might be necessary to implemented a minimal parser in C++ (would need to understand character classes and \-escaping)

[hadley]

@gagolews just wanted to check if you happened to have built any like this into stringi in the interim. Otherwise I'll start writing up more test cases

[gagolews]

> sub('(["\\\\])', "\\\\\\1", '"', perl=TRUE)
[1] "\\\""
> 
> sub('(["\\\\])', "\\\\\\1", '"', perl=FALSE)
[1] "\\\""

I didn't build anything into stringi like this.

[hadley]

This is a somewhat easier to understand reprex:

library(stringr)
library(stringi)

# OK
str_replace_all("x", "(.)", "\\\\-\\1")
#> [1] "\\-x"
stri_replace_all_regex("x", "(.)", "\\\\-$1")
#> [1] "\\-x"

# NOT OK
str_replace_all("x", "(.)", "\\\\\\1")
#> [1] "\\1"
stri_replace_all_regex("x", "(.)", "\\\\$1")
#> [1] "\\x"