Considerations for Programming DATA NONE DLLs

• Microsoft Windows Software Development Kit (SDK) versions 3.0, 3.1

SUMMARY

The Windows Software Development Kit (SDK) documentation regarding dynamic-link libraries (DLLs) is not very detailed with regard to some facets of developing DLLs. In this article, some of the lesser known considerations for developing DLLs are discussed.

MORE INFORMATION

The primary reason for using a DATA NONE DLL is to ensure that the data segment (DS) is equal to the stack segment (SS). Many of the functions provided with the Microsoft C run-time libraries expect that DS == SS. Functions with this property have been removed from the Windows SDK DLL libraries. In DATA SINGLE DLLs, the DLL will use the application's SS, but the DLL will have its own DS. DS is then not equal to SS and the C run-time functions (or other functions expecting DS == SS) may not be used. In DATA NONE DLLs, however, the DLL uses the application's DS and SS. In applications, DS == SS; therefore, for a DATA NONE DLL, DS == SS.

While this may sound like a straightforward process, there are some items declared within a DLL that can create a data segment. Windows looks at a flag to determine if the DLL has a data segment. If DATA NONE has been declared, Windows will treat that DLL as a DATA NONE DLL whether a data segment exists or not. However, the Windows loader uses information from the linker to determine what segments to load, not how the flags are set. Therefore, data segments that do exist are loaded into memory.

When additional data segments are created, because the compiler works under the assumption that DS == SS, the data segments that are loaded are not used and waste memory.

When a DATA NONE DLL does have a data segment, Windows will use the application's DS as it should, but the offset into the data segment is the offset into the DLL's data segment. Writing to any variable declared in the DLL will overwrite data in the application's data segment.

This type of bug is often difficult to track down. When the data segment is built for the DLL, items are placed in the data segment from bottom to top. So an integer variable will take up 2 bytes on the bottom of the DLL's data segment. Writing to this variable will overwrite 2 bytes at the bottom of the calling application's data segment. An application's data segment normally contains thousands of bytes; therefore, this type of error may go unnoticed and will not cause a GP-fault.

Global variables, static variables, and static text (string literals such as "Hello") declared in the DLL will create new data segments for the DATA NONE DLL.

The warning message "RC: Warning RW4002: Non-discardable Segment 2 set to PRELOAD" is quite common when DLLs are processed by the Resource Compiler. However, it is not immediately obvious from the text of the message why this is a problem. This warning is generated when a data segment has been created but flags have not been defined for that segment (such as NONE, PRELOAD, or SINGLE). The default setting for these segments is PRELOAD SINGLE.

After ruling out all the obvious ways that a new data segment can be created, it is time to look deeper into the heart of the code.

If the DLL uses the Microsoft C run-time code, a global internal variable is used by the library. This variable is referenced in LIBENTRY.ASM by the following code:

   extern __acrtused:abs 

Additional query words:

Keywords : kb16bitonly KrDll
Version : WINDOWS:3.0,3.1
Platform : WINDOWS
Issue type : kbhowto