Dear fellow programming toilers,
is the following routine correct ?
I've tested it extensively, and it appears to work.
----- Cut
FUNCTION IsGoodReadPtr(CONST P: pointer): boolean;
{ -- Checks to see if P is readable.
  -- Under DPMI, this means: can we obtain the value of P^ without causing a
  --             GPF (Exception 0Dh) ?
  -- In real mode, any pointer  NIL is considered readable. }
{$IFDEF MSDOS}
BEGIN IsGoodReadPtr:=P  NIL END;
{$ENDIF}
{$IFDEF DPMI}
ASSEMBLER;
ASM xor cl, cl              { -- Function result := FALSE,
                              -- for the time being.                    }
{ -- First we test the selector part of P:                              }
    mov bx, word ptr P+2    { -- Seg(P^) into BX.                       }
    or bx, bx               { -- BX = 0, i.e. was P of the form 0:ofs ? }
    je @@EXIT               { -- YES: not valid, so exit.               }
    { -- The above two lines were added because older versions of VERR  }
      -- and LSL return nonsense when presented with a selector equal   }
      -- to zero.                                                       }
    verr bx                 { -- Readable ?                             }
    jnz @@EXIT              { -- NO : exit, return FALSE.               }
                            { -- YES: on to the next test.              }
{ -- Now the offset part of P comes under scrutiny:
  -- is offset <= the selector's limit ?                                }
                            { -- BX still holds the selector.           }
    lsl ax, bx              { -- AX := limit.                           }
    cmp ax, word ptr P      { -- Limit >= Ofs(P^) ?                     }
    jb @@EXIT               { -- NO : P is invalid - exit.              }
    inc cl                  { -- YES: return TRUE.                      }
@@EXIT:
    mov al, cl              { -- Function result into AL.               }
END;
{$ENDIF}
----- Cut
Some required information for those not familiar with Borland Pascal:
- Anything between "{" and the next "}" is a comment.
- Pointers are 32 bit animals, 16 bit offset, 16 bit segment, stored in
  offset-segment order.
- Boolean function results are expected to be in AL.
- in C speak, "" == "!=", NIL == NULL, and "P^" == "*P".
According to the literature, LSL may fail. Can it also fail when VERR has
already given the selector a clean bill of health ?
Should the above routine be correct:
is a there a better/faster way (perhaps using DPMI functions 0202h and 0203h) 
?
Should it not:
kindly show me the error of my ways.
On a related note: has anyone experienced problems with VERR and/or LSL ?
According to the Interrupt List, both instructions contain some truly bizarre
bugs.
Peter
... Alternate Subspace Molecular Inversion Refresh Layer:
... Valid settings are: Enabled and Disabled    
--- EBO-BBS Diemen - NL
---------------