ufo_data/ufojson.cpp

// ufojson.cpp
// Copyright (C) 2023 Richard Geldreich, Jr.

#include "utils.h"
#include "markdown_proc.h"
#include "ufojson_core.h"
#include "udb.h"
#include "converters.h"

//#include "crnlib\crn_core.h"
//#include "crnlib\crn_hash_map.h"

//-------------------------------------------------------------------

static void detect_bad_urls()
{
    string_vec unique_urls;
    bool utf8_flag = false;
    if (!read_text_file("unique_urls.txt", unique_urls, true, &utf8_flag))
        panic("Can't read file unique_urls.txt");

    uint32_t total_processed = 0;

    string_vec failed_urls;
    for (const auto& url : unique_urls)
    {
        printf("-------------- URL : %u\n", total_processed);

        string_vec reply;
        bool status = invoke_curl(url, reply);
        bool not_found = false;

        if (status)
        {
            for (auto str : reply)
            {
                str = string_lower(str);

                if ((string_find_first(str, "404 not found") != -1) ||
                    (string_find_first(str, "cannot find the requested page") != -1))
                {
                    not_found = true;
                    break;
                }
            }
        }

        if ((!status) || (not_found))
        {
            failed_urls.push_back(url);
            printf("FAILED: %s\n", url.c_str());
        }
        else
        {
            printf("SUCCEEDED: %s\n", url.c_str());
        }

        total_processed++;
        if ((total_processed % 25) == 24)
        {
            if (!write_text_file("failed_urls.txt", failed_urls, false))
                panic("Unable to create file failed_urs.txt");
        }
    }

    printf("Total urls: %zu, failed: %zu\n", unique_urls.size(), failed_urls.size());

    if (!write_text_file("failed_urls.txt", failed_urls, false))
        panic("Unable to create file failed_urs.txt");
}

// Windows defaults to code page 437:
// https://www.ascii-codes.com/
// We want code page 1252
// http://www.alanwood.net/demos/ansi.html

static bool invoke_openai(const timeline_event &event, const char *pPrompt_text, json& result)
{
    markdown_text_processor tp;
    tp.init_from_markdown(event.m_desc.c_str());

    std::string desc;
    tp.convert_to_plain(desc, true);

    if ((desc.size() >= 2) && (desc.back() == '('))
        desc.pop_back();
    
    const uint32_t MAX_SIZE = 4096; // ~1024 tokens
    if (desc.size() > MAX_SIZE)
    {
        int i;
        for (i = MAX_SIZE; i >= 0; i--)
        {
            if (desc[i] == ' ')
            {
                desc.resize(i);
                break;
            }
        }
        if (i < 0)
            panic("Failed shrinking text");
    }

    uprintf("Desc: %s\n\n", desc.c_str());

    std::string prompt_str(pPrompt_text);
    prompt_str += desc;
    prompt_str += "\"";

    std::string res_str;
    const uint32_t MAX_TRIES = 3;
    for (uint32_t try_index = 0; try_index < MAX_TRIES; try_index++)
    {
        bool status = invoke_openai(prompt_str, res_str);
        if (status)
            break;
        
        if (try_index == MAX_TRIES - 1)
        {
            fprintf(stderr, "invoke_openai() failed!\n");
            return false;
        }
    }

    uprintf("Result: %s\n", res_str.c_str());

    bool success = false;
    try
    {
        result = json::parse(res_str.begin(), res_str.end());
        success = true;
    }
    catch (json::exception& e)
    {
        fprintf(stderr, "json::parse() failed (id %i): %s", e.id, e.what());
        return false;
    }

    if (!result.is_object() || !result.size())
    {
        fprintf(stderr, "Invalid JSON!\n");
        return false;
    }
    
    return true;
}

static void process_timeline_using_openai(const ufo_timeline &timeline)
{
    bool utf8_flag;
    json existing_results;
    load_json_object("openai_results.json", utf8_flag, existing_results);
        
    json final_result = json::object();

    final_result["results"] = json::array();

    auto& results_array = final_result["results"];

    uint32_t total_failures = 0, total_hits = 0, total_new_results = 0;
    //for (uint32_t i = 0; i < timeline.size(); i++)
    for (uint32_t i = 8515; i < 8516; i++)
    {
        uprintf("****** Record: %u\n", i);

        const timeline_event& event = timeline[i];
        const uint32_t event_crc32 = event.get_crc32();

        json result_obj;
        bool found_flag = false;

        if (existing_results.is_object() && existing_results.size())
        {
            const auto& results_arr = existing_results["results"];
            for (auto it = results_arr.begin(); it != results_arr.end(); ++it)
            {
                const auto& obj = *it;
                if (obj.is_object())
                {
                    uint32_t peek_obj_crc32 = obj["event_crc32"].get<uint32_t>();
                    if (peek_obj_crc32 == event_crc32)
                    {
                        if (obj["event_source_id"] == event.m_source_id)
                        {
                            result_obj = obj;
                            found_flag = true;
                            break;
                        }
                    }
                }
            }
        }

        if (found_flag)
        {
            total_hits++;
            printf("Found hit for record %u\n", i);

            results_array.push_back(result_obj);

            continue;
        }

        if (!invoke_openai(event,
            //"Compose a JSON object from the following quoted text. Give an array (with the key \"locations\") containing all the geographic locations or addresses. Then an array (with the key \"index\") with all phrases that would appear in a book index. Then an array (with the key \"names\") with all the person names. \"",
            "Compose a JSON object from the following quoted text. Give an array (with the key \"locations\") containing all the geographic locations or addresses. Then an array (with the key \"index\") with all phrases that would appear in a book index. Then an array (with the key \"names\") with all the person names. For the locations, aggressively provide as few strings as possible so they can be geocoded. For example, if the input has \"County Waterford in Cappoquin, Ireland\", provide \"County Waterford, Cappoquin, Ireland\". \"",
            result_obj))
        {
            fprintf(stderr, "Invoke AI failed!");

            total_failures++;
            continue;
        }

        result_obj["event_index"] = i;

        result_obj["event_date_str"] = event.m_date_str;

        if (event.m_time_str.size())
            result_obj["event_time"] = event.m_time_str;

        result_obj["event_source"] = event.m_source;
        result_obj["event_source_id"] = event.m_source_id;

        result_obj["event_crc32"] = event_crc32;

        results_array.push_back(result_obj);

        total_new_results++;

        if ((total_new_results % 50) == 0)
        {
            serialize_to_json_file("temp_results.json", final_result, true);
        }
    }

    if (!results_array.size())
        panic("No results");

    if (!serialize_to_json_file("new_openai_results.json", final_result, true))
        panic("failed writing results.json");

    uprintf("Total hits: %u, new results: %u, failures: %u\n", total_hits, total_new_results, total_failures);

    uprintf("Success\n");
}

static void process_timeline_using_python(const ufo_timeline& timeline)
{
    json final_result = json::object();

    final_result["results"] = json::array();

    auto& results_array = final_result["results"];

    for (uint32_t i = 0; i < timeline.size(); i++)
    //for (uint32_t i = 0; i < 5; i++)
    {
        uprintf("****** Record: %u\n", i);

        const timeline_event& event = timeline[i];

        string_vec lines;
        lines.push_back(event.m_desc);

        if (!write_text_file("input.txt", lines))
            panic("Failed writing input.txt");

        remove("locations.json");

        Sleep(50);
                
        int status = system("python.exe pextractlocs.py");
        if (status != EXIT_SUCCESS)
            panic("Failed running python.exe");

        json locs;
        bool utf8_flag;
        if (!load_json_object("locations.json", utf8_flag, locs))
        {
            Sleep(50);

            if (!load_json_object("locations.json", utf8_flag, locs))
                panic("failed reading locations.json");
        }

        for (auto it = locs.begin(); it != locs.end(); ++it)
        {
            if (it->is_string())
                uprintf("%s\n", it->get<std::string>().c_str());
        }
                
        json new_obj = json::object();
        new_obj.emplace("index", i);
        new_obj.emplace("date", event.m_date_str);
        new_obj.emplace("crc32", event.get_crc32());
        new_obj.emplace("locations", locs);

        results_array.push_back(new_obj);
    }

    if (!results_array.size())
        panic("No results");

    if (!serialize_to_json_file("new_results.json", final_result, true))
        panic("failed writing results.json");
}

enum
{
    gn_geonameid, // integer id of record in geonames database
    gn_name, // name of geographical point(utf8) varchar(200)
    gn_asciiname, // name of geographical point in plain ascii characters, varchar(200)
    gn_alternatenames, // alternatenames, comma separated, ascii names automatically transliterated, convenience attribute from alternatename table, varchar(10000)
    gn_latitude, // latitude in decimal degrees(wgs84)
    gn_longitude, // longitude in decimal degrees(wgs84)
    gn_feature_class, // see http ://www.geonames.org/export/codes.html, char(1)
    gn_feature_code, // see http ://www.geonames.org/export/codes.html, varchar(10)
    gn_country_code, // ISO - 3166 2 - letter country code, 2 characters
    gn_cc2, // alternate country codes, comma separated, ISO - 3166 2 - letter country code, 200 characters
    gn_admin1_code, // fipscode(subject to change to iso code), see exceptions below, see file admin1Codes.txt for display names of this code; varchar(20)
    gn_admin2_code, // code for the second administrative division, a county in the US, see file admin2Codes.txt; varchar(80)
    gn_admin3_code, // code for third level administrative division, varchar(20)
    gn_admin4_code, // code for fourth level administrative division, varchar(20)
    gn_population, // bigint(8 byte int)
    gn_elevation, // in meters, integer
    gn_dem, // digital elevation model, srtm3 or gtopo30, average elevation of 3''x3'' (ca 90mx90m) or 30''x30'' (ca 900mx900m) area in meters, integer.srtm processed by cgiar / ciat.
    gn_timezone, // the iana timezone id(see file timeZone.txt) varchar(40)
    gn_modification_date,
    gn_total
};

struct geoname
{
    std::string m_fields[gn_total];
};

typedef std::vector<geoname> geoname_vec;

static bool is_important_country(const std::string& s)
{
    return (s == "US") || (s == "GB") || (s == "AU") || (s == "CA") || (s == "NZ") || (s == "FR") || (s == "DE") || (s == "BR") || (s == "IT");
}

static bool is_favored_country(const std::string& s)
{
    return (s == "US") || (s == "GB") || (s == "AU") || (s == "CA") || (s == "NZ") || (s == "FR") || (s == "DE");
}

const uint32_t TOTAL_FAVORED_COUNTRY_RANKS = 8;

static int get_favored_country_rank(const std::string& s)
{
    if (s == "US")
        return 0;
    else if (s == "CA")
        return 1;
    else if (s == "GB")
        return 2;
    else if (s == "AU")
        return 3;
    else if (s == "NZ")
        return 4;
    else if (s == "FR")
        return 5;
    else if (s == "DE")
        return 6;

    return 7;
}

static bool is_country_fcode(const std::string &fcode)
{
    return ((fcode == "PCL") || (fcode == "PCLD") || (fcode == "PCLF") || (fcode == "PCLH") || (fcode == "PCLI") || (fcode == "PCLIX") || (fcode == "PCLS") || (fcode == "TERR"));
}

static void process_geodata()
{
    string_vec lines;
    lines.reserve(13000000);

    if (!read_text_file("allCountries.txt", lines, true, nullptr))
        panic("failed reading allCountries.txt");

    int_vec tab_locs;
    tab_locs.reserve(32);

    geoname_vec geonames;
    geonames.resize(13000000);

    uint32_t total_geonames = 0;
    
    uint32_t max_col_sizes[gn_total];
    clear_obj(max_col_sizes);

    std::unordered_map<char, uint32_t> accepted_class_counts;
    std::unordered_map<char, uint32_t> rejected_class_counts;

    uint32_t total_accepted = 0;

    json output_json = json::array();
        
    for (const auto& str : lines)
    {
        tab_locs.resize(0);

        for (uint32_t ofs = 0; ofs < str.size(); ofs++)
            if (str[ofs] == '\t')
                tab_locs.push_back(ofs);

        if (tab_locs.size() != 18)
            panic("Unexpected number of tabs");

        tab_locs.push_back((uint32_t)str.size());

        geoname& g = geonames[total_geonames];
        uint32_t cur_ofs = 0;
        for (uint32_t i = 0; i < gn_total; i++)
        {
            g.m_fields[i] = string_slice(str, cur_ofs, tab_locs[i] - cur_ofs);

#if 0
            const uint32_t MAX_ALT_NAMES_SIZE = 1024;
            if ((i == gn_alternatenames) && (g.m_fields[i].size() > MAX_ALT_NAMES_SIZE))
            {
                int j;
                for (j = MAX_ALT_NAMES_SIZE; j >= 0; j--)
                {
                    if (g.m_fields[i][j] == ',')
                    {
                        g.m_fields[i].resize(j);
                        break;
                    }
                }
                if (j < 0)
                    g.m_fields[i].resize(0);
            }
#endif

            max_col_sizes[i] = std::max(max_col_sizes[i], (uint32_t)g.m_fields[i].size());
            
            cur_ofs = tab_locs[i] + 1;
        }

        bool accept_flag = false;
        bool has_min_pop = false;
        if (g.m_fields[gn_population].size())
        {
            int pop = atoi(g.m_fields[gn_population].c_str());
            
            const int MIN_POP = 10;
            if (pop >= MIN_POP)
                has_min_pop = true;
        }

        const auto& code = g.m_fields[gn_feature_code];

        char feature_class = ' ';
        if (g.m_fields[gn_feature_class].size() == 1)
            feature_class = g.m_fields[gn_feature_class][0];

        switch (feature_class)
        {
        case 'T': // mountain,hill,rock,...
            if ((code == "MT") || (code == "MTS") || (code == "ATOL") || (code == "CAPE") || (code == "CNYN") || (code == "DSRT") || 
                (code == "ISL") || (code == "ISLS") || (code == "PEN") || (code == "VALS") || (code == "VALX"))
            {
                accept_flag = true;
            }
            break;
        case 'S': // spot, building, farm
            if ((code == "AIRB") || (code == "AIRF") || (code == "AIRP") || (code == "AIRQ") || (code == "BRKS") || (code == "CTRA") ||
                (code == "CTRS") || (code == "INSM") || (code == "ITTR") || (code == "PSN") || (code == "STNE") || (code == "USGE") ||
                (code == "OBS") || (code == "OBSR") || (code == "MFGM") || (code == "FT") || (code == "ASTR") || (code == "FCL") || 
                (code == "PS") || (code == "PSH") || (code == "STNB") || (code == "STNS") || (code == "UNIV"))
            {
                accept_flag = true;
            }
            break;
        case 'L': // parks,area, ...
            if ((code == "MILB") || (code == "MVA") || (code == "NVB") || (code == "BTL") || (code == "CST") || (code == "RNGA") || (code == "PRK") || (code == "RESV"))
            {
                accept_flag = true;
            }
            break;
        case 'A': // country, state, region,...
            accept_flag = true;
            //accept_flag = has_min_pop;
            break;
        case 'H': // stream, lake, ...
            if ((code == "BAY") || (code == "BAYS") || (code == "CHN") || (code == "CHNL") || (code == "CHNM") || (code == "CHNN") ||
                (code == "CNL") || (code == "CNL") || (code == "LK") || (code == "LKN") || (code == "LKS") || (code == "RSV") || (code == "SD") || (code == "STRT"))
            {
                accept_flag = true;
            }
            break;
        case 'P': // city, village,...
            if ((code == "PPLC") || (code == "PPLG") || (code == "PPLA")) // TODO
                accept_flag = true;
            else if (is_important_country(g.m_fields[gn_country_code]))
                accept_flag = true;
            else
                accept_flag = has_min_pop;
            break;
        case 'R': // road, railroad
            break;
        case 'U': // undersea
            break;
        case 'V': // forest,heath,...
            break;
        default:
            break;
        }

        if (accept_flag)
        {
            accepted_class_counts[feature_class] = accepted_class_counts[feature_class] + 1;

            json obj = json::object();

            obj["id"] = g.m_fields[gn_geonameid].size() ? atoi(g.m_fields[gn_geonameid].c_str()) : -1;
            obj["name"] = g.m_fields[gn_name];
            obj["plainname"] = g.m_fields[gn_asciiname];
            
            if (g.m_fields[gn_alternatenames].size())
                obj["altnames"] = g.m_fields[gn_alternatenames];

            if (g.m_fields[gn_feature_class].size())
                obj["fclass"] = g.m_fields[gn_feature_class];

            if (g.m_fields[gn_feature_code].size())
                obj["fcode"] = g.m_fields[gn_feature_code];

            if (g.m_fields[gn_country_code].size())
                obj["ccode"] = g.m_fields[gn_country_code];
            
            if (g.m_fields[gn_cc2].size())
                obj["cc2"] = g.m_fields[gn_cc2];
            
            if (g.m_fields[gn_admin1_code].size())
                obj["a1"] = g.m_fields[gn_admin1_code];

            if (g.m_fields[gn_admin2_code].size())
                obj["a2"] = g.m_fields[gn_admin2_code];

            if (g.m_fields[gn_admin3_code].size())
                obj["a3"] = g.m_fields[gn_admin3_code];

            if (g.m_fields[gn_admin4_code].size())
                obj["a4"] = g.m_fields[gn_admin4_code];

            obj["lat"] = g.m_fields[gn_latitude].size() ? atof(g.m_fields[gn_latitude].c_str()) : 0.0f;
            obj["long"] = g.m_fields[gn_longitude].size() ? atof(g.m_fields[gn_longitude].c_str()) : 0.0f;

            if (g.m_fields[gn_population].size())
                obj["pop"] = atoi(g.m_fields[gn_population].c_str());

            output_json.push_back(obj);

            total_accepted++;

            if ((total_accepted % 1000) == 0)
                uprintf("Total accepted: %u\n", total_accepted);
        }
        else
        {
            rejected_class_counts[feature_class] = rejected_class_counts[feature_class] + 1;
        }
        
        total_geonames++;

        if ((total_geonames % 1000000) == 0)
            uprintf("Geonames processed: %u\n", total_geonames);
    }

    if (!serialize_to_json_file("world_features.json", output_json, true))
        panic("failed writing to file world_features.json");

    uprintf("Total accepted: %u\n", total_accepted);

    for (uint32_t i = 0; i < gn_total; i++)
        uprintf("%u\n", max_col_sizes[i]);

    uprintf("Accepted:\n");
    for (const auto& s : accepted_class_counts)
        uprintf("%c %u\n", s.first, s.second);

    uprintf("Rejected:\n");
    for (const auto& s : rejected_class_counts)
        uprintf("%c %u\n", s.first, s.second);
}

static const struct
{
    const char* m_pCode;
    int m_level;
} g_geocode_levels[] = 
{
    { "ADM1", 1 },
    { "ADM1H", 1 },

    { "ADM2", 2 },
    { "ADM2H", 2 },

    { "ADM3", 3 },
    { "ADM3H", 3 },

    { "ADM4", 4 },
    { "ADM4H", 4 },

    { "ADM5", 5 },
    { "ADM5H", 5 },

    { "ADMD", 1 },
    { "ADMDH", 1 },

    { "PCL", 0 },
    { "PCLD", 0 },
    { "PCLF", 0 },
    { "PCLH", 0 },
    { "PCLI", 0 },
    { "PCLIX", 0 },
    { "PCLS", 0 },
    { "PRSH", 0 },
    { "TERR", 0 },
    { "ZN", 0 },
    { "ZNB", 0 }
};

const int MIN_GEOCODE_LEVEL = 0, MAX_GEOCODE_LEVEL = 5, NUM_GEOCODE_LEVELS = 6;

static int find_geocode_admin_level(const char* pCode)
{
    for (const auto& l : g_geocode_levels)
        if (_stricmp(pCode, l.m_pCode) == 0)
            return l.m_level;

    return -1;
}

struct country_info
{
    std::string m_iso2_name;
    std::string m_iso3_name;
    std::string m_iso_numeric;
    std::string m_fips;
    std::string m_country_name;
    std::string m_capital_name;
    int m_area;
    int m_pop;
    std::string m_continent;
    int m_geoid;
    int m_rec_index;
    std::string m_neighbors;
    int m_pop_rank;
};

typedef std::vector<country_info> country_info_vec;

static int get_admin_level(const std::string& code)
{
    if (code == "ADM1")
        return 1;
    else if (code == "ADM2")
        return 2;
    else if (code == "ADM3")
        return 3;
    else if (code == "ADM4")
        return 4;
    else if (code == "ADM5")
        return 5;

    if (code == "ADM1H")
        return 1;
    else if (code == "ADM2H")
        return 2;
    else if (code == "ADM3H")
        return 3;
    else if (code == "ADM4H")
        return 4;
    else if (code == "ADM5H")
        return 5;

    return 0;
}

class geodata
{
    enum { HASH_BITS = 23U, HASH_SHIFT = 32U - HASH_BITS, HASH_FMAGIC = 2654435769U, MAX_EXPECTED_RECS = 3000000 };

public:
    geodata()
    {
        assert((1U << HASH_BITS) >= MAX_EXPECTED_RECS * 2U);
    }

    void init()
    {
        uprintf("Reading hierarchy\n");

        load_hierarchy();

        uprintf("Reading world_features.json\n");
                
        if (!read_text_file("world_features.json", m_filebuf, nullptr))
            panic("Failed reading file");

        uprintf("Deserializing JSON file\n");
                
        bool status = m_doc.deserialize_in_place((char*)&m_filebuf[0]);
        if (!status)
            panic("Failed parsing JSON document!");

        if (!m_doc.is_array())
            panic("Invalid JSON");

        uprintf("Hashing JSON data\n");

        m_name_hashtab.resize(0);
        m_name_hashtab.resize(1U << HASH_BITS);

        const auto& root_arr = m_doc.get_array();

        //crnlib::timer tm;
        //tm.start();

        uint8_vec name_buf;
                
        m_geoid_to_rec.clear();
        m_geoid_to_rec.reserve(MAX_EXPECTED_RECS);
        
        for (uint32_t rec_index = 0; rec_index < root_arr.size(); rec_index++)
        {
            const auto& arr_entry = root_arr[rec_index];
            if (!arr_entry.is_object())
                panic("Invalid JSON");
            
            int geoid = arr_entry.find_int32("id");
            assert(geoid > 0);
            auto ins_res = m_geoid_to_rec.insert(std::make_pair(geoid, (int)rec_index));
            if (!ins_res.second)
                panic("Invalid id!");

            const auto pName = arr_entry.find_value_variant("name");
            const auto pAltName = arr_entry.find_value_variant("altnames");

            if ((pName == nullptr) || (!pName->is_string()))
                panic("Missing/invalid name field");

            {
                const char* pName_str = pName->get_string_ptr();
                size_t name_size = strlen(pName_str);

                if (name_size > name_buf.size())
                    name_buf.resize(name_size);

                for (uint32_t ofs = 0; ofs < name_size; ofs++)
                    name_buf[ofs] = utolower(pName_str[ofs]);

                uint32_t hash_val = (hash_hsieh(&name_buf[0], name_size) * HASH_FMAGIC) >> HASH_SHIFT;
                m_name_hashtab[hash_val].push_back(rec_index);
            }

            const auto pPlainName = arr_entry.find_value_variant("plainname");
            if ((pPlainName == nullptr) || (!pPlainName->is_string()))
                panic("Missing/invalid plainname field");
            
            {
                const char* pName_str = pPlainName->get_string_ptr();
                size_t name_size = strlen(pName_str);

                if (name_size > name_buf.size())
                    name_buf.resize(name_size);

                for (uint32_t ofs = 0; ofs < name_size; ofs++)
                    name_buf[ofs] = utolower(pName_str[ofs]);

                uint32_t hash_val = (hash_hsieh(&name_buf[0], name_size) * HASH_FMAGIC) >> HASH_SHIFT;
                m_name_hashtab[hash_val].push_back(rec_index);
            }

            if (pAltName)
            {
                if (!pAltName->is_string())
                    panic("Missing/invalid altname field");

                const char* pAlt_name_str = pAltName->get_string_ptr();
                size_t alt_name_size = strlen(pAlt_name_str);

                if (alt_name_size > name_buf.size())
                    name_buf.resize(alt_name_size);

                for (uint32_t ofs = 0; ofs < alt_name_size; ofs++)
                    name_buf[ofs] = utolower(pAlt_name_str[ofs]);

                uint32_t start_ofs = 0;

                for (uint32_t i = 0; i < alt_name_size; i++)
                {
                    if (pAlt_name_str[i] == ',')
                    {
                        size_t size = i - start_ofs;
                        assert(size);

                        uint32_t hash_val = (hash_hsieh(&name_buf[0] + start_ofs, size) * HASH_FMAGIC) >> HASH_SHIFT;
                        m_name_hashtab[hash_val].push_back(rec_index);

                        start_ofs = i + 1;
                    }
                }

                size_t size = alt_name_size - start_ofs;
                assert(size);

                uint32_t hash_val = (hash_hsieh(&name_buf[0] + start_ofs, size) * HASH_FMAGIC) >> HASH_SHIFT;
                m_name_hashtab[hash_val].push_back(rec_index);
            }

            std::string fclass = arr_entry.find_string_obj("fclass");
            
            if (fclass == "A")
            {
                std::string fcode(arr_entry.find_string_obj("fcode"));
                
                if ((fcode == "ADM1") || (fcode == "ADM2") || (fcode == "ADM3") || (fcode == "ADM4")) 
                {
                    std::string ccode(arr_entry.find_string_obj("ccode"));

                    std::string a[4] = {
                        arr_entry.find_string_obj("a1"),
                        arr_entry.find_string_obj("a2"),
                        arr_entry.find_string_obj("a3"),
                        arr_entry.find_string_obj("a4") };

                    std::string desc(ccode);

                    for (uint32_t i = 0; i < 4; i++)
                    {
                        if (!a[i].size())
                            break;
                        desc += "." + a[i];
                    }
                                                            
                    m_admin_map[desc].push_back(std::pair<int, int>(rec_index, get_admin_level(fcode)));
                }
            }
        }

#if 0
        if (m_admin_map.count(desc))
        {
            const int alt_admin_level = get_admin_level(fcode);

            const int cur_rec_index = m_admin_map.find(desc)->second;
            const pjson::value_variant* pCur = &m_doc[cur_rec_index];
            const int cur_admin_level = get_admin_level(pCur->find_string_obj("fcode"));

            if (alt_admin_level < cur_admin_level)
                m_admin_map[desc] = rec_index;
            else if (alt_admin_level == cur_admin_level)
            {
                uprintf("ccode: %s rec: %u geoid: %u name: %s fcode: %s, current rec: %u geoid: %u name: %s fcode: %s\n",
                    ccode.c_str(),
                    rec_index, geoid, pName->get_string_ptr(), fcode.c_str(),
                    cur_rec_index, pCur->find_int32("id"), pCur->find_string_obj("name").c_str(), pCur->find_string_obj("fcode").c_str());
            }
        }
        else
#endif

        for (auto it = m_admin_map.begin(); it != m_admin_map.end(); it++)
        {
            std::vector< std::pair<int, int> >& recs = it->second;

            std::sort(recs.begin(), recs.end(), 
                [](const std::pair<int, int>& a, const std::pair<int, int>& b) -> bool
                {
                    return a.second < b.second;
                });

#if 0
            uprintf("%s:\n", it->first.c_str());
            for (uint32_t i = 0; i < recs.size(); i++)
            {
                const int cur_rec_index = recs[i].first;
                const pjson::value_variant* pCur = &m_doc[cur_rec_index];
                                
                uprintf("admlevel: %u, rec: %u geoid: %u name: %s fcode: %s\n",
                    recs[i].second,
                    cur_rec_index, pCur->find_int32("id"), pCur->find_string_obj("name").c_str(), pCur->find_string_obj("fcode").c_str());
            }
#endif
        }

        for (uint32_t i = 0; i < m_name_hashtab.size(); i++)
        {
            if (!m_name_hashtab[i].size())
                continue;

            std::sort(m_name_hashtab[i].begin(), m_name_hashtab[i].end());
            auto new_end = std::unique(m_name_hashtab[i].begin(), m_name_hashtab[i].end());
            m_name_hashtab[i].resize(new_end - m_name_hashtab[i].begin());
        }

        load_country_info();

        //uprintf("geodata::init: Elapsed time: %f\n", tm.get_elapsed_secs());
    }

    void find(const char* pKey, std::vector<const pjson::value_variant*>& results, std::vector<const pjson::value_variant*>& alt_results) const
    {
        assert(m_name_hashtab.size());

        std::string key(pKey);
        for (char& c : key)
            c = utolower(c);

        const uint32_t hash_val = (hash_hsieh((const uint8_t *)key.c_str(), key.size()) * HASH_FMAGIC) >> HASH_SHIFT;
        
        results.resize(0);
        alt_results.resize(0);

        const uint_vec& bucket = m_name_hashtab[hash_val];

        for (uint32_t i = 0; i < bucket.size(); i++)
        {
            const uint32_t rec_index = bucket[i];
            const pjson::value_variant* pObj = &m_doc[rec_index];

            const char *pName = pObj->find_string_ptr("name");
            
            const char* pPlainName = pObj->find_string_ptr("plainname");

            if ((_stricmp(pKey, pName) != 0) && (_stricmp(pKey, pPlainName) != 0))
            {
                const std::string alt_names(pObj->find_string_obj("altnames"));

                string_vec alt_tokens;
                string_tokenize(alt_names, ",", "", alt_tokens);

                uint32_t j;
                for (j = 0; j < alt_tokens.size(); j++)
                    if (string_icompare(alt_tokens[j], pKey) == 0)
                        break;

                if (j < alt_tokens.size())
                    alt_results.push_back(pObj);
            }
            else
            {
                results.push_back(pObj);
            }
        }
    }

    const pjson::document& get_doc() const { return m_doc; }

    struct geo_result
    {
        const pjson::value_variant* m_pVariant;
        bool m_alt;
    };
    typedef std::vector<geo_result> geo_result_vec;

    enum { MAX_ADMINS = 4 };

    uint32_t count_admins(const pjson::value_variant* p) const
    {
        uint32_t num_admins;
        for (num_admins = 0; num_admins < MAX_ADMINS; num_admins++)
        {
            std::string str(p->find_string_obj(string_format("a%u", num_admins + 1).c_str()));

            if (!str.size() || (str == "00"))
                break;
        }

        return num_admins;
    }

    bool is_valid_parent(const pjson::value_variant* pParent, const pjson::value_variant* pChild) const
    {
        if (pParent == pChild)
            return false;

        if (pParent->find_string_obj("ccode") != pChild->find_string_obj("ccode"))
            return false;

        const int parent_id = pParent->find_int32("id");
        const auto find_res = m_geo_hierarchy.find(parent_id);
        if (find_res != m_geo_hierarchy.end())
        {
            const int child_id = pChild->find_int32("id");

            const int_vec& child_indices = find_res->second;

            for (uint32_t i = 0; i < child_indices.size(); i++)
                if (child_indices[i] == child_id)
                    return true;
        }

        const uint32_t num_parent_admins = count_admins(pParent);
        const uint32_t num_child_admins = count_admins(pChild);

        if (num_parent_admins > num_child_admins)
            return false;
        
        // Example: Anderson, Shasta County, California
        if (num_parent_admins == num_child_admins)
        {
            // TODO: check fcode, child should be higher ADM level vs. parent (i.e. ADM4 vs. AMD1) for 2035607 vs. 8648889
            if (pParent->find_string_obj("fclass") != "A")
                return false;
        }

        for (uint32_t admin_index = 0; admin_index < num_parent_admins; admin_index++)
        {
            std::string id(string_format("a%u", admin_index + 1));
            
            std::string admin_parent(pParent->find_string_obj(id.c_str()));
            std::string admin_child(pChild->find_string_obj(id.c_str()));

            if (admin_parent != admin_child)
                return false;
        }

        return true;
    }

    bool is_valid_candidate(const std::vector<geo_result_vec> &temp_results, uint32_t tok_index, uint32_t result_index) const
    {
        if (!tok_index)
            return true;

        const uint32_t parent_tok_index = tok_index - 1;

        for (uint32_t i = 0; i < temp_results[parent_tok_index].size(); i++)
        {
            if (!is_valid_parent(temp_results[parent_tok_index][i].m_pVariant, temp_results[tok_index][result_index].m_pVariant))
                continue;

            if (!is_valid_candidate(temp_results, parent_tok_index, i))
                continue;

            return true;
        }

        return false;
    }

    enum
    {
        cRankNamedLocAlt = 0,   // alt

        cRankNamedLoc,          // prim

        cRankNaturalFeature0,   // prim or alt
        cRankNaturalFeature1,   // prim or alt

        cRankVillageNoPopAlt,        // alt

        cRankAdminNoPop,        // not a numbered admin
        
        cRankPopVillageAlt,     // prim, 1-100
        cRankTownAlt,           // alt, 100+

        cRankCityLevel0Alt,      // alt or alt, 1k+
        cRankCityLevel1Alt,      // alt or alt, 10k+
                                        
        cRankAdminCapital4Alt,  // alt cap4
        cRankAdmin4Alt,         // alt admin4
                
        cRankAdminCapital3Alt,  // alt cap3
        cRankAdmin3Alt,         // alt amind3

        cRankCityLevel2Alt,      // alt or alt, 100k+
        cRankCityLevel3Alt,      // alt or alt, 1mk+

        cRankVillageNoPop,      // prim no pop

        cRankAdmin,             // not numbered, has pop
                
        cRankPopVillage,        // prim, 1-100
        cRankTown,              // prim, 100+
        
        cRankAdminCapital2Alt, // alt county seat
        cRankAdmin2Alt,        // alt county

        cRankAdminCapital4,     // prim cap4
        cRankAdmin4,            // prim admin4

        cRankPark,              // prim or alt
        cRankReserve,           // prim or alt
        
        cRankAdminCapital1Alt, // alt state cap
                                
        cRankCityLevel0,        // prim or alt, 1k+
        cRankCityLevel1,        // prim or alt, 10k+

        cRankAdminCapital3,     // prim cap3
        cRankAdmin3,            // prim admin3

        cRankCityLevel2,        // prim or alt, 100k+
        cRankCityLevel3,        // prim or alt, 1m+

        cRankBaseOrAirport,    // prim or alt
                                
        cRankAdminCapital2, // prim county seat
        cRankAdmin2,        // prim county 
                
        cRankAdminCapital1, // prim state cap
        
        cRankAdmin1Alt,      // alt state
                
        cRankPoliticalCapital,  // prim or alt
        cRankGovernmentCapital, // prim or alt
                
        cRankAdmin1,        // prim state
        
        // all countries prim or alt
        cRankCountryLevel0,
        cRankCountryLevel1,
        cRankCountryLevel2,
        cRankCountryLevel3,
        cRankCountryLevel4,
        cRankCountryLevel5,
        cRankCountryLevel6,
        cRankCountryLevel7,
        cRankCountryLevel8,
        cRankCountryLevel9,
        
        cRankTotal,
    };
        
    int get_rank(const pjson::value_variant* p, bool alt_match) const
    {
        int country_index = get_country_index(p);
        if (country_index >= 0)
            return cRankCountryLevel0 + m_countries[country_index].m_pop_rank;

        //uint32_t admin_count = count_admins(p);

        std::string fclass(p->find_string_obj("fclass"));
        std::string fcode(p->find_string_obj("fcode"));

        if (fclass == "H")
        {
            //if ((code == "BAY") || (code == "BAYS") || (code == "CHN") || (code == "CHNL") || (code == "CHNM") || (code == "CHNN") ||
            //    (code == "CNL") || (code == "CNL") || (code == "LK") || (code == "LKN") || (code == "LKS") || (code == "RSV") || (code == "SD") || (code == "STRT"))

            if ((fcode == "LK") || (fcode == "LKS") || (fcode == "HBR") || (fcode == "BAY") || (fcode == "BAYS"))
                return cRankNaturalFeature1;
            else
                return cRankNaturalFeature0;
        }
        else if (fclass == "S")
        {
            //if ((code == "AIRB") || (code == "AIRF") || (code == "AIRP") || (code == "AIRQ") || (code == "BRKS") || (code == "CTRA") ||
            //    (code == "CTRS") || (code == "INSM") || (code == "ITTR") || (code == "PSN") || (code == "STNE") || (code == "USGE"))

            return cRankBaseOrAirport;
        }
        else if (fclass == "L")
        {
            //if ((code == "MILB") || (code == "MVA") || (code == "NVB") || (code == "BTL") || (code == "CST") || (code == "RNGA") || (code == "PRK") || (code == "RESV"))

            if ((fcode == "MILB") || (fcode == "NVB") || (fcode == "RNGA"))
                return cRankBaseOrAirport;

            if (fcode == "RESV")
                return cRankReserve;

            return cRankPark;
        }
        else if (fclass == "A")
        {
            // countries are all fclass A, a1=00
            // fcode must be PCL, PCLD, PCLF, PCLH, PCLI, PCLIX, PCLS, TERR

            if ((fcode == "PCL") || (fcode == "PCLD") || (fcode == "PCLF") || (fcode == "PCLH") || (fcode == "PCLI") || (fcode == "PCLIX") || (fcode == "PCLS") || (fcode == "TERR"))
                return cRankCountryLevel0;

            if ((fcode == "ADM1") || (fcode == "ADM1H"))
                return alt_match ? cRankAdmin1Alt : cRankAdmin1;

            if ((fcode == "ADM2") || (fcode == "ADM2H"))
                return alt_match ? cRankAdmin2Alt : cRankAdmin2;

            if ((fcode == "ADM3") || (fcode == "ADM3H"))
                return alt_match ? cRankAdmin3Alt : cRankAdmin3;

            if ((fcode == "ADM4") || (fcode == "ADM4H"))
                return alt_match ? cRankAdmin4Alt : cRankAdmin4;

            int pop = p->find_int32("pop");

            if (!pop)
                return cRankAdminNoPop;
            else
                return cRankAdmin;
        }
        else if (fclass == "P")
        {
            if (fcode == "PPLA")
                return alt_match ? cRankAdminCapital1Alt : cRankAdminCapital1;
            else if (fcode == "PPLA2")
                return alt_match ? cRankAdminCapital2Alt : cRankAdminCapital2;
            else if (fcode == "PPLA3")
                return alt_match ? cRankAdminCapital3Alt : cRankAdminCapital3;
            else if ((fcode == "PPLA4") || (fcode == "PPLA5"))
                return alt_match ? cRankAdminCapital4Alt : cRankAdminCapital4;

            if ((fcode == "PPLC") || (fcode == "PPLCH"))
                return cRankPoliticalCapital;

            if (fcode == "PPLG")
                return cRankGovernmentCapital;

            int pop = p->find_int32("pop");

            if (pop > 1000000)
                return alt_match ? cRankCityLevel3Alt : cRankCityLevel3;
            else if (pop >= 100000)
                return alt_match ? cRankCityLevel2Alt : cRankCityLevel2;
            else if (pop >= 10000)
                return alt_match ? cRankCityLevel1Alt : cRankCityLevel1;
            else if (pop >= 1000)
                return alt_match ? cRankCityLevel0Alt : cRankCityLevel0;
            else if (pop >= 100)
                return alt_match ? cRankTownAlt : cRankTown;
            else if (pop)
                return alt_match ? cRankPopVillageAlt : cRankPopVillage;
            else
                return alt_match ? cRankVillageNoPopAlt : cRankVillageNoPop;
        }
        else if (fclass == "T")
        {
            //if ((code == "MT") || (code == "MTS") || (code == "ATOL") || (code == "CAPE") || (code == "CNYN") || (code == "DSRT") ||
            //    (code == "ISL") || (code == "ISLS") || (code == "PEN") || (code == "VALS") || (code == "VALX"))

            if ((fcode == "ISL") || (fcode == "ISLS") || (fcode == "MT") || (fcode == "MTS"))
                return cRankNaturalFeature1;
            else
                return cRankNaturalFeature0;
        }
        else if (fclass == "V")
        {
            return cRankNaturalFeature0;
        }

        return cRankNamedLoc;
    }

    bool is_country(const pjson::value_variant* p) const
    {
        int rank = get_rank(p, false);
        return (rank >= cRankCountryLevel0);
    }

    struct resolve_results
    {
        resolve_results() 
        {
            clear();
        }
        
        void clear()
        {
            m_candidates.resize(0);
            m_sorted_results.resize(0);
            m_best_result.m_pVariant = nullptr;
            m_best_sorted_result_index = 0;
            m_best_score = 0.0f;
            m_num_input_tokens = 0;
            m_strong_match = false;
        }

        geo_result m_best_result;
        
        uint32_t m_num_input_tokens;
        bool m_strong_match;
        
        geo_result_vec m_candidates;
        std::vector< std::pair<uint32_t, float> > m_sorted_results;
        uint32_t m_best_sorted_result_index;
        float m_best_score;
    };
        
    bool resolve(const std::string& str, resolve_results &resolve_res) const
    {
        uprintf("--- Candidates for query: %s\n", str.c_str());

        resolve_res.clear();

        string_vec toks;
        string_tokenize(str, ",", "", toks);

        for (auto& tok_str : toks)
        {
            string_vec temp_toks;
            string_tokenize(tok_str, " \t", "", temp_toks);

            tok_str.resize(0);
            for (const auto& s : temp_toks)
            {
                if (tok_str.size())
                    tok_str += " ";
                tok_str += s;
            }
        }

        resolve_res.m_num_input_tokens = (uint32_t)toks.size();

        // First token will be the most significant (country etc.), last token will be the least significant.
        std::reverse(toks.begin(), toks.end());

        std::vector<geo_result_vec> temp_results(toks.size());

        size_t total_results = 0;

        for (uint32_t toks_index = 0; toks_index < toks.size(); toks_index++)
        {
            const std::string& tok_str = toks[toks_index];

            std::vector<const pjson::value_variant*> results;
            std::vector<const pjson::value_variant*> alt_results;

            find(tok_str.c_str(), results, alt_results);

            if ((!toks_index) && (tok_str.size() == 2))
            {
                auto find_it = m_country_to_record_index.find(tok_str);
                if (find_it != m_country_to_record_index.end())
                {
                    const uint32_t country_rec_index = find_it->second;

                    alt_results.push_back(&m_doc[country_rec_index]);
                }
            }

            total_results += results.size() + alt_results.size();

            for (auto p : results)
            {
#if 0
                uprintf("-- full match: tok: %u, id: %u, name: \"%s\", altnames: \"%s\", lat: %f, long: %f, ccode=%s, a1=%s, a2=%s, a3=%s, a4=%s, fclass: %s, fcode: %s, pop: %i\n",
                    toks_index,
                    p->find_int32("id"),
                    p->find_string_ptr("name"), p->find_string_ptr("altnames"), p->find_float("lat"), p->find_float("long"),
                    p->find_string_ptr("ccode"),
                    p->find_string_ptr("a1"), p->find_string_ptr("a2"), p->find_string_ptr("a3"), p->find_string_ptr("a4"),
                    p->find_string_ptr("fclass"),
                    p->find_string_ptr("fcode"),
                    p->find_int32("pop"));
#endif
                                
                temp_results[toks_index].push_back({ p, false });
            }

            for (auto p : alt_results)
            {
#if 0
                uprintf("-- alt match: tok: %u, id: %u, name: \"%s\", altnames: \"%s\", lat: %f, long: %f, ccode=%s, a1=%s, a2=%s, a3=%s, a4=%s, fclass: %s, fcode: %s, pop: %i\n",
                    toks_index,
                    p->find_int32("id"),
                    p->find_string_ptr("name"), p->find_string_ptr("altnames"), p->find_float("lat"), p->find_float("long"),
                    p->find_string_ptr("ccode"),
                    p->find_string_ptr("a1"), p->find_string_ptr("a2"), p->find_string_ptr("a3"), p->find_string_ptr("a4"),
                    p->find_string_ptr("fclass"),
                    p->find_string_ptr("fcode"),
                    p->find_int32("pop"));
#endif

                temp_results[toks_index].push_back({ p, true });
            }
        }

        // 0=last token (Country)
        // size-1=first token (location/city/town)

        if (!total_results)
        {
            uprintf("No results\n");
            return false;
        }
                
        //uprintf("Candidates for query: %s\n", str.c_str());

        std::vector<uint32_t> valid_candidates;

        const uint32_t last_tok_index = (uint32_t)toks.size() - 1;
        //const bool single_token = (toks.size() == 1);

        for (uint32_t i = 0; i < temp_results[last_tok_index].size(); i++)
        {
            if (is_valid_candidate(temp_results, last_tok_index, i))
                valid_candidates.push_back(i);
        }

        std::vector< std::pair<uint32_t, float> > candidate_results[TOTAL_FAVORED_COUNTRY_RANKS];
        uint32_t total_country_rankings = 0;
        uint32_t total_candidates = 0;

        for (uint32_t candidate_index_iter = 0; candidate_index_iter < valid_candidates.size(); candidate_index_iter++)
        {
            const uint32_t candidate_index = valid_candidates[candidate_index_iter];

            const pjson::value_variant* p = temp_results[last_tok_index][candidate_index].m_pVariant;

            std::string name(p->find_string_obj("name"));
            std::string plainname(p->find_string_obj("plainname"));
            std::string ccode(p->find_string_obj("ccode"));

            float candidate_score = (float)get_rank(p, temp_results[last_tok_index][candidate_index].m_alt);

            if ((name == toks[last_tok_index]) || (plainname == toks[last_tok_index]))
                candidate_score += (float).5f;
            else if ((string_icompare(name, toks[last_tok_index].c_str()) == 0) || (string_icompare(plainname, toks[last_tok_index].c_str()) == 0))
                candidate_score += (float).2f;
            else
            {
                if (toks[last_tok_index].size() >= 5)
                {
                    if ((string_find_first(name, toks[last_tok_index].c_str()) != -1) ||
                        (string_find_first(plainname, toks[last_tok_index].c_str()) != -1))
                    {
                        candidate_score += (float).25f;
                    }
                }

                const std::string alt_names(p->find_string_obj("altnames"));

                string_vec alt_tokens;
                string_tokenize(alt_names, ",", "", alt_tokens);

                uint32_t j;
                for (j = 0; j < alt_tokens.size(); j++)
                {
                    if (alt_tokens[j] == toks[last_tok_index])
                    {
                        candidate_score += .1f;
                        break;
                    }
                }
            }
                        
            candidate_score += p->find_float("pop") / 40000000.0f;

            const int country_rank = get_favored_country_rank(ccode);
            assert(country_rank < TOTAL_FAVORED_COUNTRY_RANKS);

            if (!candidate_results[country_rank].size())
                total_country_rankings++;

            candidate_results[country_rank].push_back(std::make_pair(candidate_index, candidate_score));

            total_candidates++;
        }
                        
        // 1. If there's just one country rank group, choose the best score in that country rank group.
        // 2. If they matched against a country, choose the highest ranking country, prioritizing the favored countries first.
        // 3. Check for states, state capitals or other significant admin districts in the favored countries, in order
        // If they matched against a country's capital, choose the highest ranking capital.
        // 5. Check for significant enough cities in the favored countries, in order
        // 6. Choose the one with the highest score

        const geo_result* pBest_result = nullptr;
        float best_score = -1.0f;
        const std::vector< std::pair<uint32_t, float> >* pBest_ranking_vec = nullptr;
        uint32_t best_ranking_index = 0;

        // Sort each ranked country group by score, highest to lowest in each.
        for (auto& r : candidate_results)
        {
            if (r.size())
            {
                std::sort(r.begin(), r.end(),
                    [](const std::pair<uint32_t, float>& a, const std::pair<uint32_t, float>& b) -> bool
                    {
                        return a.second > b.second;
                    });
            }
        }

#if 0
        for (uint32_t rank_iter = 0; rank_iter < TOTAL_FAVORED_COUNTRY_RANKS; rank_iter++)
        {
            for (auto & c : candidate_results[rank_iter])
            {
                const uint32_t candidate_index = c.first;
                const float score = c.second;

                const pjson::value_variant* p = temp_results[last_tok_index][candidate_index].m_pVariant;

                uprintf("-- score: %f, country rank: %u, alt: %u, id: %u, name: \"%s\", altnames: \"%s\", lat: %f, long: %f, ccode=%s, a1=%s, a2=%s, a3=%s, a4=%s, fclass: %s, fcode: %s, pop: %i\n",
                    score,
                    rank_iter,
                    temp_results[last_tok_index][candidate_index].m_alt,
                    p->find_int32("id"),
                    p->find_string_ptr("name"), p->find_string_ptr("altnames"), p->find_float("lat"), p->find_float("long"),
                    p->find_string_ptr("ccode"),
                    p->find_string_ptr("a1"), p->find_string_ptr("a2"), p->find_string_ptr("a3"), p->find_string_ptr("a4"),
                    p->find_string_ptr("fclass"),
                    p->find_string_ptr("fcode"),
                    p->find_int32("pop"));
            }
        }
#endif
                
        if (total_country_rankings == 1)
        {
            // Only one ranked country group in the candidate results, so just choose the one with the highest score.

            for (const auto& r : candidate_results)
            {
                if (r.size())
                {
                    pBest_ranking_vec = &r;
                    break;
                }
            }
            
            assert(pBest_ranking_vec);
            
            uint32_t candidate_index = (*pBest_ranking_vec)[0].first;

            best_score = (*pBest_ranking_vec)[0].second;
            pBest_result = &temp_results[last_tok_index][candidate_index];
            best_ranking_index = 0;

            resolve_res.m_strong_match = (temp_results[last_tok_index].size() == 1);
        }
        else
        {
            // Multiple ranked country groups.
            
            // Check for US states (primary or alt)
            {
                uint32_t r_index = 0;
                const std::vector< std::pair<uint32_t, float> >& r = candidate_results[r_index];

                for (uint32_t i = 0; i < r.size(); i++)
                {
                    const uint32_t candidate_index = r[i].first;
                    const float score = r[i].second;

                    const pjson::value_variant* p = temp_results[last_tok_index][candidate_index].m_pVariant;

                    const int rank = get_rank(p, temp_results[last_tok_index][candidate_index].m_alt);

                    if ((rank == cRankAdmin1) || (rank == cRankAdmin1Alt))
                    {
                        pBest_result = &temp_results[last_tok_index][candidate_index];
                        best_score = score;
                        pBest_ranking_vec = &r;
                        best_ranking_index = i;
                        break;
                    }
                }
            }
            
            if (!pBest_result)
            {
                // First check for any country hits from any ranked country group.
                for (const auto& r : candidate_results)
                {
                    for (uint32_t i = 0; i < r.size(); i++)
                    {
                        const uint32_t candidate_index = r[i].first;
                        const float score = r[i].second;

                        const pjson::value_variant* p = temp_results[last_tok_index][candidate_index].m_pVariant;

                        const int rank = get_rank(p, temp_results[last_tok_index][candidate_index].m_alt);

                        if (rank >= cRankCountryLevel0)
                        {
                            pBest_result = &temp_results[last_tok_index][candidate_index];
                            best_score = score;
                            pBest_ranking_vec = &r;
                            best_ranking_index = i;
                            break;
                        }
                    }
                    
                    if (pBest_result)
                        break;
                }
            }

            if (!pBest_result)
            {
                // Check for country capitals or states in the favored countries
                for (uint32_t r_index = 0; r_index < (TOTAL_FAVORED_COUNTRY_RANKS - 1); r_index++)
                {
                    const std::vector< std::pair<uint32_t, float> >& r = candidate_results[r_index];

                    for (uint32_t i = 0; i < r.size(); i++)
                    {
                        const uint32_t candidate_index = r[i].first;
                        const float score = r[i].second;

                        const pjson::value_variant* p = temp_results[last_tok_index][candidate_index].m_pVariant;
                        //const bool was_alt = temp_results[last_tok_index][candidate_index].m_alt;

                        const int rank = get_rank(p, temp_results[last_tok_index][candidate_index].m_alt);
                                                
                        if ((rank == cRankAdmin1Alt) || (rank == cRankAdmin1) || (rank == cRankPoliticalCapital) || (rank == cRankGovernmentCapital))
                        {
                            pBest_result = &temp_results[last_tok_index][candidate_index];
                            best_score = score;
                            pBest_ranking_vec = &r;
                            best_ranking_index = i;
                            break;
                        }
                    }
                    
                    if (pBest_result)
                        break;
                }
            }

            if (!pBest_result)
            {
                // Check for large cities with pops > 100K, from any country, this will catch "Moscow"
                int best_pop = 99999;

                for (uint32_t r_index = 0; r_index < TOTAL_FAVORED_COUNTRY_RANKS; r_index++)
                {
                    const std::vector< std::pair<uint32_t, float> >& r = candidate_results[r_index];

                    for (uint32_t i = 0; i < r.size(); i++)
                    {
                        const uint32_t candidate_index = r[i].first;
                        const float score = r[i].second;

                        const pjson::value_variant* p = temp_results[last_tok_index][candidate_index].m_pVariant;

                        int pop = p->find_int32("pop");

                        if ((!temp_results[last_tok_index][candidate_index].m_alt) && (pop > best_pop))
                        {
                            best_pop = pop;

                            pBest_result = &temp_results[last_tok_index][candidate_index];
                            best_score = score;
                            pBest_ranking_vec = &r;
                            best_ranking_index = i;
                        }
                    }
                }
            }

            if (!pBest_result)
            {
                // Check for large enough cities in our favored countries
                for (uint32_t r_index = 0; r_index < (TOTAL_FAVORED_COUNTRY_RANKS - 1); r_index++)
                {
                    const std::vector< std::pair<uint32_t, float> >& r = candidate_results[r_index];

                    for (uint32_t i = 0; i < r.size(); i++)
                    {
                        const uint32_t candidate_index = r[i].first;
                        const float score = r[i].second;

                        const pjson::value_variant* p = temp_results[last_tok_index][candidate_index].m_pVariant;

                        const int rank = get_rank(p, temp_results[last_tok_index][candidate_index].m_alt);

                        if (rank >= cRankCityLevel2)
                        {
                            if (score > best_score)
                            {
                                pBest_result = &temp_results[last_tok_index][candidate_index];
                                best_score = score;
                                pBest_ranking_vec = &r;
                                best_ranking_index = i;
                            }
                        }
                    }
                }
            }
                        
            if (!pBest_result)
            {
                // Fall back to choosing the highest score
                for (uint32_t r_index = 0; r_index < TOTAL_FAVORED_COUNTRY_RANKS; r_index++)
                {
                    const std::vector< std::pair<uint32_t, float> >& r = candidate_results[r_index];

                    for (uint32_t i = 0; i < r.size(); i++)
                    {
                        const uint32_t candidate_index = r[i].first;
                        const float score = r[i].second;
                                                
                        if (score > best_score)
                        {
                            best_score = score;

                            pBest_result = &temp_results[last_tok_index][candidate_index];
                            
                            pBest_ranking_vec = &r;
                            best_ranking_index = i;
                        }
                    }
                }
            }
        }

#if 0
        uprintf("\n");
        uprintf("---- Query: %s\n", str.c_str());
#endif

        if (!pBest_result)
        {
            uprintf("No results\n");
            return false;
        }

        resolve_res.m_best_result = *pBest_result;
        resolve_res.m_candidates = temp_results[last_tok_index];
        resolve_res.m_sorted_results = *pBest_ranking_vec;
        resolve_res.m_best_sorted_result_index = best_ranking_index;
        resolve_res.m_best_score = best_score;

        const pjson::value_variant* pVariant = pBest_result->m_pVariant;

#if 0                        
        uprintf("Result: score:%f, alt: %u, id: %u, name: \"%s\", lat: %f, long: %f, ccode=%s, a1=%s, a2=%s, a3=%s, a4=%s, fclass: %s, fcode: %s, pop: %i\n",
            best_score,
            pBest_result->m_alt,
            pVariant->find_int32("id"),
            pVariant->find_string_ptr("name"), pVariant->find_float("lat"), pVariant->find_float("long"),
            pVariant->find_string_ptr("ccode"),
            pVariant->find_string_ptr("a1"), pVariant->find_string_ptr("a2"), pVariant->find_string_ptr("a3"), pVariant->find_string_ptr("a4"),
            pVariant->find_string_ptr("fclass"),
            pVariant->find_string_ptr("fcode"),
            pVariant->find_int32("pop"));

        std::string rdesc(get_desc(pBest_result->m_pVariant));
        uprintf("Desc: %s\n", rdesc.c_str());
#endif

        return true;
    }

    std::string get_desc(const pjson::value_variant* p) const
    {
        std::string res;

        std::string ccode(p->find_string_obj("ccode"));
        std::string fclass(p->find_string_obj("fclass"));
        std::string fcode(p->find_string_obj("fcode"));
        
        std::string a[4] = { p->find_string_obj("a1"), p->find_string_obj("a2"), p->find_string_obj("a3"), p->find_string_obj("a4") };

        uint32_t num_admins = count_admins(p);

        res += p->find_string_obj("name");

        const int start_admin = num_admins - 1;

        for (int i = start_admin; i >= 0; i--)
        {
            std::string x(ccode);
            for (int j = 0; j <= i; j++)
                x += "." + a[j];

            auto find_res(m_admin_map.find(x));
            if (find_res != m_admin_map.end())
            {
                const std::vector< std::pair<int, int> >& recs = find_res->second;
                
                assert(recs.size());
                                
                int cur_level = recs[0].second;
                for (uint32_t j = 0; j < recs.size(); j++)
                {
                    if (recs[j].second != cur_level)
                        break;

                    int rec_index = recs[j].first;
                                        
                    const pjson::value_variant* q = &m_doc[rec_index];

                    if (i == (int)(num_admins - 1))
                    {
                        if (!is_valid_parent(q, p))
                        {
                            goto skip;
                        }
                    }

                    if (!string_ends_in(res, q->find_string_obj("name").c_str()))
                    {
                        if (j)
                            res += "/";
                        else
                            res += ", ";

                        res += q->find_string_obj("name");
                    }
                }
            }

        skip:;
        }

        auto find_res(m_country_to_record_index.find(ccode));
        if (find_res != m_country_to_record_index.end())
        {
            int country_rec_index = find_res->second;

            if (res != m_doc[country_rec_index].find_string_obj("name"))
                res += ", " + m_doc[country_rec_index].find_string_obj("name");
        }

        return res;
    }

private:
    pjson::document m_doc;
    uint8_vec m_filebuf;
    std::vector<uint_vec> m_name_hashtab;

    std::unordered_map<int, int> m_geoid_to_rec;
            
    country_info_vec m_countries;
    std::unordered_map<int, int> m_rec_index_to_country_index;
    std::unordered_map<int, int> m_geoid_to_country_index;

    // maps iso2 name to m_countries[] record index
    std::unordered_map<std::string, uint32_t> m_country_to_record_index;

    // maps parent to child or children geoid's, used to handle places that are in multiple counties (for example)
    std::unordered_map<int, int_vec> m_geo_hierarchy;

    // maps CC+ADM1+ADM2+etc. to (rec index, admin_level)
    std::unordered_map<std::string, std::vector< std::pair<int, int> > > m_admin_map;

    int get_country_index(const pjson::value_variant* p) const
    {
        int geoid = p->find_int32("id");
        auto find_res = m_geoid_to_country_index.find(geoid);

        if (find_res == m_geoid_to_country_index.end())
            return -1;

        return find_res->second;
    }

    static void extract_tab_fields(const std::string& str, string_vec& fields) 
    {
        std::vector<int> tab_locs;
        tab_locs.resize(0);

        for (uint32_t ofs = 0; ofs < str.size(); ofs++)
            if (str[ofs] == '\t')
                tab_locs.push_back(ofs);

        tab_locs.push_back((uint32_t)str.size());

        fields.resize(tab_locs.size());

        uint32_t cur_ofs = 0;
        for (uint32_t i = 0; i < fields.size(); i++)
        {
            fields[i] = string_slice(str, cur_ofs, tab_locs[i] - cur_ofs);
            string_trim(fields[i]);
            cur_ofs = tab_locs[i] + 1;
        }
    }

    void load_hierarchy()
    {
        string_vec lines;
        if (!read_text_file("geoHierarchy.txt", lines, false, nullptr))
            panic("Failed reading geoHierarchy.txt");

        m_geo_hierarchy.clear();
        m_geo_hierarchy.reserve(500000);

        for (std::string &str : lines)
        {
            string_vec fields;
            extract_tab_fields(str, fields);

            if (fields.size() < 2)
                panic("Invalid geoHierarchy.txt");

            int parent_index = atoi(fields[0].c_str());
            int child_index = atoi(fields[1].c_str());

            m_geo_hierarchy[parent_index].push_back(child_index);
        }
    }

    void load_country_info()
    {
        string_vec country_lines;
        country_lines.reserve(256);

        if (!read_text_file("countryInfo.txt", country_lines, false, nullptr))
            panic("failed reading countryInfo.txt");

        std::vector< std::pair<int, int> > sorted_by_pop;

        for (uint32_t country_rec_index = 0; country_rec_index < country_lines.size(); country_rec_index++)
        {
            const std::string& str = country_lines[country_rec_index];
            if (!str.size())
                continue;
            if (str[0] == '#')
                continue;

            std::vector<int> tab_locs;
            tab_locs.resize(0);

            for (uint32_t ofs = 0; ofs < str.size(); ofs++)
                if (str[ofs] == '\t')
                    tab_locs.push_back(ofs);

            const uint32_t N = 18;
            if (tab_locs.size() != N)
                panic("Unexpected number of tabs");

            tab_locs.push_back((uint32_t)str.size());

            string_vec fields(N + 1);

            enum
            {
                cCountryISO, cCountryISO3, cCountryISONumeric, cCountryfips, cCountryName, cCountryCapital,
                cCountryAreaSqKm, cCountryPopulation, cCountryContinent, cCountryTld, cCountryCurrencyCode, cCountryCurrencyName,
                cCountryPhone, cCountryPostalCodeFormat, cCountryPostalCodeRegex, cCountryLanguages, cCountrygeonameid, cCountryNeighbours, cCountryEquivalentFipsCode
            };

            uint32_t cur_ofs = 0;
            for (uint32_t i = 0; i < fields.size(); i++)
            {
                fields[i] = string_slice(str, cur_ofs, tab_locs[i] - cur_ofs);
                string_trim(fields[i]);
                cur_ofs = tab_locs[i] + 1;
            }

            const int country_index = (int)m_countries.size();

            m_countries.resize(m_countries.size() + 1);
            country_info& new_c = m_countries.back();

            new_c.m_iso2_name = fields[cCountryISO];
            new_c.m_iso3_name = fields[cCountryISO3];
            new_c.m_iso_numeric = fields[cCountryISONumeric];
            new_c.m_fips = fields[cCountryfips];
            new_c.m_country_name = fields[cCountryName];
            new_c.m_capital_name = fields[cCountryCapital];
            new_c.m_area = atoi(fields[cCountryAreaSqKm].c_str());
            new_c.m_pop = atoi(fields[cCountryPopulation].c_str());
            new_c.m_continent = fields[cCountryContinent];
            new_c.m_geoid = atoi(fields[cCountrygeonameid].c_str());
            new_c.m_pop_rank = 0;

            if (!m_geoid_to_rec.count(new_c.m_geoid))
                panic("Failed finding geoid");

            new_c.m_rec_index = m_geoid_to_rec[new_c.m_geoid];
            new_c.m_neighbors = fields[cCountryNeighbours];

            if (m_rec_index_to_country_index.count(new_c.m_rec_index) != 0)
                panic("Invalid rec index");

            m_rec_index_to_country_index[new_c.m_rec_index] = country_index;
            m_geoid_to_country_index[new_c.m_geoid] = country_index;

            auto ins_res = m_country_to_record_index.insert(std::make_pair(new_c.m_iso2_name, new_c.m_rec_index));
            if (!ins_res.second)
                panic("Failed inserting country");

            sorted_by_pop.push_back(std::make_pair(new_c.m_pop, country_index));

#if 0
            {
                const pjson::value_variant* p = &m_doc[new_c.m_rec_index];

                uprintf("a1: %s\n", p->find_string_ptr("a1"));
                uprintf("fclass: %s\n", p->find_string_ptr("fclass"));
                uprintf("fcode: %s\n", p->find_string_ptr("fcode"));
#if 0
                uprintf("-- id: %u, name: \"%s\", lat: %f, long: %f, ccode=%s, a1=%s, a2=%s, a3=%s, a4=%s, fclass: %s, fcode: %s, pop: %i\n",
                    p->find_int32("id"),
                    p->find_string_ptr("name"), p->find_float("lat"), p->find_float("long"),
                    p->find_string_ptr("ccode"),
                    p->find_string_ptr("a1"), p->find_string_ptr("a2"), p->find_string_ptr("a3"), p->find_string_ptr("a4"),
                    p->find_string_ptr("fclass"),
                    p->find_string_ptr("fcode"),
                    p->find_int32("pop"));
#endif
            }

            // countries are all fclass A, a1=00
            // fcode must be PCL, PCLD, PCLF, PCLH, PCLI, PCLIX, PCLS, TERR
#endif

        }

        std::sort(sorted_by_pop.begin(), sorted_by_pop.end(),
            [](const std::pair<int, int>& a, const std::pair<int, int>& b) -> bool
            {
                return a.first < b.first;
            });

        for (uint32_t i = 0; i < sorted_by_pop.size(); i++)
        {
            uint32_t country_index = sorted_by_pop[i].second;
            m_countries[country_index].m_pop_rank = (i * 3 + ((int)sorted_by_pop.size() / 2)) / ((int)sorted_by_pop.size() - 1);
        }
    }
};

static bool has_match(const std::string& str, const std::string& pat, bool ignore_case)
{
    if (ignore_case)
        return (string_ifind_first(str, pat.c_str()) != -1);
    else
        return (string_find_first(str, pat.c_str()) != -1);
}

static bool has_match(const string_vec& str_vec, const std::string& pat, bool ignore_case)
{
    for (const auto& str : str_vec)
        if (has_match(str, pat, ignore_case))
            return true;
    return false;
}

// Main code
int wmain(int argc, wchar_t* argv[])
{
    assert(cTotalPrefixes == sizeof(g_date_prefix_strings) / sizeof(g_date_prefix_strings[0]));
        
    string_vec args;
    convert_args_to_utf8(args, argc, argv);

    // Set ANSI Latin 1; Western European (Windows) code page for output.
    SetConsoleOutputCP(1252);
    //SetConsoleOutputCP(CP_UTF8);
        
    converters_init();

    udb_init();

    bool status = false, utf8_flag = false;

    unordered_string_set unique_urls;

    bool single_file_output = false;

    string_vec filter_strings;
    bool filter_ignore_case = false;
    bool filter_all_flag = false;
    std::string title_str("All events");
    
    int arg_index = 1;
    while (arg_index < argc)
    {
        std::string arg(args[arg_index]);
        uint8_t t = arg[0];
        arg_index++;

        const uint32_t num_args_remaining = argc - arg_index;
        
        if (t == '-')
        {
            if (arg == "-filter")
            {
                if (num_args_remaining < 1)
                    panic(string_format("Expected parameter after option: %s", arg.c_str()).c_str());
                else
                {
                    if (!args[arg_index].size())
                        panic("Invalid filter string");

                    filter_strings.push_back(args[arg_index++]);
                }
            }
            else if (arg == "-title")
            {
                if (num_args_remaining < 1)
                    panic(string_format("Expected parameter after option: %s", arg.c_str()).c_str());
                else
                {
                    title_str = args[arg_index++];
                }
            }
            else if (arg == "-filter_all")
            {
                filter_all_flag = true;
            }
            else if (arg == "-filter_ignore_case")
            {
                filter_ignore_case = false;
            }
            else if (arg == "-single_file")
            {
                single_file_output = true;
            }
            else
            {
                panic(string_format("Unrecognized option: %s", arg.c_str()).c_str());
            }
        }
        else
        {
            panic(string_format("Unrecognized option: %s", arg.c_str()).c_str());
        }
    }
                            
#if 1
    uprintf("Convert Eberhart:\n");
    if (!convert_eberhart(unique_urls))
        panic("convert_eberthart() failed!");
    uprintf("Success\n");

    uprintf("Convert Nuclear:\n");
    if (!convert_nuk())
        panic("convert_nuk() failed!");
    uprintf("Success\n");

    uprintf("Convert Hatch UDB:\n");
    if (!udb_convert())
        panic("udb_convert() failed!");
    uprintf("Success\n");

    uprintf("Convert NICAP:\n");
    if (!convert_nicap(unique_urls))
        panic("convert_nicap() failed!");
    uprintf("Success\n");

    uprintf("Convert Johnson:\n");
    if (!convert_johnson())
        panic("convert_johnson() failed!");
    uprintf("Success\n");
        
    uprintf("Convert Trace:\n");
    if (!convert_magnonia("trace.txt", "trace.json", "Trace", " [Trace Cases](https://www.thenightskyii.org/trace.html)"))
        panic("convert_magnonia() failed!");
    uprintf("Success\n");

    uprintf("Convert Magnonia:\n");
    if (!convert_magnonia("magnonia.txt", "magnonia.json"))
        panic("convert_magnonia() failed!");
    uprintf("Success\n");
    
    uprintf("Convert Hall:\n");
    if (!convert_hall())
        panic("convert_hall() failed!");
    uprintf("Success\n");

    uprintf("Convert Bluebook Unknowns:\n");
    if (!convert_bluebook_unknowns())
        panic("convert_bluebook_unknowns failed!");
    uprintf("Success\n");
#endif
    
    uprintf("Total unique URL's: %u\n", (uint32_t)unique_urls.size());

    string_vec urls;
    for (const auto& s : unique_urls)
        urls.push_back(s);
    std::sort(urls.begin(), urls.end());
    write_text_file("unique_urls.txt", urls, false);
    uprintf("Wrote unique_urls.txt\n");

    ufo_timeline timeline;

    status = timeline.load_json("maj2.json", utf8_flag, "Maj2", true);
    if (!status)
        panic("Failed loading maj2.json");
    for (uint32_t i = 0; i < timeline.size(); i++)
        timeline[i].m_source_id = string_format("%s_%u", timeline[i].m_source.c_str(), i);

#if 1
    status = timeline.load_json("hatch_udb.json", utf8_flag, nullptr, false);
    if (!status)
        panic("Failed loading hatch_udb.json");

    status = timeline.load_json("nicap_db.json", utf8_flag, nullptr, false);
    if (!status)
        panic("Failed loading nicap_db.json");
        
    status = timeline.load_json("trace.json", utf8_flag, nullptr, false);
    if (!status)
        panic("Failed loading trace.json");

    status = timeline.load_json("magnonia.json", utf8_flag, nullptr, false);
    if (!status)
        panic("Failed loading magnolia.json");

    status = timeline.load_json("bb_unknowns.json", utf8_flag, nullptr, false);
    if (!status)
        panic("Failed loading bb_unknowns.json");

    status = timeline.load_json("ufo_evidence_hall.json", utf8_flag, nullptr, false);
    if (!status)
        panic("Failed loading ufo_evidence_hall.json");
    
    status = timeline.load_json("nuclear_tests.json", utf8_flag, nullptr, false);
    if (!status)
        panic("Failed loading nuclear_tests.json");

    status = timeline.load_json("eberhart.json", utf8_flag, nullptr, false);
    if (!status)
        panic("Failed loading eberhart.json");

    status = timeline.load_json("johnson.json", utf8_flag, nullptr, false);
    if (!status)
        panic("Failed loading johnson.json");
#endif

    for (uint32_t i = 0; i < timeline.size(); i++)
    {
        if (!timeline[i].m_begin_date.sanity_check())
            panic("Date failed sanity check");

        if (timeline[i].m_end_date.is_valid())
        {
            if (!timeline[i].m_end_date.sanity_check())
                panic("Date failed sanity check");
        }

        if (timeline[i].m_alt_date.is_valid())
        {
            if (!timeline[i].m_alt_date.sanity_check())
                panic("Date failed sanity check");
        }

        // roundtrip test
        event_date test_date;

        if (!test_date.parse(timeline[i].m_date_str.c_str(), false))
            panic("Date failed sanity check");

        if (test_date != timeline[i].m_begin_date)
            panic("Date failed sanity check");

        std::string test_str(timeline[i].m_begin_date.get_string());
        if (test_str != timeline[i].m_date_str)
        {
            fprintf(stderr, "Date failed roundtrip: %s %s %s\n", timeline[i].m_source_id.c_str(), timeline[i].m_date_str.c_str(), test_str.c_str());
            panic("Date failed sanity check");
        }

        if (timeline[i].m_end_date.is_valid())
        {
            if (!test_date.parse(timeline[i].m_end_date_str.c_str(), false))
                panic("Date failed sanity check");

            if (test_date != timeline[i].m_end_date)
                panic("Date failed sanity check");

            std::string test_str2(timeline[i].m_end_date.get_string());
            if (test_str2 != timeline[i].m_end_date_str)
            {
                fprintf(stderr, "Date failed roundtrip: %s %s %s\n", timeline[i].m_source_id.c_str(), timeline[i].m_end_date_str.c_str(), test_str2.c_str());
                panic("End date failed sanity check");
            }
        }
        else if (timeline[i].m_end_date_str.size())
            panic("Date failed sanity check");

        if (timeline[i].m_alt_date.is_valid())
        {
            if (!test_date.parse(timeline[i].m_alt_date_str.c_str(), false))
                panic("Date failed sanity check");

            if (test_date != timeline[i].m_alt_date)
                panic("Date failed sanity check");

            std::string test_str3(timeline[i].m_alt_date.get_string());
            if (test_str3 != timeline[i].m_alt_date_str)
            {
                fprintf(stderr, "Date failed roundtrip: %s %s %s\n", timeline[i].m_source_id.c_str(), timeline[i].m_alt_date_str.c_str(), test_str3.c_str());
                panic("Alt date failed sanity check");
            }
        }
        else if (timeline[i].m_alt_date_str.size())
            panic("Date failed sanity check");

    }
        
    uprintf("Load success, %zu total events\n", timeline.get_events().size());

    timeline.sort();

    if (filter_strings.size())
    {
        ufo_timeline new_timeline;
                
        for (uint32_t i = 0; i < timeline.size(); i++)
        {
            const timeline_event& event = timeline[i];

            uint32_t total_matched = 0;
            for (uint32_t j = 0; j < filter_strings.size(); j++)
            {
                bool match_flag = false;

                const std::string& filter_string = filter_strings[j];

                if (has_match(event.m_date_str, filter_string, filter_ignore_case) ||
                    has_match(event.m_time_str, filter_string, filter_ignore_case) ||
                    has_match(event.m_alt_date_str, filter_string, filter_ignore_case) ||
                    has_match(event.m_end_date_str, filter_string, filter_ignore_case) ||
                    has_match(event.m_desc, filter_string, filter_ignore_case) ||
                    has_match(event.m_type, filter_string, filter_ignore_case) ||
                    has_match(event.m_refs, filter_string, filter_ignore_case) ||
                    has_match(event.m_locations, filter_string, filter_ignore_case) ||
                    has_match(event.m_attributes, filter_string, filter_ignore_case) ||
                    has_match(event.m_see_also, filter_string, filter_ignore_case) ||
                    has_match(event.m_rocket_type, filter_string, filter_ignore_case) ||
                    has_match(event.m_rocket_altitude, filter_string, filter_ignore_case) ||
                    has_match(event.m_rocket_range, filter_string, filter_ignore_case) ||
                    has_match(event.m_atomic_type, filter_string, filter_ignore_case) ||
                    has_match(event.m_atomic_kt, filter_string, filter_ignore_case) ||
                    has_match(event.m_atomic_mt, filter_string, filter_ignore_case) ||
                    has_match(event.m_source_id, filter_string, filter_ignore_case) ||
                    has_match(event.m_source, filter_string, filter_ignore_case))
                {
                    match_flag = true;
                }

                if (match_flag)
                    total_matched++;
            }

            if ( ((filter_all_flag) && (total_matched == filter_strings.size())) ||
                 ((!filter_all_flag) && (total_matched > 0)) ) 
            {
                new_timeline.get_events().push_back(event);
            }
        }

        if (!new_timeline.size())
            panic("No events founds.\n");
        else
            uprintf("Found %zu event(s)\n", new_timeline.size());

        timeline.get_events().swap(new_timeline.get_events());
    }

    // Write majestic.json, then load it and verify that it saved and loaded correctly.
    {
        timeline.set_name("Majestic Timeline");
        timeline.write_file("majestic.json", true);

        ufo_timeline timeline_comp;
        bool utf8_flag_comp;
        if (!timeline_comp.load_json("majestic.json", utf8_flag_comp, nullptr, false))
            panic("Failed loading majestic.json");

        if (timeline.get_events().size() != timeline_comp.get_events().size())
            panic("Failed loading timeline events JSON");

        for (size_t i = 0; i < timeline.get_events().size(); i++)
            if (timeline[i] != timeline_comp[i])
                panic("Failed comparing majestic.json");
    }

    if (single_file_output)
    {
        timeline.write_markdown("timeline.md", title_str.c_str(), -10000, 10000, true);
    }
    else
    {
        timeline.write_markdown("timeline.md", "Part 1: Distant Past up to and including 1949", -10000, 1949, false);
        timeline.write_markdown("timeline_part2.md", "Part 2: 1950 up to and including 1959", 1950, 1959, false);
        timeline.write_markdown("timeline_part3.md", "Part 3: 1960 up to and including 1969", 1960, 1969, false);
        timeline.write_markdown("timeline_part4.md", "Part 4: 1970 up to and including 1979", 1970, 1979, false);
        timeline.write_markdown("timeline_part5.md", "Part 5: 1980 to Present", 1980, 10000, false);
    }

    uprintf("Processing successful\n");

    return EXIT_SUCCESS;
}