Prev: Extable download
Next: Sudoku Player
From: Han on 26 Jun 2010 00:01
Within CK&DIPATCH1 is a jump to #20E34h in the case where an integer
is allowed as an alternate argument type for real numbers. The code
jumps to #20E34h when, upon converting the integer to a real number,
an overflow or underflow occurs. The code is as follows:

L20E34
sstdsp_overflow D0=(5) (=SystemFlags)+5
A=DAT0 B
?ABIT=1 0
GOYES +
GOSBVL =GETPTR
LC(5) =ofloerr
GOVLNG =GPErrjmpC
+ LC(5) =%MAXREAL
?D=0 S
GOYES +
LC(5) =%-MAXREAL
+ D0=(5) (=SystemFlags)+6
A=DAT0 B
ABIT=1 0
DAT0=A B
GOTO sstdsp_push%

There is a check in the system flags whether an error is generated or
+/-MAXREAL constants are pushed. The first bug is that the routine
checks if the overflow flag is set, and if so, it pushes the
constants. It should instead check if the flag is cleared!

Secondly, the routine sstdsp_push% overwrites the pointer to the
original integer with the pointer to the real (in TEMPOB), and takes
its input from A[A]. The problem is that A[A] is used to check the
system flags, and the actual address to push is in C[A]. So the
routine above is missing an opcode: A=C A
From: Han on 26 Jun 2010 00:34
Additionally, here's how to test:

TESTPRG
::
CK&DISPATCH1
BINT1 %1
;

Make sure to turn on exact mode. Also check that flag -21 is clear
(overflow returns +/-MAXREAL). Create an integer larger than 1E499 by
typing:

1E499 R->I 1000 *

With this integer on the stack, run TESTPRG and you will get an
overlflow error. This should not happen with flag -21 cleared! We are
supposed to see %MAXREAL on the stack!

Now, set flag -21 (overflow results in an error) and run TESTPRG. You
will instead get a bad argument error (instead of an overflow error).
Moreover, the integer is now turned into PTR 505 (on my calculator,
anyway).
From: HM on 26 Jun 2010 12:05
Did the test on a 50g with ROM 2.09.
Got exactly the same results as you reported.
 | 
Pages: 1
Prev: Extable download
Next: Sudoku Player