Fix for 31456 -- replace with optional matching group fails when there is no match

[tlienart]

This PR suggests a fix for #31456 .

I don't know enough about Julia's regex system to know if that one line could have negative side-effects so I apologise in advance if it's a dumb idea.

Edit: added tests for the respective cases.

[JeffBezanson]

This doesn't look right --- isn't match_data always a Ptr?

We might have to modify substring_length_bynumber to return 0 in this error case, but I'm not sure whether that's the best way to do it.

[tlienart]

Ok yes. Looking at the PCRE2 doc for substring (https://code.woboq.org/qt5/qtbase/src/3rdparty/pcre2/src/pcre2_substring.c.html) L315 onwards, it returns a PCRE2_ERROR_UNSET code (-55) when it can't find a match.

So locally I did the following:

1. modify substring_length_bynumber as suggested with

function substring_length_bynumber(match_data, number)
    s = Base.RefValue{Csize_t}()
    rc = ccall((:pcre2_substring_length_bynumber_8, PCRE_LIB), Cint,
          (Ptr{Cvoid}, UInt32, Ref{Csize_t}), match_data, number, s)
    if rc < 0
        rc == -55 && return 0
        error("PCRE error: $(err_message(rc))")
    end
    convert(Int, s[])
end

2. modify _write_capture so that it doesn't go further when it has a length of 0 (otherwise the next PCRE function fails):

function _write_capture(io, re, group)
    len = substring_length_bynumber(re.match_data, group)
    len == 0 && return
    ensureroom(io, len+1)
    PCRE.substring_copy_bynumber(re.match_data, group,
        pointer(io.data, io.ptr), len+1)
    io.ptr += len
    io.size = max(io.size, io.ptr - 1)
end

And:

julia> replace("The quick fox quickly.", r"(quick)?\sfox(es)?\s(run)?" => s"\1 bus\2 \3")
"The quick bus quickly."

Should I try that in the PR or should I close it and let someone who knows more about PCRE take over? I've added that to the PR, feel free to close the PR if you think that's not the right approach, happy for someone more competent to do it (or tell me how it should be done)

[vtjnash]

Suggested change

	# PCRE2_ERROR_UNSET see https://code.woboq.org/qt5/qtbase/src/3rdparty/pcre2/src/pcre2_substring.c.html
	rc == -55 && return 0
	rc == ERROR_UNSET && return 0

[tlienart]

closing in favour of the updated #40391 (sorry for opening a new one).