Writing UDFs for Firebird Embedded SQL Server

This article was written mainly for developers who use Firebird Embedded SQL Server in .Net framework applications and want to speed up or optimize DB queries.

We will describe how to create your own native Firebird extension and show some approaches how to use it in managed code applications.

1. What is UDFs
2. What do you require
3. Sample project structure
4. About the sample
5. Creating UDFs-based binary data parser
6. Outside initialization
7. Conclusion

What is UDFs

User-defined functions (UDFs) are host-language programs for performing frequently needed tasks, supplementing built-in SQL functions such as MIN() and MAX(). UDFs are extensions to the Firebird server and execute as part of the server process. Another words UDFs are compiled functions and linked to a dynamically-linked library.

What do you require

To effectively read this article you need some advanced knowledge of C++, C# and Firebird SQL. Also you require having VS 2005 with C# and C++ installed to build sample. The sample already has “batteries inside”: all files required for Firebird embedded server and ADO.NET provider are placed in the sample’s Firebird folder and used in post-build event.

Sample project structure

.Firebird – folder with all files required for Firebird Embedded SQL Server
|→ .include – folder with *.h files required for UDFs compiling
|→ .lib – folder with static libraries required for UDFs linking
.MainApp – sample managed application
. SampleUdf – sample UDFs dll

About the sample

The sample project shows how to transfer binary data (BLOB) from one table using UDFs-based parser object to another table:

[Code from .MainAppbatch.sql]

SQL

CREATE TABLE "RowDataTable" (
    "Id" INTEGER NOT NULL PRIMARY KEY
      "Value" BLOB
  )

CREATE TABLE "FSTable" (
   "Id" INTEGER NOT NULL PRIMARY KEY
  , "Name" VARCHAR(256)
  , "FullPath" VARCHAR(256)
  , "CreationTime" TIMESTAMP 
  , "Attributes" INTEGER
  , "Size" BIGINT
  )

Creating UDFs-based binary data parser

UDFs Declaration

Firebird SQL has following syntax to declare UDF:

SQL

DECLARE EXTERNAL FUNCTION name [datatype | CSTRING (int)
[, datatype | CSTRING (int) ...]]
RETURNS {datatype [BY VALUE] | CSTRING (int)} [FREE_IT]
[RETURNS PARAMETER n]
ENTRY_POINT 'entryname'
MODULE_NAME 'modulename';

Argument	Description
`name`	Name of the UDF to use in SQL statements; can be different from the name of the function specified after the `ENTRY_POINT` keyword
`datatype`	Datatype of an input or return parameter • All input parameters are passed to the UDF by reference • Return parameters can be passed by value • Cannot be an array element
`RETURNS`	Specifies the return value of a function. Pay attention that syntax does not allow to declare UDF that returns nothing
`BY VALUE`	Specifies that a return value should be passed by value rather than by reference
`CSTRING (int)`	Specifies the UDF that returns a null-terminated string int bytes in length
`FREE_IT`	Frees memory allocated for the return value after the UDF finishes running. This parameter should be used with ib_util_malloc memory allocation function in UDF implementation, It’s contained in: Header : ib_util.h Library: ib_util_ms.lib DLL: ib_util.dll
`RETURNS PARAMETER n`	Specifies that the function returns the nth input parameter; is required for returning Blobs
`'entryname'`	Quoted string specifying the name of the UDF in the source code and as stored in the UDF library
`'modulename'`	Quoted file specification identifying the dll that contains the UDF

All dlls with UDFs should be placed to UDF folder in the application root where fbembedded.dll is stored. When declaration query is going for execution Firebird engine does not require UDF dll to be placed in UDF folder in that moment. But when executing some stored procedure creation query that contains UDF call engine will check required external function in dll.

Here are some UDF declaration examples from sample project:

[Code from .MainAppbatch.sql]

SQL

DECLARE EXTERNAL FUNCTION CreateParser
    BLOB
    RETURNS INTEGER BY VALUE
  ENTRY_POINT 'SampleUdf_CreateParser'  MODULE_NAME 'SampleUdf'

DECLARE EXTERNAL FUNCTION DestroyParser
    INTEGER
    RETURNS INTEGER BY VALUE
  ENTRY_POINT 'SampleUdf_DestroyParser'  MODULE_NAME 'SampleUdf'

DECLARE EXTERNAL FUNCTION GetName
    INTEGER
    RETURNS CSTRING(256)
  ENTRY_POINT 'SampleUdf_GetName'  MODULE_NAME 'SampleUdf'

DECLARE EXTERNAL FUNCTION GetCreationTime
    INTEGER
    RETURNS TIMESTAMP FREE_IT
  ENTRY_POINT 'SampleUdf_GetCreationTime'  MODULE_NAME 'SampleUdf'

DECLARE EXTERNAL FUNCTION GetSize
    INTEGER
    RETURNS BIGINT
  ENTRY_POINT 'SampleUdf_GetSize'  MODULE_NAME 'SampleUdf'

UDFs implementation

As you may guess, I’m using integer type parameter to send parser object that was created in SampleUdf.dll heap. With parser object everything is very simple:

[Code from .SampleUdfmain.cpp]

C++

class SampleParser
{
    std::vector<char> _buffer;
    size_t _pathLen;
    char * _name;
public:
    SampleParser(std::vector<char> & buffer)
    {
        _buffer.swap(buffer);
        char * path = (char*)&_buffer.at(0);
        _pathLen = strlen(path);
        if(_pathLen < 1 ||  _pathLen >= _buffer.size())
            throw std::exception("Invalid buffer format");
        _name = strrchr(path, '');
        if(!_name) _name = path;
        else ++_name;
        //ON_MESSAGE("--SampleParser created--")
    }
    ~SampleParser()
    {
        //ON_MESSAGE("--SampleParser destroyed--")
    }

    char * GetName() { return _name; }
    char * GetFullPath() { return (char*)&_buffer.at(0); }
    __int64 * GetCreationTime() { return (__int64*)&_buffer.at(_pathLen + 1); }
    int * GetAttributes() { return (int*)&_buffer.at(_pathLen + 1 + sizeof(__int64)); }
    __int64 * GetSize() { return (__int64*)&_buffer.at(_pathLen + 1 + sizeof(__int64) + sizeof(int)); }
};

The next UDF demonstrates how parser is created and also the work with BLOB data:

C++

extern "C" __declspec(dllexport) SampleParser * SampleUdf_CreateParser(BLOBCALLBACK data)
{
    try
    {
        std::vector<char> buffer(data->blob_total_length);
        ISC_UCHAR * p = (ISC_UCHAR*)&buffer.front();
        for(ISC_LONG i=0; i < data->blob_number_segments; ++i)
        {
            ISC_USHORT length = 0;
            data->blob_get_segment(data->blob_handle, p, data->blob_max_segment, &length);
            p+= length;
        }
        return new SampleParser(buffer);
    }
    catch(std::exception & ex)
    {
        ON_ERROR(ex.what());
    }
    catch(...)
    {
        ON_ERROR("Unknown error");
    }
    return 0;
}

And now let’s show how to use parser object. The following function also shows how to:

convert FILETIME structure to Firebird TIMESTAMP
use embedded server memory allocation, when you transfer memory ownership to the database engine

C++

extern "C" __declspec(dllexport) ISC_TIMESTAMP * SampleUdf_GetCreationTime(int * ptr)
{
    try
    {
        SampleParser * self = (SampleParser*)(*ptr);

        FILETIME localTime;
        if(!::FileTimeToLocalFileTime((const FILETIME*)self->GetCreationTime(), &localTime))
            return 0;
        SYSTEMTIME st;
        if(!::FileTimeToSystemTime(&localTime, &st))
            return 0;

        ISC_TIMESTAMP * timeStamp = (ISC_TIMESTAMP*)ib_util_malloc(sizeof(ISC_TIMESTAMP));

        timeStamp->timestamp_time = (st.wHour * 3600000 + st.wMinute * 60000 + st.wSecond * 1000 + st.wMilliseconds) * 10;

        WORD day = st.wDay;
        WORD month = st.wMonth;
        WORD year = st.wYear;

        //some magic calculations from ADO.NET Provider code
        if (month > 2)
            month -= 3;
        else
        {
            month += 9;
            year -= 1;
        }
        WORD c = year / 100;
        WORD ya = year - 100 * c;

        timeStamp->timestamp_date = ((146097 * c) / 4 + (1461 * ya) / 4 + (153 * month + 2) / 5 + day + 1721119 - 2400001);
        return timeStamp;
    }
    catch(std::exception & ex)
    {
        ON_ERROR(ex.what());
    }
    catch(...)
    {
        ON_ERROR("Unknown error");
    }
    return 0;
}

As you can see all input and output parameters in UDF implementations are pointers, except output parameters declared with BY VALUE modifier. But this modifier can be used not with all data types (for example it’s incompatible with TIMESTAMP, but compatible with BIGINT – it’s strange behavior for the types of the same size, isn’t it?)

Using UDFs

The next stored procedure demonstrates how to use created parser:

[Code from .MainAppbatch.sql]

SQL

CREATE PROCEDURE TransferData
RETURNS ("counter" INTEGER)
AS
DECLARE VARIABLE "tmp" INTEGER;
DECLARE VARIABLE "parserPtr" INTEGER;
DECLARE VARIABLE "Value" BLOB;
DECLARE VARIABLE "Name" VARCHAR(256);
DECLARE VARIABLE "FullPath" VARCHAR(256);
DECLARE VARIABLE "CreationTime" TIMESTAMP;
DECLARE VARIABLE "Attributes" INTEGER;
DECLARE VARIABLE "Size" BIGINT;
BEGIN
    "counter" = 0;
    FOR SELECT "Value" FROM "RowDataTable" INTO :"Value" DO    BEGIN
        SELECT CreateParser(:"Value") FROM rdb$database INTO :"parserPtr";            
        IF ("parserPtr" IS NOT NULL) THEN BEGIN
            SELECT GetName(:"parserPtr") FROM rdb$database INTO :"Name";
            SELECT GetFullPath(:"parserPtr") FROM rdb$database INTO :"FullPath";
            SELECT GetCreationTime(:"parserPtr") FROM rdb$database INTO :"CreationTime";
            SELECT GetAttributes(:"parserPtr") FROM rdb$database INTO :"Attributes";
            SELECT GetSize(:"parserPtr") FROM rdb$database INTO :"Size";
            
            "tmp" = GEN_ID("FSTable_Generator", 1);
            INSERT INTO "FSTable" ("Id", "Name", "FullPath", "CreationTime", "Attributes", "Size")
            VALUES (:"tmp", :"Name", :"FullPath", :"CreationTime", :"Attributes", :"Size");
            "counter" = "counter" + 1;

            SELECT DestroyParser(:"parserPtr") FROM rdb$database INTO :"tmp";            
        END
    END
    SUSPEND;
END

About strange form of calling :

SQL

SELECT <UDF_Name>(<Parameters_List>) FROM rdb$database INTO <Output_parameter>

It’s the only way to make it working in embedded dll server 🙂

Outside initialization

Thanks to the windows caching there is a possibility to load and initialize UDFs dll before it is loaded by Firebird database engine. In my sample I used this possibility to make some useful callbacks:

[Code from .MainAppSampleProvider.cs]

private delegate void MessageCallbackDelegate([MarshalAs(UnmanagedType.LPStr)] string message);

        private static MessageCallbackDelegate messageCallback;
        private static MessageCallbackDelegate errorCallback;

        [DllImport("udf/SampleUdf")]
        private static extern void RegisterCallbacks(MessageCallbackDelegate messageCallback, MessageCallbackDelegate errorCallback);

        static SampleProvider()
        {
            messageCallback = MessageCallback;
            errorCallback = ErrorCallback;
            RegisterCallbacks(messageCallback, errorCallback);
        }

[Code from .SampleUdfmain.cpp]

typedef void (__stdcall * FCallback)(const char * message);
FCallback g_messageCallback = 0;
FCallback g_errorCallback = 0;
//---------------------------------------------------------------------------
#define ON_MESSAGE(mess) { if(g_messageCallback) g_messageCallback(mess); }
#define ON_ERROR(mess) { if(g_errorCallback) g_errorCallback(mess); }
//---------------------------------------------------------------------------
extern "C" __declspec(dllexport) void RegisterCallbacks(FCallback messageCallback, FCallback errorCallback)
{
    g_messageCallback = messageCallback;
    g_errorCallback = errorCallback;
}