ParadoxSaveParser/ParadoxSaveParser.Benchmarks/Extractors.cs

using System;
using System.Collections.Generic;
using System.IO;
using System.Text.Json;
using System.Text.RegularExpressions;
using ParadoxSaveParser.Lib;
using PCRE;

namespace ParadoxSaveParser.Benchmarks;

/// <summary>
///     Every extractor solves the same task: pull the <c>technology</c> block
///     (adm/dip/mil levels) of every country out of an EU4 save, i.e. the path
///     <c>countries.*.technology</c>.
///     They all return the same checksum so the benchmarks can be verified against each other:
///     <c>number_of_countries * 1_000_000 + sum_of_all_tech_levels</c>.
/// </summary>
public static partial class Extractors
{
    public static long Checksum(long count, long techSum) => count * 1_000_000 + techSum;

    // ---------------------------------------------------------------- this project

    /// <summary>The parser of this repo: single streaming pass, guided by a compiled search expression.</summary>
    public static long SearchExpression(byte[] pdx, ISearchExpression query)
    {
        using var stream = new MemoryStream(pdx, false);
        var root = new SaveParserEU4(stream, query).Parse();
        return ChecksumOfParsedTree(root);
    }

    /// <summary>Same query, but the parser reads from the given stream instead of a preloaded byte[].</summary>
    public static long SearchExpressionFromStream(Stream stream, ISearchExpression query)
    {
        var root = new SaveParserEU4(stream, query).Parse();
        return ChecksumOfParsedTree(root);
    }

    /// <summary>Same parser without a query: parses the whole save into memory (worst case reference point).</summary>
    public static long FullParseThenSelect(byte[] pdx)
    {
        using var stream = new MemoryStream(pdx, false);
        var root = new SaveParserEU4(stream, null).Parse();
        return ChecksumOfParsedTree(root);
    }

    private static long ChecksumOfParsedTree(Dictionary<string, object> root)
    {
        long count = 0, sum = 0;
        var countries = (Dictionary<string, object>)root["countries"];
        foreach (var country in countries.Values)
        {
            if (country is not Dictionary<string, object> c
                || !c.TryGetValue("technology", out var t)
                || t is not Dictionary<string, object> tech)
                continue;
            count++;
            foreach (var level in tech.Values)
                sum += (long)level;
        }

        return Checksum(count, sum);
    }

    // ---------------------------------------------------------------- .NET Regex

    /// <summary>
    ///     Flat scan: matches technology blocks anywhere in the file.
    ///     Fastest regex option, but it does not verify that the match really sits under <c>countries</c>.
    /// </summary>
    public const string FlatPattern =
        @"\n\t\ttechnology=\{\n\t\t\tadm_tech=(\d+)\n\t\t\tdip_tech=(\d+)\n\t\t\tmil_tech=(\d+)\n\t\t\}";

    /// <summary>
    ///     Path-aware: pairs a depth-1 country tag with the next technology block via a lazy gap.
    ///     This is what a regex user writes when the key must be associated with the value.
    /// </summary>
    public const string PathAwarePattern =
        @"\n\t([A-Z0-9]{3})=\{[\s\S]*?\n\t\ttechnology=\{\n\t\t\tadm_tech=(\d+)\n\t\t\tdip_tech=(\d+)\n\t\t\tmil_tech=(\d+)\n\t\t\}";

    /// <summary>
    ///     Balancing-group pattern isolating the whole <c>countries={...}</c> block.
    ///     .NET-only feature; PCRE would need recursion instead.
    /// </summary>
    public const string CountriesBlockPattern =
        @"\ncountries=\{(?>[^{}]+|\{(?<d>)|\}(?<-d>))*(?(d)(?!))\n\}";

    public static long RegexScan(Regex regex, string text)
    {
        long count = 0, sum = 0;
        foreach (Match m in regex.Matches(text))
        {
            count++;
            int g = m.Groups.Count - 3;
            sum += int.Parse(m.Groups[g].ValueSpan)
                   + int.Parse(m.Groups[g + 1].ValueSpan)
                   + int.Parse(m.Groups[g + 2].ValueSpan);
        }

        return Checksum(count, sum);
    }

    /// <summary>Two stages: cut out the countries block with balanced braces, then scan inside it.</summary>
    public static long RegexTwoStage(Regex blockRegex, Regex innerRegex, string text)
    {
        var block = blockRegex.Match(text);
        if (!block.Success)
            throw new Exception("countries block not found");
        return RegexScan(innerRegex, block.Value);
    }

    [GeneratedRegex(FlatPattern)]
    public static partial Regex FlatSourceGen();

    [GeneratedRegex(PathAwarePattern)]
    public static partial Regex PathAwareSourceGen();

    // ---------------------------------------------------------------- PCRE.NET

    public static long PcreScan(PcreRegex regex, string text)
    {
        long count = 0, sum = 0;
        foreach (var m in regex.Matches(text))
        {
            count++;
            int g = m.Groups.Count - 3;
            sum += int.Parse(m[g].Value)
                   + int.Parse(m[g + 1].Value)
                   + int.Parse(m[g + 2].Value);
        }

        return Checksum(count, sum);
    }

    // ---------------------------------------------------------------- System.Text.Json

    private static readonly byte[] CountriesUtf8 = "countries"u8.ToArray();
    private static readonly byte[] TechnologyUtf8 = "technology"u8.ToArray();

    /// <summary>Hand written streaming reader over the JSON twin: the closest JSON analogue of the search expression.</summary>
    public static long Utf8JsonReaderScan(byte[] json)
    {
        long count = 0, sum = 0;
        var reader = new Utf8JsonReader(json, isFinalBlock: true, state: default);
        reader.Read(); // StartObject (root)

        while (reader.Read() && reader.TokenType == JsonTokenType.PropertyName)
        {
            bool isCountries = reader.ValueTextEquals(CountriesUtf8);
            reader.Read();
            if (!isCountries)
            {
                reader.Skip();
                continue;
            }

            // inside countries: property name = country tag
            while (reader.Read() && reader.TokenType == JsonTokenType.PropertyName)
            {
                reader.Read();
                if (reader.TokenType != JsonTokenType.StartObject)
                {
                    reader.Skip();
                    continue;
                }

                while (reader.Read() && reader.TokenType == JsonTokenType.PropertyName)
                {
                    bool isTech = reader.ValueTextEquals(TechnologyUtf8);
                    reader.Read();
                    if (!isTech)
                    {
                        reader.Skip();
                        continue;
                    }

                    count++;
                    while (reader.Read() && reader.TokenType == JsonTokenType.PropertyName)
                    {
                        reader.Read();
                        sum += reader.GetInt64();
                    }
                }
            }

            break; // countries handled, nothing else is needed
        }

        return Checksum(count, sum);
    }

    /// <summary>DOM approach: parse the whole JSON document, then navigate. The .NET equivalent of what jq does.</summary>
    public static long JsonDocumentScan(byte[] json)
    {
        long count = 0, sum = 0;
        using var doc = JsonDocument.Parse(json, new JsonDocumentOptions { MaxDepth = 1024 });
        foreach (var country in doc.RootElement.GetProperty("countries").EnumerateObject())
        {
            if (country.Value.ValueKind != JsonValueKind.Object
                || !country.Value.TryGetProperty("technology", out var tech))
                continue;
            count++;
            foreach (var level in tech.EnumerateObject())
                sum += level.Value.GetInt64();
        }

        return Checksum(count, sum);
    }
}