Remove System.Net.Http dependency on System.Text.RegularExpressions

[stephentoub]

This dependency is the only reason we end up with a 105K System.Text.RegularExpressions.dll as part of a trimmed new MVC app. Regex is used in the case where on Windows a bypass list is provided, in which case each item in the list is changed into a regex, which is then evaluated against each url provided to SocketsHttpHandler. But the patterns usable are simple: the only special character recognized is an asterisk, which can map to zero or more of any character. So, we can instead employ a simple processor for such patterns, which then eliminates the need to reference System.Text.RegularExpressions.dll from System.Net.Http.dll. It also happens to be faster.

cc: @ViktorHofer, @jkotas, @wfurt

[danmoseley]

I was hoping I could write it in a simple way like this, which works fine, then transform into tail-call. But tail-call can't work with the alternation.

        private static bool Match(string t, int it, string p, int ip)
        {
            if (ip == p.Length) // if there's no more pattern, it's a match iff there's no more text
                return it == t.Length;

            if (it == t.Length) // if there's no more text, it's a match iff pattern is '*' and it's a match for subpattern
                return p[ip] == '*' && Match(t, it, p, ip + 1);

            if (p[ip] == '*') // if pattern is *, it's a match iff it's a match for subtext (* consumes and continues) 
                              // or it's a match for subpattern (* consumes nothing)
                return Match(t, it + 1, p, ip) || Match(t, it, p, ip + 1);

            if (char.ToLowerInvariant(p[ip]) == char.ToLowerInvariant(t[it]))
                // if pattern and text match, it's a match if it's a match for subtext + subpattern
                return Match(t, it + 1, p, ip + 1);

            return false;
        }

[danmoseley]

It adds "$" at the end but not "^" at the start. Does that mean that it would match suffixes-only, while your new code will not?

[stephentoub]

That's a good catch. That seems wrong in the previous code (e.g. if the pattern were "a.com" it would say that "ba.com" matched). @wfurt, was that intentional? The new code could be changed to match by adding a * at the beginning of the pattern if necessary.

[wfurt]

yes, I think this is bug. With curl, .example.com would also match www.example.com (https://curl.haxx.se/libcurl/c/CURLOPT_NOPROXY.html) but for windows it does not work that way. *a.com should match a.com and ba.com but a.com rule should not match ba.com

[stephentoub]

Ok, thanks, then I'll leave this as-is. Is fixing it in master sufficient, or should a fix for this (presumably prepending "^" +) be considered for 3.1?

[stephentoub]

I was hoping I could write it in a simple way like this

Yeah, but this can end up having really problematic performance in the worst case. Try running this with your code:

Match(new string('a', 100), 0, "a*a*a*a*a*a*a*a*b", 0)