code checker crap

Hi,

We use a code checker at work which complains about the following code:

Code:

---

short Rank;
....
Rank = SHRT_MIN ;

---

Code:

---

// SHRT_MIN is defined as
#define SHRT_MIN        (-32767-1)

---

The complaint is
"Values shall not be implicitly converted to a narrower type.
(Category: CONVERSIONS, Level: 1)
Value implicitly converted to narrower type 'short'"

It's even a level 1 meaning that is a nasty bug according to the tool :)

The tool says that because the expression (-32767-1) yields an int as type and that int is assigned to a short. So the fix to shut the tool up is

Code:

---

Rank = (short)SHRT_MIN;

---

and management seems to think this improves code quality. What do you people think? The cast makes better quality code or not?

Isn't SHRT_MIN already defined in <limits.h>?

I certainly don't think the cast improves the quality at all.

Neither does gcc warn about any conversion when I do this:

Code:

---

#include <iostream>

int main()
{
  short a = -32767-1;
  std::cout << a << std::endl;
}

---

--
Mats

Quote:

---

Originally Posted by laserlight

Isn't SHRT_MIN already defined in <limits.h>?

---

yeah, its the one from the standard library. The code checker only checks code after it is preprocessed so it never sees SHRT_MIN but only (-32767-1)

I get a warning that SHRT_MIN is being redefined. Why not this (without a redefine):

#define SHRTMIN (short) (-32767-1)

Todd

Quote:

---

yeah, its the one from the standard library. The code checker only checks code after it is preprocessed so it never sees SHRT_MIN but only (-32767-1)

---

Oh, I see. Has it been pointed out to management that this is a problem with the code checker, since it purports to correct the standard library implementation? I too do not think that a cast is useful, but if it should be present, then the change should be made to <limits.h>.

Quote:

---

Originally Posted by laserlight

Oh, I see. Has it been pointed out to management that this is a problem with the code checker, since it purports to correct the standard library implementation? I too do not think that a cast is useful, but if it should be present, then the change should be made to <limits.h>.

---

But there is no conflict between -32768 (or -32767-1) and short (assuming of course that short is 16 bits, and not some other size).

--
Mats

Quote:

---

But there is no conflict between -32768 (or -32767-1) and short (assuming of course that short is 16 bits, and not some other size).

---

Yes, so what is your point?

Quote:

---

Originally Posted by laserlight

Yes, so what is your point?

---

Sorry, my point is that the code-checker is wrong, and not the original code. Possibly the difference between > and >=? (or < and <=).

--
Mats

Quote:

---

Sorry, my point is that the code-checker is wrong, and not the original code.

---

That is my point as well ;)

Quote:

---

Possibly the difference between > and >=? (or < and <=).

---

I do not think that the problem is so much of incorrect interpretation of the requirements for the standard minimum implementation limits, but that the code checker cannot recognise that the result is a constant that will fit into the range of the destination type, so there surely cannot be a potential bug lurking there, much less a nasty bug.

Quote:

---

Originally Posted by laserlight

That is my point as well ;)


I do not think that the problem is so much of incorrect interpretation of the requirements for the standard minimum implementation limits, but that the code checker cannot recognise that the result is a constant that will fit into the range of the destination type, so there surely cannot be a potential bug lurking there, much less a nasty bug.

---

That is, of course another interpretation.

Does it also complain if you assign it -32768, or -1?

--
Mats

Quote:

---

Originally Posted by matsp

That is, of course another interpretation.

Does it also complain if you assign it -32768, or -1?

--
Mats

---

yes it does because the checker sees any numerical constant as int type and assigning int to short without a cast is bad according to the checker. I agree that short x = int y; is not good code but with constants its a whole other story.