BUG: Bad Results Displayed by CodeView for Huge COMMON Blocks

• Microsoft CodeView for MS-DOS, versions 2.3, 3.0, and 3.11
• Microsoft CodeView for OS/2, versions 2.3, 3.0, and 3.11

SYMPTOMS

Microsoft CodeView versions 2.30, 3.00, and 3.11 erroneously calculate the addresses of variables in FORTRAN COMMON blocks when these variables are preceded in the COMMON block by 65,535 or more bytes of data. The segment values of the addresses of these variables are calculated incorrectly, generating incorrect results for any CodeView function that uses symbolic names to display information. In particular, watch-expressions will show incorrect values for the contents of these variables.

RESOLUTION

If viewing a specific data item is necessary, it should be moved to the beginning of the COMMON block or placed in a separate COMMON block. If this is not possible, the LOCFAR intrinsic function can be used with FORTRAN versions 4.0, 4.01, 4.1, and 5.0 to print out the correct memory address of the data item, which can then be used with a CodeView dump command, a watchpoint, or a breakpoint to view the required information.

STATUS

Microsoft has confirmed this to be a bug in CodeView versions 2.3, 3.0, and 3.11 for MS-DOS and OS/2.

This is not an issue with FORTRAN PowerStation.

MORE INFORMATION

The following program example illustrates the problem:

Sample Code

      parameter(n=65535)               ! 64K less one byte

      integer*1 arr1, arr2

      common /test/ arr1(n), arr2(10)

      arr1(1) = 1
      arr2(1) = 2

      write(*,'(a,z)') ' locfar(arr1) = ',locfar(arr1)
      write(*,'(a,z)') ' locfar(arr2) = ',locfar(arr2)

      end

Within CodeView, the offset of a variable preceded by 65,535 bytes in a COMMON block is always correctly determined, but the segment address is not. In the example above, if arr1 is increased in size, the memory locations of the two arrays are no longer mapped directly on top of each other by CodeView because the offsets will no longer be identical.

CodeView does not appear to have a general problem with huge data items or items in COMMON blocks. The problem only occurs with data items in COMMON blocks when the starting address of the data item begins on or beyond the boundary of the second segment of the COMMON block. If a huge (greater than 64K bytes) data item has a starting address in the first segment of the COMMON block, all the elements of that data item are correctly displayed by CodeView.