PRB: Structure Packing Causes Some Win32 Programs to Fail (117388)


The information in this article applies to:

  • Microsoft Visual C++, 32-bit Editions 1.0
  • Microsoft Visual C++, 32-bit Editions 2.0
  • Microsoft Visual C++, 32-bit Editions 2.1
  • Microsoft Visual C++, 32-bit Editions 2.2
  • Microsoft Visual C++, 32-bit Editions 4.0
  • Microsoft Visual C++, 32-bit Editions 4.1
  • Microsoft Visual C++, 32-bit Editions 4.2
  • Microsoft Visual C++, 32-bit Editions 5.0
  • Microsoft Visual C++, 32-bit Editions 6.0
  • Microsoft Win32 Software Development Kit (SDK) 3.1
  • Microsoft Win32 Software Development Kit (SDK) 3.5
This article was previously published under Q117388

SYMPTOMS

The default packing (alignment) for structures in Windows NT is 8 bytes. This setting is the default for projects created using the Visual Workbench. Programs built with other packing build without error, but the resulting executable may produce run-time errors. These errors include incorrect data, error return codes from Win32 APIs, or program failure. On RISC platforms, such as MIPS and Alpha, doubles and long ints that are not 8-byte-aligned cause alignment faults.

CAUSE

The default packing on Windows NT is 8-byte packing. This packing is used both for efficiency reasons and for portability (some RISC processors allow only 8-byte packing). Because the operating system is expecting data packed on 8-byte boundaries, it may incorrectly process structures passed from programs that were not packed on 8-byte boundaries. Not all programs exhibit this behavior. One example that does give problems with a different kind of packing is shown in the "MORE INFORMATION" section.

RESOLUTION

If a program builds without errors or warnings but does not run correctly, one thing to check is that all modules are compiled with /Zp8 (or no /Zp switch at all).

MORE INFORMATION

Many MS-DOS and Windows programs are built with 1- or 2-byte packing. Old makefiles may have a /Zp1 or /Zp2 setting included. When porting programs with Win32, this setting should be changed. In addition, check for a /Zp setting in the CL environment variable in the AUTOEXEC.BAT file. This setting overrides the setting in the Visual Workbench or in a makefile.

While 8-byte packing is the standard for Windows NT, in many cases 4-byte packing works. Macro Assembler, version 6.11 and earlier, does not allow 8-byte packing. Use the highest available packing in this case. In some cases, you may not be able to get the program working and it may have to be rewritten in C or C++.

REFERENCES

For more information on alignment, see the "Programming Techniques" book included with the Visual C++, 32-bit Edition, documentation set.

For general porting guidelines, please see the article, "Port Your 16-bit Applications to Windows NT Without Ripping Your Hair Out," in the "Microsoft Systems Journal," August 1993.
Modification Type:MajorLast Reviewed:12/2/2003
Keywords:kbCompiler kbprb KB117388
©2003 Microsoft Corporation. All rights reserved.