PRB: Span-Dependent Value Behavior Changes in N-Pass Assembly

• Microsoft Macro Assembler for MS-DOS, versions 6.0, 6.0a, 6.0b, 6.1, 6.1a, and 6.11
• Microsoft Macro Assembler for OS/2, versions 6.0, 6.0a, 6.0b

SYMPTOMS

When the Microsoft Macro Assembler (MASM) assembles code that depends on the difference between the values of two labels (a span-dependent value), the results of an assembly by MASM versions 6.0 or later differs from the results produces by an earlier version of MASM.

Specifically, the difference occurs when the application uses a span- dependent value in one of the following cases:

• In the expression for a preprocessor directive, like IF
• In the expression for a predefined macro, like REPEAT or WHILE
• In the arguments to a MACRO or TEXTEQU statement preceded by % (percent sign)

CAUSE

MASM versions 5.1 and earlier are two-pass assemblers; MASM versions 6.0 and later are N-pass assemblers.

RESOLUTION

Avoid using a span-dependent value in a preprocessor directive or in a predefined macro. If the code requires a span-dependent value, create and review a listing file to determine that each span-dependent value is evaluated as desired.

Avoid using the % operator with span-dependent values. The assembler evaluates an expression containing the % operator on the first pass. In many cases, the assembler produces the desired results when you remove the % operator and delay expression evaluation to a later pass.

STATUS

This behavior is under review and may or may not change in a future release.

MORE INFORMATION

MASM versions 5.1 and earlier are two-pass assemblers. The source file is read twice to determine the size of the instructions and the location of labels.

MASM versions 6.0 and later are N-pass assemblers. The assembler scans the source until it can determine the size of all instructions and the location of all labels. However, an N-pass assembler must determine certain values on its first pass through the application code. These values include expressions in preprocessor directives, expressions in predefined macros, and arguments preceded by the % operator. If one of these values is based on the difference between the values of two labels, the results may differ from those generated by MASM versions 5.1 and earlier.

The difference occurs because the assembler evaluates expressions and arguments based on the undetermined label locations available during the first assembly pass that may change in a subsequent assembly pass. Typically, the value changes when a forward reference occurs between the two labels. Because the assembler adds padding bytes for the forward reference and removes the padding in subsequent assembly passes, the second label has a larger value during the first assembly pass than in subsequent passes.

The code example below demonstrates this behavior. Specify the /Fl assembler option switch to create a listing file. Review the listing file to see how the assembler processes a span-dependent value.

The following information is part of the README.TXT file distributed with MASM version 6.1:

   Span-Dependent Expressions used in Macros
   -----------------------------------------

   MASM 6.1 evaluates macro expressions only on the first pass of assembly,
   but code and data are reevaluated on subsequent passes. Because of this,
   macro expressions which depend on the span between two addresses may not
   evaluate correctly. For example, the following code will not evaluate
   correctly:

      Label1:
         JMP Label2
      Label2:
         REPEAT Label2 - Label1  ; Evaluates incorrectly
         INC AX
      ENDM

   View the listing file to determine if a questionable macro expression
   was evaluated as desired.

   Span-Dependent Text Equates
   ---------------------------

   The TEXTEQU operator is evaluated on the first assembly pass. If TEXTEQU
   is used with an expression that depends on the difference between two
   addresses, the resulting constant may be incorrect. For example, the
   following code will not evaluate correctly:

      Label1:
         JMP Label2
      Label2:
         WrongNum TEXTEQU %Label2-Label1 ; WrongNum will be incorrect

Sample Code

; Assemble options needed in MASM 6.x: /c /Fl

_text   SEGMENT para public 'CODE'
        ASSUME cs:_text

        jmp SHORT forward

        mov ax, 4C00h
        int 21h
_text   ENDS

IF count LT 6

   %OUT two-pass assembler   ; MASM 5.1 and earlier shows this message
ELSE
   %OUT n-pass assembler     ; MASM 6.0 and later shows this message

tst1    SEGMENT para public

        WHILE index LT count
          DB 1          ; This will be repeated 10 times in MASM 6.x.

        ENDM
tst1    ENDS

tst2    SEGMENT para public
        REPT count      ; This will be repeated 10 times in MASM 6.x,
          DB 1          ; twice in MASM 5.1.
        ENDM
tst2    ENDS

mac3    MACRO arg1
        macvar = arg1
        ENDM

tst3    SEGMENT para public
        mac3 %count     ; This will set macvar to 10 in MASM 6.x
tst3    ENDS            ; and to 2 in MASM 5.1.

txt4    TEXTEQU %count

tst4    SEGMENT para public
txtvar  DB txt4         ; This will set txtvar to 10 in MASM 6.x.
tst4    ENDS