fix: Replace boost::regex with RE2 in parse_duration function

[abhinavmuk04]

Summary:
Some implementations of functions in Velox rely on boost::regex as regular expression engine. That is not recommended. RE2 is the preferred engine.

boost relies on backtracking and it can lead to significantly slower performance compared to RE2. RE2 has O(n) complexity for a given string of length n, as it uses finite-state machines.

This change replaces boost::regex with RE2 in the ParseDurationFunction in DateTimeFunctions.h. The implementation now uses RE2::FullMatch instead of boost::regex_search, and directly captures the value and unit strings using RE2's API, simplifying the code by eliminating the need for boost::smatch.

Differential Revision: D84391289

[meta-codesync]

@abhinavmuk04 has exported this pull request. If you are a Meta employee, you can view the originating Diff in D84391289.

[netlify]

✅ Deploy Preview for meta-velox canceled.

Name	Link
🔨 Latest commit	99cdebd
🔍 Latest deploy log	https://app.netlify.com/projects/meta-velox/deploys/68f7f22063388c0008ea16b7

[MBkkt]

@abhinavmuk04 Hi, nice changes!

I have some suggestions:

1. We can use raw-string literals to avoid complicating regex pattern
2. We can avoid std::string copy for matched strings (valueStr and unit)
3. Better to use from_chars, because it works faster and doesn't throw exception, instead return result

template <typename T>
struct ParseDurationFunction {
  VELOX_DEFINE_FUNCTION_TYPES(T);

  std::unique_ptr<RE2> durationRegex_;

  FOLLY_ALWAYS_INLINE void initialize(
      const std::vector<TypePtr>& /*inputTypes*/,
      const core::QueryConfig& /*config*/,
      const arg_type<Varchar>* /*amountUnit*/) {
    durationRegex_ =
        std::make_unique<RE2>(R"(^\s*(\d+(?:\.\d+)?)\s*([a-zA-Z]+)\s*$)");
  }

  FOLLY_ALWAYS_INLINE void call(
      out_type<IntervalDayTime>& result,
      const arg_type<Varchar>& amountUnit) {
    std::string_view valueStr;
    std::string_view unit;
    if (!RE2::FullMatch(
            std::string_view{amountUnit}, *durationRegex_, &valueStr, &unit)) {
      VELOX_USER_FAIL(
          "Input duration is not a valid data duration string: {}", amountUnit);
    }

    double value{};
    auto [_, error] = std::from_chars(
        valueStr.data(), valueStr.data() + valueStr.size(), value);
    if (error != std::errc{}) {
      VELOX_USER_FAIL(
          "Input duration value is not a valid number: {}", valueStr);
    }

    result = valueOfTimeUnitToMillis(value, unit);
  }
};

[MBkkt]

@abhinavmuk04 will you check #15134 ?

[abhinavmuk04]

@abhinavmuk04 will you check #15134 ?

@MBkkt I see in this PR you seem to be handling the parse_duration function. Is this something you are working on? If so, I can let go of this PR

[abhinavmuk04]

@abhinavmuk04 will you check #15134 ?

@MBkkt I see in this PR you seem to be handling the parse_duration function. Is this something you are working on? If so, I can let go of this PR

@MBkkt wanted to follow up on this

[MBkkt]

@abhinavmuk04 Not really. If you will merge your PR, I will just rebase. If you don't merge this PR, I will wait for review for mine, to merge it and get rid of boost::regex.

Some explanation:

I just noticed your PR and thought about it will be nice to do 3 things:

1. I did some review suggestions to your PR (avoid string copies, use faster "to double conversion", single time automaton construction, etc)
2. Removed second usage of boost::regex
3. Removed boost::regex from CMake

[kevinwilfong]

LGTM Thanks @MBkkt for all the suggestions!

[kevinwilfong]

Sorry, I noticed you didn't apply the changes from @MBkkt

[kevinwilfong]

It looks like you had them in a previous commit (the one I was initially looking at) and accidentally reverted them.

[meta-codesync]

This pull request has been merged in ec15078.