Add C++ Pattern Validation

[asraa]

Envoy now uses RE2 as a safe regex engine instead of std::regex (envoyproxy/envoy#7878). Because PGV already requires patterns to use RE2 syntax, one option is to use RE2 for C++ patterns as well. This implements it, for use in strings, bytes, repeated items, and may key/value pattern validation.

Implements #22

WIP: I ran in to difficulty creating the regex because a regex containing a null character would get cut off... for example, the ascii character test used the pattern, ^[\x00-x7f]+$, and consuming this as a string resulted in creating a null-terminated string pattern ^[ instead of the actual pattern. I think this might be a problem across most of the C++ code? That's why there's a terrible string construction in the pattern creation.

Signed-off-by: Asra Ali asraa@google.com

[asraa]

@akonradi

[asraa]

@rmichela @rodaine would either one of you be able to give this a pass over? thank you!

[rodaine]

This looks fine to me, but I'm not terribly familiar with C++ enough to say if this is ideal or not.

Looking in Envoy, the RE2 usages use a StringPiece. Is that applicable here? https://github.com/envoyproxy/envoy/blob/master/source/common/common/regex.cc#L47

[akonradi]

The construction of a std::string here doesn't seem terrible; I don't think there's a better way to deal with null-terminated strings unless you want to introduce absl/std::string_view. That would save a copy on startup.

[asraa]

(Sorry just getting back to this after KubeCon) Oh, thanks Chris and Alex. I think we can get the best of both worlds with that re2::StringPiece. Should avoid the copy. I'll update and test using that for both the matcher and the value.

[asraa]

I used re2::StringPiece where I could to avoid copies. I do have a question, seems like the pattern is to panic if a regex is invalid, so I eliminated re2::Quiet, it'll throw an error if it's an invalid pattern like the go code does.

[akonradi]

I assume we're validating the regex when generating the code, so we shouldn't see invalid regexes during validation. I think the behavior of failing loudly is reasonable.