What is the correct prototype of md_atof?

What is the correct prototype of md_atof?

[Kazu Hirata]

Hi,

gas/tc.h has

  char *md_atof PARAMS ((int what_statement_type, char *literalP, int *sizeP));

but so many gas/config/tc-*.c has

  char *
  md_atof (type, litP, sizeP)
       int type; <- Notice it's int!!
       char *litP;
       int *sizeP;

I noticed this problem while converting tc-h8300.c into ISO-C.  To me
"char" seems to be correct here.  What do others think?

Kazu Hirata

Re: What is the correct prototype of md_atof?

[Ian Lance Taylor]

Kazu Hirata <kazu@cs.umass.edu> writes:

> gas/tc.h has
> 
>   char *md_atof PARAMS ((int what_statement_type, char *literalP, int *sizeP));
> 
> but so many gas/config/tc-*.c has
> 
>   char *
>   md_atof (type, litP, sizeP)
>        int type; <- Notice it's int!!
>        char *litP;
>        int *sizeP;
> 
> I noticed this problem while converting tc-h8300.c into ISO-C.  To me
> "char" seems to be correct here.  What do others think?

Looks like `int' in both places, so what's the problem?

The code uses `int' because in traditional C, with no prototypes, a
char is always promoted to int anyhow.  If you write `char' in a
prototype, but then don't use a prototype when defining the function,
the 'char' in the function definition will be promoted to `int'.
Since the function definition then does not match the prototype, there
is a potential problem.

If we convert to C90 and use a prototype style definition, then we can
safely switch to `char'.

Ian

Re: What is the correct prototype of md_atof?

[Kazu Hirata]

Hi Ian,

> > gas/tc.h has
> > 
> >   char *md_atof PARAMS ((int what_statement_type, char *literalP, int *sizeP));
> > 
> > but so many gas/config/tc-*.c has
> > 
> >   char *
> >   md_atof (type, litP, sizeP)
> >        int type; <- Notice it's int!!
> >        char *litP;
> >        int *sizeP;
> > 
> > I noticed this problem while converting tc-h8300.c into ISO-C.  To me
> > "char" seems to be correct here.  What do others think?
> 
> Looks like `int' in both places, so what's the problem?

Oops.  I meant to say that so many gas/config/tc-*.c has char for the
first argument of md_atof.

> The code uses `int' because in traditional C, with no prototypes, a
> char is always promoted to int anyhow.  If you write `char' in a
> prototype, but then don't use a prototype when defining the function,
> the 'char' in the function definition will be promoted to `int'.
> Since the function definition then does not match the prototype, there
> is a potential problem.
> 
> If we convert to C90 and use a prototype style definition, then we can
> safely switch to `char'.

The only reason why I said "char" is correct is that the first
argument contains a character.  "int" wouldn't be wrong, but I am not
comfortable with the mismatch between the definition and the
prototype.  Should I change every md_atof to use char for the first
argument?

Kazu Hirata

Re: What is the correct prototype of md_atof?

[Ian Lance Taylor]

Kazu Hirata <kazu@cs.umass.edu> writes:

> > If we convert to C90 and use a prototype style definition, then we can
> > safely switch to `char'.
> 
> The only reason why I said "char" is correct is that the first
> argument contains a character.  "int" wouldn't be wrong, but I am not
> comfortable with the mismatch between the definition and the
> prototype.  Should I change every md_atof to use char for the first
> argument?

Yes, provided every md_atof uses a prototype style definition.

Ian