gsub regex in R - ignore newline symbol - r

Question

Here's a reproducible example
S0 <- "\n3 4 5"
S1 <- "\n3 5"
I want to use gsub and the following regex pattern (outside of R it works - tested in regex101) to return the digits. This regex should ignore \ and n whether they occur together or not.
([^\\n])(\s{1})?
I am not looking for a way to match the digits with a fundamentally different pattern - I'd like to know how to get the above pattern to work in R. The following do not work for me
gsub("([^\\\n])(\\s{1})?", "\\1", S0)
gsub("([^[\\\]n])(\\s{1})?", "\\1", S1)
The output should be
#S0 - 345
#S1 - 3 5

Answer 1

Since you specifically want that regex to work you could match and optional \n (using (\n)?):
gsub("(\n)?([^\\n])(\\s{1})", "\\2", S0)
#[1] "345"
gsub("(\n)?([^\\n])(\\s{1})", "\\2", S1)
#[1] "3 5"
Note that you were right, if you use a regex tester like: https://regex101.com/ it works without the extra "(\n)?". However, I think in R you have to match more for capture groups to work properly.

Answer 2

Your ([^\\n])(\s{1})? pattern in regex101 (PCRE) matches different strings than the same pattern used in gsub without perl=TRUE (that is, when it is handled by the TRE regex library). They would work the same if you used perl=TRUE and use gsub("([^\\\\n])(\\s{1})?", "\\1", S1, perl=TRUE).
What is so peculair with the PCRE Regex ([^\\n])(\s{1})?
This pattern in a regex tester with PCRE option matches:
([^\\n]) - any char other than \ and n (put into Group 1)
(\s{1})? - matches and captures into Group 2 any single whitespace char, optionally, 1 or 0 times.
Note this pattern does not match any non-newline char with the first capturing group, it would match any non-newline if it were [^\n].
Now, the same regex with gsub will be
gsub("([^\n])(\\s{1})?", "\\1", S1) # OR
gsub("([^\\\\n])(\\s{1})?", "\\1", S1, perl=TRUE)
Why different number of backslashes? Because the first regex is handled with TRE regex library and in these patterns, inside bracket expressions, no regex escapes are parsed as such, the \ and n are treated as 2 separate chars. In a PCRE pattern, the one with perl=TRUE, the [...] are called character classes and inside them, you can define regex escapes, and thus the \ regex escape char should be doubled (that is, inside the R string literal it should be quadrupled as you need a \ to escape \ for the R engine to "see" a backslash).
Actually, if you want to match a newline, you just need to use \n in the regex pattern, you may either use "\n" or "\\n" as both TRE and PCRE regex engines parse LF and a \n regex escape as a newline char matching pattern. These four are equivalent:
gsub("\n([^\n])(\\s{1})?", "\\1", S1)
gsub("\\n([^\n])(\\s{1})?", "\\1", S1)
gsub("\n([^\\\\n])(\\s{1})?", "\\1", S1, perl=TRUE)
gsub("\\n([^\\\\n])(\\s{1})?", "\\1", S1, perl=TRUE)
If the \n must be optional, just add ? quantifier after it, no need wrapping it with a group:
gsub("\n?([^\n])(\\s{1})?", "\\1", S1)
^
And simplifying it further:
gsub("\n?([^\n])\\s?", "\\1", S1)
And also, if by [^\n] you want to match any char but a newline, just use . with (?n) inline modifier:
gsub("(?n)(.)(\\s{1})?", "\\1", S1)
See R demo online.

Answer 3

A couple of issues. The is not a backslash in your S object (it's an escape-operator rather than a character) and there is a predefined digit character class that can be negated:
gsub("[^[:digit:]]", "", S)
[1] "345"
If in the other hand you wanted to exclude the newline character and the spaces, it would be done by removing one of the escape operators, since they are not needed except for the small group of special characters that exist in the character class context:
gsub("[\n ]", "", S)
[1] "345"