Microsoft KB Archive/104861

= FORTRAN PowerStation README.TXT: Linker, Library, VWB =

Article ID: 104861

Article Last Modified on 10/2/2003

-

APPLIES TO


 * Microsoft FORTRAN PowerStation 1.0 Standard Edition
 * Microsoft Fortran PowerStation 1.0a for MS-DOS

-



This article was previously published under Q104861



SUMMARY
The following information is from parts 3, 4, and 5 of the Microsoft FORTRAN PowerStation README.TXT file located in the \F32\README directory.   Part 3: Using the LINK32 Linker Part 4: Using the Run-Time Library Part 5: Using FORTRAN Visual Workbench



MORE INFORMATION
Part 3: Using the LINK32 Linker ---

Notes on Changes in File Formats

The FORTRAN PowerStation 1.0 compiler produces 32-bit COFF object files. The FORTRAN 5.1 compiler produces 16-bit OMF object files. These file formats are not compatible, and therefore you cannot link OMF object files with COFF object files.

Debugging Unresolved External References

If the linker reports an unresolved external name, and the routine does exist in your code, you should check its parameter list. The number following the "@" in the name given in the link error indicates the number of bytes in the parameter list of the calling statement. Verify that the calling statement and the subprogram declaration agree in their number and types of arguments.

Naming the Executable File Produced by LINK32 -

In order to use the FL32 command to compile and link in one step and to give the resulting executable file a specific name, use the /Fe option as in the following command line:

FL32 TEST.FOR /FeMYTEST

This will result in an executable file called MYTEST.EXE. The command line

FL32 TEST.FOR /link -out:MYTEST.EXE

will result in the following error:

Fatal error: Cannot open input file -- File not found "test.exe"

This is caused by the binding process failing to find the executable with the changed name. If the executable is then run, it will generate the error

This program cannot be run in DOS mode.

because it was not bound. Using the command

BINDMSF MYTEST.EXE

will correct this problem.

Getting Help from the MS-DOS Command Line -

To get the syntax for LINK32 help, type:

LINK32 -HELP

To get the syntax for LINK32 LINK options, type:

LINK32

To get the syntax for LINK32 LIB options, type:

LINK32 -LIB

To get the syntax for LINK32 DUMP options, type:

LINK32 -DUMP

Part 4: Using the Run-Time Library --

Floating-Point Coprocessor Limitation -

Some older 80386 systems have 287 coprocessors instead of 387s; some floating-point operations will not work correctly in these cases. To fix this, you must force use of the floating-point emulator by setting the TNT environment variable as follows. At the MS-DOS (or MS-DOS session) command line, enter "SET TNT=-FPEM" before running your compiled program. To cancel the forced emulation, you can enter "SET TNT=-NOFPEM".

Floating-Point Options --

Earlier versions of Microsoft FORTRAN had a variety of floating-point options:

- Emulator - Coprocessor only - Alternate math

FORTRAN PowerStation detects the presence of a coprocessor and will use it if it is present. Otherwise, it will emulate the floating-point coprocessor.

Floating-Point Exceptions and Run-Time Math Exceptions --

The following table shows the default behavior for floating-point exceptions and run-time math exceptions. Entries in the table are:

0.         A value of zero (value)    The value resulting from the operation 1#INF      IEEE infinity 1#IND      IEEE NaN (Invalid) M6nnn      Run-time error number M6nnn

When one of the values 0., (value), 1#INF, or 1#IND occurs, the program continues to run. When a run-time error (M6nnn) occurs, the program stops after reporting the error. The behavior depends on which compiler option you specify and whether you use the default run-time function MATHERRQQ or replace it.

Compiler Option and Resulting Default Exception Behavior

Compiler Option

(none)     /Op       /Ox /Op    /Ox Exception              -- - (FP exceptions) invalid                M6101       M6101     M6101      M6101 denormalized           (value)     (value)   (value)    (value) zerodivide             M6103       M6103     M6103      M6103 overflow               M6104       M6104     M6104      M6104 underflow              0. 0.       0.         0. inexact                 (value)     (value)   (value)    (value)

(Math exceptions) domain                 M6201       M6201     M6201      (undefined) singular               M6202       M6202     M6202      (undefined) overflow               M6203       M6203     M6203      (undefined) total loss of precision M6205      M6205     M6205      (undefined)

This behavior results from the default call of SETCONTROLLFPQQ (done by the run-time library when your program starts) that includes the exception control flags

FPCW$DENORMAL + FPCW$UNDERFLOW + FPCW$INEXACT

and the inclusion of the default run-time subroutine MATHERRQQ. (The behavior of math exceptions for /Ox is undefined; either an IEEE value results or a run-time error occurs.)

If you want behavior consistent with IEEE precision and exception handling, compile with /Op and explicitly call SETCONTROLFPQQ with all exception control flags:

FPCW$INVALID + FPCW$DENORMAL + FPCW$ZERODIVIDE + FPCW$OVERFLOW + FPCW$UNDERFLOW + FPCW$INEXACT

In addition, you must link a replacement subroutine MATHERRQQ that just returns a successful return code. The next table shows the IEEE behavior that results.

IEEE Precision and Exception Behavior

/Op + Flags + MATHERRQQ Exception - (FP exceptions) invalid                            1#IND denormalized                       (value) zerodivide                         1#INF overflow                           1#INF underflow                          0. inexact                            (value)

(Math exceptions) domain                             1#IND singular                           1#IND overflow                           1#INF total loss of precision            0.

The sample programs include a program SPECIAL.FOR that illustrates tailoring the floating-point environment to be the closest possible to IEEE. It calls SETCONTROLFPQQ with the appropriate flags and provides a replacement MATHERRQQ.

Part 5: Using FORTRAN Visual Workbench --

Fonts used in FORTRAN Visual Workbench --

FORTRAN Visual Workbench will only allow screen fonts. Most simulated fonts cannot be displayed. Therefore, many boldface fonts are not available. All TrueType fonts, bold and italic, can be used. Occasionally the caret will be positioned incorrectly for some bold or italic non-TrueType fonts. If you select a font and the caret is positioned incorrectly, chose a different font.

Programs with Very Long Paths -

If a program has a very long path, then choosing Execute from the Project menu may bring up a message box titled "F32EXEC.PIF," which says that the program cannot be found, although the .EXE file was created and can be run outside of FORTRAN Visual Workbench.

Time-Out Problems in FORTRAN Visual Workbench -

If you have time-out problems with FORTRAN Visual Workbench, check to see if WinExclusive=1 is in the 386enh section of SYSTEM.INI. If this is set and Windows has the focus, then Windows gets all the timeslices. (Therefore Wintee can't initialize an MS-DOS VM.)

WinExclusive is set from the "Exclusive in Foreground" check box in Windows Control Panel, 386Enhanced. For FORTRAN Visual Workbench, the box should NOT be checked.

Using Extended Characters in FORTRAN Visual Workbench -

Because FORTRAN Visual Workbench runs in Microsoft Windows and produces programs that run in MS-DOS, two distinct character sets are used: ANSI and OEM. The ANSI set is normally used in Windows; the OEM set is used in MS-DOS. Both character sets contain extended characters (such as accented letters and line-drawing symbols), but the numerical representation of these characters is different. Windows provides a special font called Terminal that uses the OEM character set. If you use extended characters in your files, you need to know about both.

FORTRAN Visual Workbench is an ANSI editor; it does not convert characters to the OEM character set. If you want to see the true OEM characters, use the TERMINAL font. If you want to enter OEM characters, add a zero to the beginning of the character code. (For example, "ALT+0130" produces an accented "e"; typing "ALT+130" produces the ANSI character.) No translation is performed when files are loaded or saved. The editor displays the character representation of what was in the file. When a character is entered, it is translated to ANSI and saved as that value. This applies for all characters on the keyboard, including an e-accent on an international keyboard and an ALT+number entered from the numeric key pad. The only time a translation is not done is when an ALT+0+number is entered. When looking at a file from MS-DOS, you will get the OEM representation of the characters.

Microsoft recommends that you do not use extended characters in filenames.

The sample program GRDEMO in the ..\F32\SAMPLES\DEMO directory uses extended characters to produce a boxed menu on the screen. The file MENU.FOR uses extended characters to draw a box around the menu. The line characters display as such only if the font is set to Terminal.

Sample Programs in Book, Help, and Disk ---

Some of the sample programs on the distribution disks are newer versions than those that appear in the "Language Guide."

Additional query words: 1.00

Keywords: KB104861

-

[mailto:TECHNET@MICROSOFT.COM Send feedback to Microsoft]

© Microsoft Corporation. All rights reserved.