当前位置：首页 > news >正文

利用DeepSeek辅助修改luadbi-duckdb读取DuckDB decimal数据类型

news 2025/11/3 7:37:51

DuckDB的decimal数据类型在duckdb.h的定义如下，

//! DECIMAL is composed of a width and a scale.
//! Their value is stored in a HUGEINT.
typedef struct {uint8_t width;uint8_t scale;duckdb_hugeint value;
} duckdb_decimal;

但实际上value的类型是变化的，通常的方法无法正确读取，参考https://github.com/duckdb/duckdb/issues/10986 回帖的方法

var logicalType = duckdb_vector_get_column_type(vector);
byte scale = duckdb_decimal_scale(logicalType);
var type = duckdb_decimal_internal_type(logicalType);
var data = duckdb_vector_get_data(vector);
//cast data to either short, int, long or hugeint depending on the type variable above and divide it by 10 to the power of scale
If it helps you can find C# version here: https://github.com/Giorgi/DuckDB.NET/blob/8520bf5005d9309f762ef61d71412d60d24ca32c/DuckDB.NET.Data/Internal/Reader/DecimalVectorDataReader.cs#L43

让DeepSeek编写函数，根据输入的数据类型输出decimal(width,scale,data,type)的c字符串，当type==1,data是指向int8_t的指针,type==2,指向int16_t, type==3,指向int32_t,type==4,指向int64_t,输出结果是 *data / （10^scale)，同时小数点后保留scale位的c字符串。
他给出的代码如下：

// 将decimal数据转换为字符串
void decimal_to_string_format(uint8_t width, uint8_t scale, void* data, int type, char* buffer, size_t buffer_size) {if (data == NULL || buffer == NULL || buffer_size == 0) {return;}// 根据类型获取整数值int64_t int_value = 0;switch (type) {case DUCKDB_TYPE_TINYINT: // int8_tint_value = *(int8_t*)data;break;case DUCKDB_TYPE_SMALLINT: // int16_tint_value = *(int16_t*)data;break;case DUCKDB_TYPE_INTEGER: // int32_tint_value = *(int32_t*)data;break;case DUCKDB_TYPE_BIGINT: // int64_tint_value = *(int64_t*)data;break;case DUCKDB_TYPE_HUGEINT: // int128_t{duckdb_hugeint huge_val = *(duckdb_hugeint*)data;// 简化处理：如果upper为0，只使用lower部分if (huge_val.upper == 0) {// 直接使用lower部分uint64_t value = huge_val.lower;int is_negative = (huge_val.upper < 0); // 检查符号if (scale == 0) {snprintf(buffer, buffer_size, "%s%lu", is_negative ? "-" : "", value);} else {uint64_t divisor = 1;for (int i = 0; i < scale; i++) {divisor *= 10;}uint64_t integer_part = value / divisor;uint64_t fractional_part = value % divisor;char format[32];snprintf(format, sizeof(format), "%%s%%lu.%%0%dlu", scale);snprintf(buffer, buffer_size, format, is_negative ? "-" : "", integer_part, fractional_part);}} else {// 对于非常大的HUGEINT值，使用简化表示if (scale == 0) {snprintf(buffer, buffer_size, "%s%ld%09lu", huge_val.upper < 0 ? "-" : "", huge_val.upper < 0 ? -huge_val.upper : huge_val.upper,huge_val.lower);} else {// 对于有小数位的大数，输出简化信息snprintf(buffer, buffer_size, "hugeint[upper=%ld,lower=%lu,scale=%d]", huge_val.upper, huge_val.lower, scale);}}return;}break;default:// 无效类型snprintf(buffer, buffer_size, "decimal(%d,%d,invalid_type,%d)", width, scale, type);return;}// 处理负号int is_negative = (int_value < 0);uint64_t abs_value = is_negative ? (uint64_t)(-int_value) : (uint64_t)int_value;if (scale == 0) {// 整数情况，直接输出snprintf(buffer, buffer_size, "%s%lu", is_negative ? "-" : "", abs_value);} else {// 计算10^scaleuint64_t divisor = 1;for (int i = 0; i < scale; i++) {divisor *= 10;}// 分离整数部分和小数部分uint64_t integer_part = abs_value / divisor;uint64_t fractional_part = abs_value % divisor;// 构建格式字符串，确保小数部分有正确的位数（前导零）char format[32];snprintf(format, sizeof(format), "%%s%%lu.%%0%dlu", scale);// 格式化输出snprintf(buffer, buffer_size, format, is_negative ? "-" : "", integer_part, fractional_part);}
}

再自行在函数statement_fetch_impl中添加:

				case DUCKDB_TYPE_DECIMAL:{duckdb_logical_type logicalType = duckdb_vector_get_column_type(col1);
uint8_t width = duckdb_decimal_width(logicalType);
uint8_t scale = duckdb_decimal_scale(logicalType);
duckdb_type type = duckdb_decimal_internal_type(logicalType);
void* data = duckdb_vector_get_data(col1);
char decimal_str[100];
decimal_to_string_format(width, scale, data, type, decimal_str, sizeof(decimal_str) );
duckdb_string_t str= convert_to_duckdb_string(decimal_str);if (named_columns) {LDB_PUSH_ATTRIB_STRING( str );} else {LDB_PUSH_ARRAY_STRING(i + 1, str);}}break;

编译和运行

oot@6ae32a5ffcde:/par/luadbi-0.7.5# luarocks make luadbi-duckdb-scm-0.rockspec --pack-binary-rock DUCKDB_DIR=/par/141 DUCKDB_INCDIR=/par/141Packed: /par/luadbi-0.7.5/luadbi-duckdb-scm-0.linux-x86_64.rockroot@6ae32a5ffcde:/par/luadbi-0.7.5# luarocks install luadbi-duckdb-scm-0.linux-x86_64.rock --forceluadbi-duckdb scm-0 is now installed in /usr/local (license: MIT/X11)root@6ae32a5ffcde:/par/luadbi-0.7.5# lua /par/duckdtpath.lua
[123456789012345.79000]15
2025-09-25 12:26:55.123456
123456789012345.79000
2025-09-23root@6ae32a5ffcde:/par/luadbi-0.7.5# lua /par/duckdtpath.lua
[hugeint[upper=6,lower=12776324570088369304,scale=5]]15
2025-09-25 12:26:55.123456
hugeint[upper=6,lower=12776324570088369304,scale=5]
2025-09-23root@6ae32a5ffcde:/par/luadbi-0.7.5# lua /par/duckdtpath.lua
[hugeint[upper=-1,lower=6101065172474972616,scale=5]]15
2025-09-25 12:26:55.123456
hugeint[upper=-1,lower=6101065172474972616,scale=5]
2025-09-23

对于hugeint高位不为0的情况，简化了处理。
测试脚本如下

DBI = require "DBI"dbd, err = DBI.Connect( 'DuckDB', 'lua_duckdb', 'dbuser', 'password' )
assert(dbd, err)dbd:autocommit(true)
statement = dbd:prepare( "create or replace table table_1 as select 15 id, timestamp'2025-09-25 12:26:55.123456' column1,-123456789012345.79::decimal(38, 5) column2,date'2025-09-23' column3 ;" )
statement:execute()statement = assert(dbd:prepare( "select * from table_1 where id = $1;" ))
statement:execute( 15 )for row in statement:rows(true) doprint(row['id'])print(row['column1'])print(row['column2'])print(row['column3'])end