-Wconversion producing incorrect warning

-Wconversion producing incorrect warning

[Kevin P. Fleming]

When compiling this program:

#include <sys/types.h>
#include <inttypes.h>

typedef uint16_t t;

int f(t x)
{
	return x*x;
}

int main(int argc, char *argv[])
{
	t b = 12;

	return f(b);
}

using GCC 3.3.5, 3.4.3 and 4.0.1 with '-Wconversion' in effect, I get 
this warning:

luigi.c: In function `main':
luigi.c:15: warning: passing arg 1 of `f' with different width due to 
prototype

I don't understand why this is true... the prototype and the argument 
are of the same type.

Re: -Wconversion producing incorrect warning

[John Love-Jensen]

Hi Kevin,

The ABI of your platform probably converts the uint16_t to an unsigned int
on the stack.

HTH,
--Eljay

Re: -Wconversion producing incorrect warning

[Kevin P. Fleming]

John Love-Jensen wrote:

> The ABI of your platform probably converts the uint16_t to an unsigned int
> on the stack.

That's what I suspected, as the person who provided that code reported 
that's what was causing the warning.

What concerns me is that the warning message states this is happening 
because of the prototype, but that's not technically true... on a 
platform that actually chose to pass the 16-bit quantity on the stack 
the warning would not occur, so this is a platform-specific problem, not 
anything wrong with the code itself.

Re: -Wconversion producing incorrect warning

[John Love-Jensen]

Hi Kevin,

> What concerns me is that the warning message states this is happening
> because of the prototype, but that's not technically true... on a
> platform that actually chose to pass the 16-bit quantity on the stack
> the warning would not occur, so this is a platform-specific problem, not
> anything wrong with the code itself.

Correct.  On that platform, the warning is occurring because of the
platform's ABI.

The warning does not indicate that there is anything wrong with the code
itself.  The warning is justified, warranted, and desirable if/when/where
-Wconversion is specified, since the warning tells you that something funny
is happening that may not be apparent from the code itself, on that
particular platform.

Sincerely,
--Eljay

Re: -Wconversion producing incorrect warning

[Kevin P. Fleming]

John Love-Jensen wrote:

> The warning does not indicate that there is anything wrong with the code
> itself.  The warning is justified, warranted, and desirable if/when/where
> -Wconversion is specified, since the warning tells you that something funny
> is happening that may not be apparent from the code itself, on that
> particular platform.

We've worked around it for now by just passing the arguments 
by-reference instead. In this application it makes very little 
difference, and may in fact be slightly faster.

Re: -Wconversion producing incorrect warning

[Segher Boessenkool]

> using GCC 3.3.5, 3.4.3 and 4.0.1 with '-Wconversion' in effect, I get 
> this warning:
>
> luigi.c: In function `main':
> luigi.c:15: warning: passing arg 1 of `f' with different width due to 
> prototype
>
> I don't understand why this is true... the prototype and the argument 
> are of the same type.

Yes, so the parameter will be passed as whatever your ABI
specifies for parameters of unsigned 16-bit type.

The problem here is, that -Wconversion does _not_ warn
about type conversions in your code!  It is a warning for
people trying to port old K&R C code to ANSI C.  What the
warning means is, "if there would not have been a prototype
for f() in scope, the parameter would have been passed as
a 32-bit type".

GCC does not currently have a warning with the behaviour
you might have expected, and the name of "-Wconversion"
confuses a lot of people.  Perhaps we should change its
name...


Segher

Re: -Wconversion producing incorrect warning

[John Love-Jensen]

Hi Segher,

> The problem here is, that -Wconversion does _not_ warn about type conversions
in your code!

From: http://gcc.gnu.org/onlinedocs/gcc-4.0.2/gcc/Warning-Options.html

-------------------------------------------------------------------
-Wconversion

Warn if a prototype causes a type conversion that is different from what
would happen to the same argument in the absence of a prototype. This
includes conversions of fixed point to floating and vice versa, and
conversions changing the width or signedness of a fixed point argument
except when the same as the default promotion.

Also, warn if a negative integer constant expression is implicitly converted
to an unsigned type. For example, warn about the assignment x = -1 if x is
unsigned. But do not warn about explicit casts like (unsigned) -1.
-------------------------------------------------------------------

What kind of type conversions in your code are you referring to?

--Eljay

Re: -Wconversion producing incorrect warning

[Segher Boessenkool]

>> The problem here is, that -Wconversion does _not_ warn about type 
>> conversions
> in your code!

[snip]

> What kind of type conversions in your code are you referring to?

Not my code.  The type conversions I am referring to, as does
the documentation piece I snipped, is conversion of arguments
to functions specifically, and any type conversions generally.

Hrm, I think I don't understand your question :-)


Segher