Page 1 of 2
MAX with two arguments
Posted: Sat Aug 22, 2020 1:40 am
by Ivan Denisov
Oleg found
Code: Select all
MODULE TestCP;
CONST
max* = MAX(INTEGER, 0);
END TestCP.
This invalid code gives not error.
Re: MAX with two arguments
Posted: Sat Aug 22, 2020 8:27 am
by adimetrius
Great finding, Ivan!
Looks like a small omission in DevCPB.StFct, I think it can be fixed as follows:
Code: Select all
ELSIF (fctno = minfn) OR (fctno = maxfn) THEN
IF (parno < 1) OR (parno = 1) & (p.hint # 1) THEN err(65)
(* add this branch *) ELSIF (p.hint = 1) & (parno > 1) THEN err(64)
END;
p.hint is filled out in StPar0 when analyzing the first actual parameter to MAX/MIN, and .hint = 1 indicates the 1st actual parameter was a type. parno is the total number of actual parameters passed to the predeclared procedure. 65 is 'not enough parameters', 64 is 'too many parameters'.
The fix also covers MAX(REAL, ?), etc., as well as MIN.
Re: MAX with two arguments
Posted: Sun Aug 23, 2020 2:09 am
by Zorko
adimetrius, I tested your fix and found it perfectly. Thank you!
In descritpion of CPlang of MIN and MAX with two arguments:
Code: Select all
MAX(x, y) <= INTEGER INTEGER the larger value of x and y
integer type LONGINT
<= SHORTREAL SHORTREAL
numeric type REAL
SHORTCHAR SHORTCHAR
character type CHAR
MIN(x, y) <= INTEGER INTEGER the smaller of x and y
integer type LONGINT
<= SHORTREAL SHORTREAL
numeric type REAL
SHORTCHAR SHORTCHAR
character type CHAR
As you can see, the documentation describes support for the characters. However I can't compile such code with definition MIN/MAX(characters) as a constant:
Code: Select all
CONST
max* = MAX("s", "a"); <-- operand inapplicable to (this) function
I'm sure it's a small omission, too.
In expression it compiles:
Code: Select all
VAR c1, c2: CHAR;
BEGIN
c1 := MIN(c1, c2);
I do propose this fix:
CPB.ConstOp
Code: Select all
| min:
- IF f IN intSet + realSet THEN
+ IF f IN intSet + realSet + charSet THEN
| max:
- IF f IN intSet + realSet THEN
+ IF f IN intSet + realSet + charSet THEN
Re: MAX with two arguments
Posted: Mon Aug 24, 2020 8:29 am
by Zinn
Sorry Ivan, that is not a mistake. In the Language Report capitel 10.3 Predeclared Procedures there define
MAX(T) T = basic type T maximum value of type T
T = SET INTEGER maximum element of a set
MAX(x, y) <= INTEGER INTEGER the larger value of x and y
integer type LONGINT
<= SHORTREAL SHORTREAL
numeric type REAL
SHORTCHAR SHORTCHAR
character type CHAR
So there is no need to change the Compiler. It works as it is defined.
- Helmut
Re: MAX with two arguments
Posted: Mon Aug 24, 2020 9:20 am
by adimetrius
Helmut,
will all due respect, this is not how I read the Language report capitel 10.3.
Also, for MAX(a, b) to be defined, '<' (less) has to be defined for a and b also (precisely speaking, '<' has to be defined for pairs a, b and b, a), because
IF a <= b THEN MAX(a, b) = a
ELSE MAX(a, b) = b
Can you say that INTEGER < 0? Or is it rather that INTEGER > 0? Maybe INTEGER = 0? Clearly, '<' is not defined for INTEGER, 0.
Thus, mathematically speaking, MAX(INTEGER, 0) does not make sense. Though MAX(MAX(INTEGER), 0) makes perfect sense.
Also, I've tried to search for 'MAX(INTEGER,' accross BlackBox sources, with no result.
All this to say, I don't think MAX(INTEGER, 0) was meant to be valid by the language authors.
Re: MAX with two arguments
Posted: Tue Aug 25, 2020 9:36 am
by Robert
Permitting the expression
looks like an error to me, and should be treated as a compiler bug.
I have not followed this discussion very closely, so have two questions:
1 - Do the proposed changes to DevCPB fully resolve this issue?
2 - Is the problem with characters the same issue, or a separate bug?
Re: MAX with two arguments
Posted: Tue Aug 25, 2020 11:05 pm
by cfbsoftware
Do the same issues apply to the MIN function? If they do then any possible rework to MAX should be considered for MIN as well. If they don't then it is likely that similar code patterns for MIN could be used.
There are a significant number of variations of valid / invalid parameter combinations for these two functions. The risk of introducing new problems with any modifications must be considered and extensive test cases developed to ensure that this has not happened. The original problem reported only occurs when attempts are made to compile badly written code that is unlikely to occur in real life. I would not want to see new problems introduced that affect correctly written code.
Re: MAX with two arguments
Posted: Wed Aug 26, 2020 10:56 am
by luowy
in my opinion,it is a bug, the compiler needs to be fixed;
accoding language report 10.3 as Zinn point out
Code: Select all
The following table lists the predeclared procedures. Some are generic procedures, i.e., they apply to several types of operands. v stands for a variable, x and y for expressions, and T for a type. The first matching line gives the correct result type.
the MAX(T) and MAX(x, y) is clearly explained that the two parameters are expression, not a type;
this issue apply to the MIN funtion also;
the bug at DevCPB.StPar1:
Code: Select all
| minfn: (*MIN*)
IF p.hint # 1 (*p.class # Ntype*) THEN Op(min, p, x) ELSE err(64) END
| maxfn: (*MAX*)
IF p.hint # 1 (*p.class # Ntype*) THEN Op(max, p, x) ELSE err(64) END
when the first parameter is a type, the "p.class" will be "Nconst" instead of "Ntype", so use "p.hint = 1" to check it ;
a simple test
Code: Select all
MODULE ObxMAX;
IMPORT Log:=StdLog;
PROCEDURE Do();
VAR x,y:INTEGER;b:BOOLEAN;
BEGIN
(* invalid*)
x := MAX(INTEGER, 0);
x := MAX(INTEGER, y);
x := MAX(0, INTEGER);
x := MAX(y, INTEGER);
x := MIN(INTEGER, 0);
x := MIN(INTEGER, y);
x := MIN(0, INTEGER);
x := MIN(y, INTEGER);
(*valid*)
x := MAX(INTEGER);
x := MAX(x, 0);
x := MAX(x, y);
x := MIN(INTEGER);
x := MIN(x, 0);
x := MIN(x, y);
END Do;
END ObxMAX.
Re: MAX with two arguments
Posted: Wed Aug 26, 2020 4:54 pm
by adimetrius
I like luowy's suggested fix better than my own, although they achieve the same result.
What about Zorko's post? He pointed out a situation when MAX/MIN is applied to two constants, which is fully evaluated at compile time, and that's done in a different procedure.
Re: MAX with two arguments
Posted: Thu Sep 03, 2020 7:10 am
by Josef Templ
There is yet another little bug in the docu of MIN/MAX(T), I think:
Code: Select all
MAX(T) T = basic type T maximum value of type T
T = SET INTEGER maximum element of a set
The order is wrong.
SET must be first according to the rule that the first matching line defines the result because SET is a basic type.
- Josef