regex match within parenthesis - regex

Question

I'm attempting to use some regular expressions that I made for Python also work with R.
Here is what I have in Python (using the excellent re module), with my expected 3 matches:
import re
line = 'VARIABLES = "First [T]" "Second [L]" "Third [1/T]"'
re.findall('"(.*?)"', line)
# ['First [T]', 'Second [L]', 'Third [1/T]']
Now with R, here is my best attempt:
line <- 'VARIABLES = "First [T]" "Second [L]" "Third [1/T]"'
m <- gregexpr('"(.*?)"', line)
regmatches(line, m)[[1]]
# [1] "\"First [T]\"" "\"Second [L]\"" "\"Third [1/T]\""
Why did R match the whole pattern, rather than just within the parenthesis? I was expecting:
[1] "First [T]" "Second [L]" "Third [1/T]"
Furthermore, perl=TRUE didn't make any difference. Is it safe to assume that R's regex does not consider matching only the parenthesis, or is there some trick that I'm missing?
Summary of solution: thanks #flodel, it appears to work well with other patterns too, so it appears to be a good general solution. A compact form of the solution using an input string line and regular expression pattern pat is:
pat <- '"(.*?)"'
sub(pat, "\\1", regmatches(line, gregexpr(pat, line))[[1]])
Furthermore, perl=TRUE should be added to gregexpr if using PCRE features in pat.

Answer 1

If you print m, you'll see gregexpr(..., perl = TRUE) gives you the positions and lengths of matches for a) your full pattern including the leading and closing quotes and b) the captured (.*).
Unfortunately for you, when m is used by regmatches, it use the positions and lengths of the former.
There are two solutions I can think of.
Pass your final output through sub:
line <- 'VARIABLES = "First [T]" "Second [L]" "Third [1/T]"'
m <- gregexpr('"(.*?)"', line, perl = TRUE)
z <- regmatches(line, m)[[1]]
sub('"(.*?)"', "\\1", z)
Or use substring using the positions and lengths of the captured expressions:
start.pos <- attr(m[[1]], "capture.start")
end.pos <- start.pos + attr(m[[1]], "capture.length") - 1L
substring(line, start.pos, end.pos)
To further your understanding, see what happens if your pattern is trying to capture more than one thing. Also see that you can give names to your captures groups (what the doc refers to as Python-style named captures), here "capture1" and "capture2":
m <- gregexpr('"(?P<capture1>.*?) \\[(?P<capture2>.*?)\\]"', line, perl = TRUE)
m
start.pos <- attr(m[[1]], "capture.start")
end.pos <- start.pos + attr(m[[1]], "capture.length") - 1L
substring(line, start.pos[, "capture1"],
end.pos[, "capture1"])
# [1] "First" "Second" "Third"
substring(line, start.pos[, "capture2"],
end.pos[, "capture2"])
# [1] "T" "L" "1/T"

Answer 2

1) strapplyc in the gsubfn package acts in the way you were expecting:
> library(gsubfn)
> strapplyc(line, '"(.*?)"')[[1]]
[1] "First [T]" "Second [L]" "Third [1/T]"
2) Although it involves delving into m's attributes, its possible to make regmatches work by reconstructing m to refer to the captures rather than the whole match:
at <- attributes( m[[1]] )
m2 <- list( structure( c(at$capture.start), match.length = at$capture.length ) )
regmatches( line, m2 )[[1]]
3) If we knew that the strings always ended in ] and were willing to modify the regular expression then this would work:
> m3 <- gregexpr('[^"]*]', line)
> regmatches( line, m3 )[[1]]
[1] "First [T]" "Second [L]" "Third [1/T]"