#ifndef ADBC_ARROW_SCHEMA_HPP
#define ADBC_ARROW_SCHEMA_HPP
#pragma once

#include <stdio.h>
#include <cmath>
#include <cstdbool>
#include <cstdint>
#include <vector>
#include <arrow-adbc/adbc.h>
#include <erl_nif.h>
#include "adbc_consts.h"
#include "adbc_arrow_metadata.hpp"
#include "adbc_column.hpp"
#include "nif_utils.hpp"

static int arrow_schema_to_nif_term(ErlNifEnv *env, struct ArrowSchema * schema, struct ArrowArray * array, uint64_t level, std::vector<ERL_NIF_TERM> &out_terms, ERL_NIF_TERM &value_type, ERL_NIF_TERM &metadata, ERL_NIF_TERM &error);

static int get_struct_schema(ErlNifEnv *env, struct ArrowSchema * schema, struct ArrowArray * array, uint64_t level, std::vector<ERL_NIF_TERM> &children, ERL_NIF_TERM &error) {
    if (schema->n_children > 0 && schema->children == nullptr) {
        error = erlang::nif::error(env, "invalid ArrowSchema, schema->children == nullptr while schema->n_children > 0");
        return 1;
    }
    children.resize(schema->n_children);
    for (int64_t child_i = 0; child_i < schema->n_children; child_i++) {
        struct ArrowSchema * child_schema = schema->children[child_i];
        std::vector<ERL_NIF_TERM> childrens;
        ERL_NIF_TERM child_type;
        ERL_NIF_TERM child_metadata;
        if (arrow_schema_to_nif_term(env, child_schema, nullptr, level + 1, childrens, child_type, child_metadata, error) != 0) {
            return 1;
        }

        if (level == 0) {
            using record_type = NifRes<struct ArrowArrayStreamRecord>;
            auto * record = record_type::allocate_resource(env, error);
            if (record == nullptr) {
                return 1;
            }
            if (record->val.allocate_schema_and_values()) {
                error = erlang::nif::error(env, "out of memory");
                return 1;
            }
            struct ArrowArray * child_array = array->children[child_i];
            ArrowSchemaDeepCopy(child_schema, record->val.schema);
            ArrowArrayMove(child_array, record->val.values);
            memset(array->children[child_i], 0, sizeof(struct ArrowArray));
            ERL_NIF_TERM data_ref = record->make_resource(env);

            children[child_i] = make_adbc_column(env, child_schema, child_type, child_metadata, data_ref);
        } else {
            children[child_i] = make_adbc_column(env, child_schema, child_type, child_metadata);
        }
    }

    return 0;
}

static int get_run_end_encoded_schema(ErlNifEnv *env, struct ArrowSchema * schema, uint64_t level, ERL_NIF_TERM &run_ends_schema, ERL_NIF_TERM &error) {
    if (schema->n_children != 2) {
        return erlang::nif::error(env, "invalid ArrowSchema (run_end_encoded), schema->n_children != 2");
    }
    if (schema->children == nullptr) {
        return erlang::nif::error(env, "invalid ArrowArray (run_end_encoded), schema->children == nullptr");
    }
    if (strcmp("run_ends", schema->children[0]->name) != 0) {
        return erlang::nif::error(env, "invalid ArrowSchema (run_end_encoded), its first child is not named run_ends");
    }
    if (strcmp("values", schema->children[1]->name) != 0) {
        return erlang::nif::error(env, "invalid ArrowSchema (run_end_encoded), its second child is not named values");
    }

    std::vector<ERL_NIF_TERM> children(2);
    for (int64_t child_i = 0; child_i < 2; child_i++) {
        std::vector<ERL_NIF_TERM> childrens;
        ERL_NIF_TERM child_type;
        ERL_NIF_TERM child_metadata;
        if (arrow_schema_to_nif_term(env, schema->children[child_i], nullptr, level + 1, childrens, child_type, child_metadata, error) != 0) {
            return 1;
        }

        children[child_i] = make_adbc_column(env, schema->children[child_i], child_type, child_metadata);
    }

    ERL_NIF_TERM run_ends_keys[] = { kAtomRunEnds, kAtomValues };
    ERL_NIF_TERM run_ends_values[] = { children[0], children[1] };
    // only fail if there are duplicated keys
    // so we don't need to check the return value
    enif_make_map_from_arrays(env, run_ends_keys, run_ends_values, 2, &run_ends_schema);
    return 0;
}

static int get_map_schema(ErlNifEnv *env, struct ArrowSchema * schema, uint64_t level, ERL_NIF_TERM &map_kv_schema, ERL_NIF_TERM &error) {
    // From https://arrow.apache.org/docs/format/CDataInterface.html#data-type-description-format-strings
    //
    //   As specified in the Arrow columnar format, the map type has a single child type named entries,
    //   itself a 2-child struct type of (key, value).
    if (schema->children == nullptr) {
        return erlang::nif::error(env, "invalid ArrowSchema (map), schema->children == nullptr");
    }
    if (schema->n_children != 1) {
        return erlang::nif::error(env, "invalid ArrowSchema (map), schema->n_children != 1");
    }

    struct ArrowSchema * entries_schema = schema->children[0];
    if (strcmp("entries", entries_schema->name) != 0) {
        return erlang::nif::error(env, "invalid ArrowSchema (map), its single child is not named entries");
    }
    if (entries_schema->n_children != 2) {
        return erlang::nif::error(env, "invalid ArrowSchema (map), its entries n_children != 2");
    }

    ERL_NIF_TERM key_schema, value_schema;
    int kv = 0;
    for (int64_t child_i = 0; child_i < 2; child_i++) {
        struct ArrowSchema * entry_schema = entries_schema->children[child_i];
        std::vector<ERL_NIF_TERM> childrens;
        ERL_NIF_TERM child_type;
        ERL_NIF_TERM child_metadata;
        if (strcmp("key", entry_schema->name) == 0) {
            if (arrow_schema_to_nif_term(env, entry_schema, nullptr, level + 1, childrens, child_type, child_metadata, error) != 0) {
                return 1;
            }

            key_schema = make_adbc_column(env, entry_schema, child_type, child_metadata);
            kv |= 0x1;
        } else if (strcmp("value", entry_schema->name) == 0) {
            if (arrow_schema_to_nif_term(env, entry_schema, nullptr, level + 1, childrens, child_type, child_metadata, error) != 0) {
                return 1;
            }

            value_schema = make_adbc_column(env, entry_schema, child_type, child_metadata);
            kv |= 0x2;
        } else {
            return 1;
        }
    }

    if (kv != 3) {
        return 1;
    }

    ERL_NIF_TERM map_kv_keys[] = { kAtomKey, kAtomValue };
    ERL_NIF_TERM map_kv_values[] = { key_schema, value_schema };
    // only fail if there are duplicated keys
    // so we don't need to check the return value
    enif_make_map_from_arrays(env, map_kv_keys, map_kv_values, 2, &map_kv_schema);
    return 0;
}

static int get_list_element_schema(ErlNifEnv *env, struct ArrowSchema * schema, uint64_t level, ERL_NIF_TERM &element_schema, ERL_NIF_TERM &error) {
    if (schema->children == nullptr) {
        return erlang::nif::error(env, "invalid ArrowSchema (list), schema->children == nullptr");
    }
    if (schema->n_children != 1) {
        return erlang::nif::error(env, "invalid ArrowSchema (list), schema->n_children != 1");
    }

    struct ArrowSchema * items_schema = schema->children[0];
    if (strcmp("item", items_schema->name) != 0) {
        return erlang::nif::error(env, "invalid ArrowSchema (list), its single child is not named item");
    }

    std::vector<ERL_NIF_TERM> childrens;
    ERL_NIF_TERM child_type;
    ERL_NIF_TERM child_metadata;
    if (arrow_schema_to_nif_term(env, items_schema, nullptr, level + 1, childrens, child_type, child_metadata, error) != 0) {
        return 1;
    }
    element_schema = make_adbc_column(env, items_schema, child_type, child_metadata);
    return 0;
}

static int arrow_schema_to_nif_term(ErlNifEnv *env, struct ArrowSchema * schema, struct ArrowArray * array, std::vector<ERL_NIF_TERM> &out_terms, ERL_NIF_TERM &error) {
    ERL_NIF_TERM type_term, metadata;
    int level = 0;
    return arrow_schema_to_nif_term(env, schema, array, level, out_terms, type_term, metadata, error);
}

static int arrow_schema_to_nif_term(ErlNifEnv *env, struct ArrowSchema * schema, struct ArrowArray * array, uint64_t level, std::vector<ERL_NIF_TERM> &out_terms, ERL_NIF_TERM &type_term, ERL_NIF_TERM &metadata, ERL_NIF_TERM &error) {
    if (schema == nullptr) {
        error = erlang::nif::error(env, "invalid ArrowSchema (nullptr) when invoking next");
        return 1;
    }
    if (level == 0 && array == nullptr) {
        error = erlang::nif::error(env, "invalid ArrowArray (nullptr) is nullptr at top-level entry");
        return 1;
    }

    char err_msg_buf[256] = { '\0' };
    const char* format = schema->format ? schema->format : "";
    ERL_NIF_TERM children_term{};
    size_t format_len = strlen(format);

    type_term = kAtomNil;
    std::vector<ERL_NIF_TERM> children;
    NANOARROW_RETURN_NOT_OK(arrow_metadata_to_nif_term(env, schema->metadata, &metadata));

    if (schema->dictionary != nullptr) {
        // NANOARROW_TYPE_DICTIONARY
        //
        // For dictionary-encoded arrays, the ArrowSchema.format string 
        // encodes the index type. The dictionary value type can be read 
        // from the ArrowSchema.dictionary structure.
        //
        // The same holds for ArrowArray structure: while the parent 
        // structure points to the index data, the ArrowArray.dictionary 
        // points to the dictionary values array.
        type_term = kAdbcColumnTypeDictionary;

        std::vector<ERL_NIF_TERM> childrens;
        ERL_NIF_TERM child_type;
        ERL_NIF_TERM child_metadata;
        if (arrow_schema_to_nif_term(env, schema->dictionary, nullptr, level + 1, childrens, child_type, child_metadata, error) != 0) {
            return 1;
        }

        type_term = enif_make_tuple2(env, kAdbcColumnTypeDictionary, child_type);
        children_term = make_adbc_column(env, schema->dictionary, type_term, child_metadata);
        return 0;
    }

    auto iter = primitiveFormatMapping.find(format);
    if (iter != primitiveFormatMapping.end()) {
        if (iter->second.size() == 1) {
            type_term = iter->second[0];
        } else {
            type_term = enif_make_tuple_from_array(env, iter->second.data(), (unsigned)iter->second.size());
        }
        children_term = make_adbc_column(env, schema, type_term, metadata);
    }

    bool format_processed = true;
    if (format_len == 1) {
        format_processed = iter != primitiveFormatMapping.end();
    } else if (format_len == 2) {
        if (strncmp("+s", format, 2) == 0) {
            // NANOARROW_TYPE_STRUCT
            if (get_struct_schema(env, schema, array, level, children, error) != 0) {
                return 1;
            }

            children_term = enif_make_list_from_array(env, children.data(), (unsigned)children.size());
            type_term = enif_make_tuple2(env, kAdbcColumnTypeStruct, children_term);
        } else if (strncmp("+r", format, 2) == 0) {
            // NANOARROW_TYPE_RUN_END_ENCODED (maybe in nanoarrow v0.6.0)
            // https://github.com/apache/arrow-nanoarrow/pull/507
            ERL_NIF_TERM run_ends_schema;
            if (get_run_end_encoded_schema(env, schema, level, run_ends_schema, error) != 0) {
                return 1;
            }

            type_term = enif_make_tuple2(env, kAdbcColumnTypeRunEndEncoded, run_ends_schema);
            children_term = make_adbc_column(env, schema, type_term, metadata);
        } else if (strncmp("+m", format, 2) == 0) {
            // NANOARROW_TYPE_MAP
            ERL_NIF_TERM map_kv_schema;
            if (get_map_schema(env, schema, level, map_kv_schema, error) != 0) {
                return 1;
            }

            type_term = enif_make_tuple2(env, kAdbcColumnTypeMap, map_kv_schema);
            children_term = make_adbc_column(env, schema, type_term, metadata);
        } else if (strncmp("+l", format, 2) == 0) {
            // NANOARROW_TYPE_LIST
            ERL_NIF_TERM elem_schema;
            if (get_list_element_schema(env, schema, level, elem_schema, error) != 0) {
                return 1;
            }

            type_term = enif_make_tuple2(env, kAdbcColumnTypeList, elem_schema);
            children_term = make_adbc_column(env, schema, type_term, metadata);
        } else if (strncmp("+L", format, 2) == 0) {
            // NANOARROW_TYPE_LARGE_LIST
            ERL_NIF_TERM elem_schema;
            if (get_list_element_schema(env, schema, level, elem_schema, error) != 0) {
                return 1;
            }

            type_term = enif_make_tuple2(env, kAdbcColumnTypeLargeList, elem_schema);
            children_term = make_adbc_column(env, schema, type_term, metadata);
        } else {
            format_processed = false;
        }
    } else if (format_len >= 3) {
        // handle all formats that start with `t` (temporal types)
        if (format[0] == 't') {
            if (format_len == 3) {
                // primitiveFormatMapping contains all temporal that have 3 characters
                format_processed = iter != primitiveFormatMapping.end();
            } else if (format_len >= 4 && format[1] == 's' && format[3] == ':') {
                // according to the arrow spec:
                //   The timezone string is appended as-is after the colon character :, 
                //   without any quotes. If the timezone is empty, the colon : must still be included.
                // so the format length for timestamps must be >= 4
            
                // possible format strings:
                // tss: - timestamp [seconds]
                // tsm: - timestamp [milliseconds]
                // tsu: - timestamp [microseconds]
                // tsn: - timestamp [nanoseconds]
                //
                // if there're any timezone infomation
                // it should be in the format like `tsu:timezone`
                // NANOARROW_TYPE_TIMESTAMP
                ERL_NIF_TERM term_unit;
                ERL_NIF_TERM term_timezone = kAtomNil;
                switch (format[2]) {
                    case 's': // seconds
                        term_unit = kAtomSeconds;
                        break;
                    case 'm': // milliseconds
                        term_unit = kAtomMilliseconds;
                        break;
                    case 'u': // microseconds
                        term_unit = kAtomMicroseconds;
                        break;
                    case 'n': // nanoseconds
                        term_unit = kAtomNanoseconds;
                        break;
                    default:
                        format_processed = false;
                }

                if (format_processed) {
                    if (format_len > 4 && format[3] == ':') {
                        std::string timezone(&format[4]);
                        term_timezone = erlang::nif::make_binary(env, timezone);
                    }
                    type_term = enif_make_tuple3(env, kAtomTimestamp, term_unit, term_timezone);
                    children_term = make_adbc_column(env, schema, type_term, metadata);
                }
            } else {
                format_processed = false;
            }
        } else {
            if (format_len == 3 && strncmp("+vl", format, 3) == 0) {
                // NANOARROW_TYPE_LIST(VIEW)
                ERL_NIF_TERM elem_schema;
                if (get_list_element_schema(env, schema, level, elem_schema, error) != 0) {
                    return 1;
                }

                type_term = enif_make_tuple2(env, kAdbcColumnTypeListView, elem_schema);
                children_term = make_adbc_column(env, schema, type_term, metadata);
            } else if (format_len == 3 && strncmp("+vL", format, 3) == 0) {
                // NANOARROW_TYPE_LARGE_LIST(VIEW)
                ERL_NIF_TERM elem_schema;
                if (get_list_element_schema(env, schema, level, elem_schema, error) != 0) {
                    return 1;
                }

                type_term = enif_make_tuple2(env, kAdbcColumnTypeLargeListView, elem_schema);
                children_term = make_adbc_column(env, schema, type_term, metadata);
            } else if (strncmp("+w:", format, 3) == 0) {
                // NANOARROW_TYPE_FIXED_SIZE_LIST
                unsigned n_items = 0;
                for (size_t i = 3; i < format_len; i++) {
                    n_items = n_items * 10 + (format[i] - '0');
                }
                type_term = kAdbcColumnTypeFixedSizeList(n_items);
                
                ERL_NIF_TERM elem_schema;
                if (get_list_element_schema(env, schema, level, elem_schema, error) != 0) {
                    return 1;
                }

                type_term = enif_make_tuple2(env, type_term, elem_schema);
                children_term = make_adbc_column(env, schema, type_term, metadata);
            } else if (strncmp("w:", format, 2) == 0) {
                // NANOARROW_TYPE_FIXED_SIZE_BINARY
                size_t nbytes = 0;
                for (size_t i = 2; i < format_len; i++) {
                    nbytes = nbytes * 10 + (format[i] - '0');
                }
                type_term = kAdbcColumnTypeFixedSizeBinary(nbytes);
                children_term = make_adbc_column(env, schema, type_term, metadata);
            } else if (format_len > 4 && (strncmp("+ud:", format, 4) == 0)) {
                // NANOARROW_TYPE_DENSE_UNION
                type_term = kAdbcColumnTypeDenseUnion;
                children_term = make_adbc_column(env, schema, type_term, metadata);
            } else if (format_len > 4 && (strncmp("+us:", format, 4) == 0)) {
                // NANOARROW_TYPE_SPARSE_UNION
                type_term = kAdbcColumnTypeSparseUnion;
                children_term = make_adbc_column(env, schema, type_term, metadata);
            } else if (strncmp("d:", format, 2) == 0) {
                // NANOARROW_TYPE_DECIMAL128
                // NANOARROW_TYPE_DECIMAL256
                //
                // format should match `d:P,S[,N]`
                // where P is precision, S is scale, N is bits
                // N is optional and defaults to 128
                int precision = 0;
                int scale = 0;
                int bits = 128;
                int * d[3] = {&precision, &scale, &bits};
                int index = 0;
                for (size_t i = 2; i < format_len; i++) {
                    if (format[i] == ',') {
                        if (index < 2) {
                            index++;
                        } else {
                            format_processed = false;
                            break;
                        }
                        continue;
                    }

                    *d[index] = *d[index] * 10 + (format[i] - '0');
                }

                if (format_processed) {
                    type_term = kAdbcColumnTypeDecimal(bits, precision, scale);
                    children_term = make_adbc_column(env, schema, type_term, metadata);
                }
            } else {
                format_processed = false;
            }
        }
    } else {
        format_processed = false;
    }

    if (!format_processed) {
        snprintf(err_msg_buf, sizeof(err_msg_buf)/sizeof(err_msg_buf[0]), "not yet implemented for format: `%s`", schema->format);
        error = erlang::nif::error(env, erlang::nif::make_binary(env, err_msg_buf));
        return 1;
        // printf("not implemented for format: `%s`\r\n", schema->format);
        // printf("length: %lld\r\n", values->length);
        // printf("null_count: %lld\r\n", values->null_count);
        // printf("offset: %lld\r\n", values->offset);
        // printf("n_buffers: %lld\r\n", values->n_buffers);
        // printf("n_children: %lld\r\n", values->n_children);
        // printf("buffers: %p\r\n", values->buffers);
    }

    out_terms.clear();
    out_terms.emplace_back(children_term);
    return 0;
}

#endif  // ADBC_ARROW_ARRAY_HPP