Announcement

Collapse
No announcement yet.

Exception handlers

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Paul Dixon
    replied
    There's no real conversion to do to make it work in PBWin10..
    The PRINTs and WAITKEYs are there to show how the demonstration is progressing and they'd all be removed in any real application.

    To run it in PBWin, replace them with MSBBOX or TXT.WINDOW/TXT.PRINT/TXT.WAITKEY

    Code:
    'PBWin10 program
    'This code attempts to handle exceptions not normally trapped by PowerBASIC.
    'It consists of 3 routines.
    '1) PBMain just a short routine to show how to use the other 2 functions
    '2) SetupExceptionHandler which installs or removes the exception handler
    '3) ExceptionHandler  which is called, via the operating system, when an exception occurs.
    '
    'Floating point exceptions need to be explicitly allowed by clearing one or more of 6 mask bits in the FPU control word.
    'Other exceptions happen anyway but this program shows how to intercept them to do something useful before the default
    '  exception handler of the OS gets them and reports a the error.
    '
    ' It is possible, though sometimes difficult, to resume execution of the program from after the point of the exception.
    '  This code shows how to resume at the next address after a PrivelidgedInstruction exception.
    '  It also shows how to resume at another address such as an error handler. This should be in the same SUB/FUNCTION as the error.
    '
    'The default in this program is to enable all FP exceptions apart from INEXACT_RESULT as this will trip up most people
    '  more than it helps them! e.g. 1/3 will trigger the exception as the result cannot be exactly represented in binary.
    '  If you want to trap INEXACT_RESULT then set %FPTrapInexactResultException = 1   below.
    '
    'The only ASM here is to set the FPU control register bits so don't be put off by it.
    '
    'Three distict outcomes are shown in this code.
    ' If there is a priveliged instruction exception caused by CLI or STI (clear or set interrupt flag) then the inxtruction is ignored
    '  and execution continues at the next address as if the instruction wasn't there. Other priviliged instructions are passed to the
    '  default handler (usually the OS).
    '
    'If there is a floating point exception then it is reported along with a dump of the FP registers, it's then passed to the default handler.
    '
    'If there is an integer divide by zero then the error is reported and execution continues at the error handler specified in the call
    '   when the Exception Handler was installed.
    '
    'There are lots of potential problems here. This code is just posted to demonstrate how exceptions can be dealt with.
    'In particular, returning execution to the ResumeAddress will only work if the exception occurs in the same function as the ResumeAddress
    ' but calls to outside functions in maths routines are very common so care needs to be taken.
    '
    'Just for demonstration purposes I print things out in the ExceptionHandler to inform the user what has happened. Normally you wouldn't
    '  do this if you are returning execution to the program.
    
    
    #COMPILE EXE
    #DIM ALL
    #REGISTER NONE
    
    #INCLUDE "win32api.inc"
    
    
    %FPTrapInexactResultException = 0    'set to non zero if you want to trap FPInexactResults (specialised use only)
    %enable = 1
    %disable= 0
    
    %CLI= &hfa     'opcode of STI instruction    'these are just to test the PRIVILEGED_INSTRUCTION exception
    %STI= &hfb     'opcode of CLI instruction
    
    
    GLOBAL ExeptionHandlerResumeAddress AS DWORD
    
    
    
    FUNCTION SetupExceptionHandler(EnableTrap AS LONG,ResumeAddress AS DWORD) AS LONG
    'EnableTrap = 0 for uninstall handler, not 0 for install handler
    'ResumeAddress = address WITHIN THE ERRING FUNCTION/SUB to which you would like the code to return after the exception is handled
    '              =0 for don't resume.
    
    STATIC hHandler AS LONG
    
    LOCAL HandlerPointer AS LONG
    LOCAL NewControlWord, mask AS INTEGER    'control word is a 16 bit register
    LOCAL OriginalControlWord AS INTEGER
    
    ExeptionHandlerResumeAddress = ResumeAddress
    
        IF EnableTrap THEN
            'install the exception handler
            HandlerPointer = CODEPTR(ExceptionHandler)
            hHandler = AddVectoredExceptionHandler( 1  , HandlerPointer )
    
            IF hHandler = 0 THEN
               'Failed to add handler
                FUNCTION = 0
            ELSE
                'Handler installed
                FUNCTION = 1
    
                'now set the FPU control register to enable exceptions on all conditions
                !fstcw OriginalControlWord  'save the original control word
                !mov ax,OriginalControlWord 'get the control word
    
                #IF %FPTrapInexactResultException = 0
                !and ax,&hffe0              'zero the exception mask bits except for InexactResult
                #ELSE
                !and ax,&hffc0              'zero the exception mask bits (lower 6 bits or control word) to allow all 6 exceptions
                #ENDIF
    
                !mov NewControlWord,ax      'save the new control word
                !fldcw NewControlWord       'set the FPU to use the new control word
    
            END IF
    
        ELSE
            'remove the exception handler
            hHandler = RemoveVectoredExceptionHandler(hHandler)
            IF hHandler = 0 THEN
                'Couldn't remove exception handler
                FUNCTION = 0
    
            ELSE
                'Exception handler removed
                FUNCTION = 1
                !fldcw OriginalControlWord  'restore the original control word
    
            END IF
    
        END IF
    
    END FUNCTION
    
    
    
    FUNCTION ExceptionHandler(BYVAL pException_Pointers AS Exception_Pointers PTR) AS LONG
    
    LOCAL ThisExceptionPointer AS Exception_Pointers
    LOCAL ThisExceptionRecordPointer AS Exception_Record PTR
    LOCAL ThisExceptionRecord AS Exception_Record
    LOCAL temp AS BYTE PTR
    LOCAL r AS LONG
    LOCAL pt AS DWORD
    LOCAL FPControlWord AS INTEGER
    LOCAL FPControlWordNew AS INTEGER
    LOCAL WhatToDoNext  AS LONG
    
    
    ThisExceptionPointer = @pException_Pointers
    ThisExceptionRecordPointer = ThisExceptionPointer.ExceptionRecord
    ThisExceptionRecord = @ThisExceptionRecordPointer
    
    WhatToDoNext=%EXCEPTION_CONTINUE_SEARCH 'set the default for all exceptions I don't handle here.
                                            'the 2 options are:
                                            '%EXCEPTION_CONTINUE_SEARCH which passes the exception on to the next handler in line (the OS?)
                                            '%EXCEPTION_CONTINUE_EXECUTION which resumes execution of the code
                                            'If execution is resumed, the address to resume at can be set within the handler
    
    !fnstcw FPControlWord   'get FP control word
    !fnclex                 'clear all FP exeptions. Got to do this as following lines use the FPU and may cause a recurring exception
    !mov ax,FPControlWord
    !or ax,&h3f             'mask all exception while in the exception routine
    !mov FPControlWord,ax
    !fldcw FPControlWord    'load new FP control word
    
    'now handle each possible exception
    
    SELECT CASE ThisExceptionRecord.ExceptionCode
        CASE  BITS(DWORD,%STATUS_FLOAT_OVERFLOW)
    
            TXT.PRINT "There has been a FP overflow at address &h";HEX$(ThisExceptionRecord.ExceptionAddress)
            TXT.PRINT "here are the FPU details at the time of the exception"
    
            TXT.PRINT "ControlWord   ";HEX$(@[email protected])
            TXT.PRINT "StatusWord    ";HEX$(@[email protected])
            TXT.PRINT "TagWord       ";HEX$(@[email protected])
            TXT.PRINT "ErrorOffset   ";HEX$(@[email protected])
            TXT.PRINT "ErrorSelector ";HEX$(@[email protected])
            TXT.PRINT "DataOffset    ";HEX$(@[email protected])
            TXT.PRINT "DataSelector  ";HEX$(@[email protected])
    
            pt=VARPTR(@[email protected](0))
    
            REDIM reg(0 TO 7) AS EXT AT pt
            FOR r& = 0 TO 7
                TXT.PRINT "ST(";r&;")=";reg(r&)
            NEXT
    
            TXT.PRINT
            TXT.PRINT "Press a key to pass error to default exception handler"
            TXT.WAITKEY$
    
    
        CASE BITS(DWORD,%STATUS_FLOAT_UNDERFLOW)
            TXT.PRINT "There has been a FP underflow at address &h";HEX$(ThisExceptionRecord.ExceptionAddress)
            TXT.WAITKEY$
    
        CASE BITS(DWORD,%STATUS_FLOAT_INVALID_OPERATION)
            TXT.PRINT "There has been an Invalid FP Operation at address &h";HEX$(ThisExceptionRecord.ExceptionAddress)
            TXT.WAITKEY$
    
        CASE BITS(DWORD,%STATUS_FLOAT_INEXACT_RESULT)
    
    
        CASE BITS(DWORD,%STATUS_FLOAT_DIVIDE_BY_ZERO)
    
    
        CASE BITS(DWORD,%STATUS_FLOAT_DENORMAL_OPERAND)
    
    
        CASE BITS(DWORD,%STATUS_ILLEGAL_INSTRUCTION)
    
    
        CASE BITS(DWORD,%STATUS_ACCESS_VIOLATION)
    
    
        CASE BITS(DWORD,%STATUS_PRIVILEGED_INSTRUCTION)
            TXT.PRINT "STATUS_PRIVILEGED_INSTRUCTION exception has occurred"
    
            temp = @[email protected]    'get the EIP register to see where the instruction was that caused the exception
    
            IF @temp = %STI OR  @temp = %CLI THEN   '@temp = the instruction itself
                'if its set/clear interrupt flag I'll just increment the return address pointer to ignore the exception
                INCR @[email protected]    'return execution to the byte after the one that caused the exception
    
                TXT.PRINT "You can't mess with the interrupt flag. I'll ignore it."
                TXT.PRINT "Execution will now continue"
                TXT.PRINT
                WhatToDoNext=%EXCEPTION_CONTINUE_EXECUTION
            ELSE
                'it's not an interrupt flag instruction so pass it to the default handler
                TXT.PRINT "You aren't allowed to use that instruction at address &h";HEX$(ThisExceptionRecord.ExceptionAddress)
                TXT.PRINT "I'm going to crash now."
    
                WhatToDoNext=%EXCEPTION_CONTINUE_SEARCH
            END IF
    
    
    
        CASE BITS(DWORD,%STATUS_STACK_OVERFLOW)
    
    
        CASE BITS(DWORD,%STATUS_FLOAT_MULTIPLE_FAULTS)
    
    
        CASE BITS(DWORD,%STATUS_INTEGER_OVERFLOW)
    
    
        CASE BITS(DWORD,%STATUS_INTEGER_DIVIDE_BY_ZERO)
            TXT.PRINT "Integer divide by zero at address &h";HEX$(ThisExceptionRecord.ExceptionAddress)
            TXT.PRINT
            'as a test we'll try to return to the faulty function's error trap
            IF ExeptionHandlerResumeAddress <>0 THEN
                @[email protected] = ExeptionHandlerResumeAddress
            END IF
    
            WhatToDoNext=%EXCEPTION_CONTINUE_EXECUTION
    
    END SELECT
    
    
    IF WhatToDoNext =%EXCEPTION_CONTINUE_SEARCH THEN
        FUNCTION = %EXCEPTION_CONTINUE_SEARCH
    ELSE
        FUNCTION = %EXCEPTION_CONTINUE_EXECUTION
        !fldcw FPControlWord   'restore the FPUControl word if I'm going to return to executing my code
    
    END IF
    
    
    END FUNCTION
    
    
    
    FUNCTION PBMAIN () AS LONG
    
    LOCAL sum AS SINGLE
    LOCAL ResumeAddress AS LONG
    
    LOCAL hTxtWin AS LONG
    TXT.WINDOW("Exception handler tests",1,1,40,80) TO hTxtWin
    
    ResumeAddress = CODEPTR(ErrorTrap)
    
    
    IF SetupExceptionHandler(%enable,ResumeAddress) = 0 THEN
        TXT.PRINT "Failed to add exception handler"
    ELSE
        TXT.PRINT "Exception Handler installed "
        TXT.PRINT "Press a key to cause a Privileged Instruction exception"
        TXT.PRINT
        TXT.WAITKEY$
    
    '############################################
    '# cause a Privileged Instruction exception
    '############################################
    
    !cli
    
    
        TXT.PRINT  "Press a key to cause an Integer Divide by zero exception"
        TXT.PRINT
        TXT.WAITKEY$
    
    '############################################
    '# cause an integer divide by zero exception
    '############################################
    LOCAL a,b,c AS LONG
    
    a=0
    b=1
    c=b\a
    
        'in this example the code doesn't reach this point but in real code you should uninstall the exception handler
        ''like this when you're finished with it.
        IF SetupExceptionHandler(%disable,0)= 0 THEN
            TXT.PRINT  "Couldn't remove exception handler"
        ELSE
            TXT.PRINT  "Exception handler removed"
        END IF
    
    
    END IF
    GOTO xit
    
    ErrorTrap:
    TXT.PRINT  "I've resumed execution at the error trap."
    TXT.PRINT  "press a key to cause a fatal FP overflow"
    TXT.PRINT
    TXT.WAITKEY$
    
    '############################################
    '# cause a FP overflow exception
    '############################################
    
    
        sum = 1
        DO
            sum = sum + sum        'cause an overflow exception by counting too high
        LOOP
    
    
    xit:
    TXT.WAITKEY$
    END FUNCTION

    Leave a comment:


  • Michael Mattias
    replied
    Has anyone taken the time to convert this to PBWin 10?
    You mean the code linked in post #1? Follow the link... post #4 there as well as post #7 here both say code works fine with PB/C 6, which is the same "guts" as PB/Win 10.

    And see also my post #9 here .. which notes problems are not due to compiler versions, they are due to differences in header files.

    Leave a comment:


  • Mike Gorden
    replied
    Has anyone taken the time to convert this to PBWin 10?

    Leave a comment:


  • Chris Holbrook
    replied
    Nice work Paul.

    By knowing that these errors are reported in apps using your enhancement, we can now make assertions about quality which may even open new doors for those of us who sell PB based products.

    Leave a comment:


  • Michael Mattias
    replied
    >Version for CC 4.04 works with CC 5.05 and with CC 3.04 b

    As Paul points out, "Works/doesn't work with compiler version ..." is not accurate! It's NOT the "compiler version" which is the source of any difficulties!

    His code should work with ANY version of compiler, provided you use the same Windows' headers files he did.

    Compile-time errors on "different headers" is really a Good Thing. Where you really get into trouble with "different headers" is when there are subtle differences which allow the program to complile cleanly but execute differently!

    MCM

    Leave a comment:


  • Arthur Gomide
    replied
    > I haven't tested it with other versions.

    Version for CC 4.04 works with CC 5.05 and with CC 3.04 but with 2 minor adjustments in the declaration of local variables.

    Leave a comment:


  • Paul Dixon
    replied
    The original program was written for PBCC4.
    Changes to the later versions of the win32api.inc files prevented the original from compiling with later versions of the compiler.
    I've fixed the code to run with later compilers by extracting all the required definitions from the PBCC4 .inc files and including them directly in the program.

    It now appears to work OK in PBCC6. I haven't tested it with other versions.

    Paul.

    Leave a comment:


  • Arthur Gomide
    replied
    Paul,

    I was really impressed with the ease of implementation and objectivity of your example. No doubt about the usefulness. It would be interesting if you could adapt this example for new compilers and so more people can see for themselves all that you demonstrated so well here.

    Thank you for this good explanation!

    Arthur.

    Leave a comment:


  • Mike Doty
    replied
    Great code, even traps stack overflow.

    Leave a comment:


  • Steve Bouffe
    replied
    Thanks. Installed now.

    Leave a comment:


  • Paul Dixon
    replied
    Steve,
    you must have removed the #REGISTER NONE line from the code.
    If you want to use registers in your code then add #REGISTER NONE as the first line in FUNCTION SetupExceptionHandler()

    Paul.

    Leave a comment:


  • Steve Bouffe
    replied
    I'm trying to use the code with PB Win

    Getting a compile error 540 here

    !fldcw NewControlWord 'set the FPU to use the new control word

    "Float opcode with a register variable"

    Leave a comment:


  • Paul Dixon
    started a topic Exception handlers

    Exception handlers

    Just posted in the source code forum, an example of installing an exception handler in Windows.
    Windows has the facility to do this but the calls aren't widely known.
    For more information seach the internet for "Structured exception handling".



    Paul.
Working...
X
😀
🥰
🤢
😎
😡
👍
👎