More efficient chomp for String (#58192)

We can avoid all the computation of valid indices for this simple
function, since it only strips CR (0xD) and LF (0xA), which are known to
be single bytes. This is approximately four times faster.

---------

Co-authored-by: Dilum Aluthge <[email protected]>

Author: jakobnissen

base/strings/util.jl

@@ -319,14 +319,20 @@ end
 """
     chomp(s::AbstractString)::SubString
 
-Remove a single trailing newline from a string.
+Remove a single trailing newline (i.e. "\\r\\n" or "\\n") from a string.
 
 See also [`chop`](@ref).
 
 # Examples
 ```jldoctest
 julia> chomp("Hello\\n")
 "Hello"
+
+julia> chomp("World\\r\\n")
+"World"
+
+julia> chomp("Julia\\r\\n\\n")
+"Julia\\r\\n"
 ```
 """
 function chomp(s::AbstractString)
@@ -336,17 +342,22 @@ function chomp(s::AbstractString)
     (j < 1 || s[j] != '\r') && (return SubString(s, 1, j))
     return SubString(s, 1, prevind(s,j))
 end
-function chomp(s::String)
-    i = lastindex(s)
-    if i < 1 || codeunit(s,i) != 0x0a
-        return @inbounds SubString(s, 1, i)
-    elseif i < 2 || codeunit(s,i-1) != 0x0d
-        return @inbounds SubString(s, 1, prevind(s, i))
+
+@assume_effects :removable :foldable function chomp(s::Union{String, SubString{String}})
+    cu = codeunits(s)
+    ncu = length(cu)
+    len = if iszero(ncu)
+        0
     else
-        return @inbounds SubString(s, 1, prevind(s, i-1))
+        has_lf = @inbounds(cu[ncu]) == 0x0a
+        two_bytes = ncu > 1
+        has_cr = has_lf & two_bytes & (@inbounds(cu[ncu - two_bytes]) == 0x0d)
+        ncu - (has_lf + has_cr)
     end
+    off = s isa String ? 0 : s.offset
+    par = s isa String ? s : s.string
+    @inbounds @inline SubString{String}(par, off, len, Val{:noshift}())
 end
-
 """
     lstrip([pred=isspace,] str::AbstractString)::SubString
     lstrip(str::AbstractString, chars)::SubString