Re: Correct arguments to isspace/isalnum

[Ralf Wildenhues]

* Bob Friesenhahn wrote on Mon, Sep 27, 2004 at 09:18:08PM CEST:
> On Mon, 27 Sep 2004, Ralf Wildenhues wrote:
> >* Bob Friesenhahn wrote on Mon, Sep 27, 2004 at 08:32:31PM CEST:
> >>On Mon, 27 Sep 2004, Ralf Wildenhues wrote:
> >>
> >>>Most systems accept this error, because it's so widespread.
> >>>No excuse stating that sign conversion is difficult in C ;)
> >>
> >>It seems that the correct answer to this depends on which standard you
> >>are conforming to and whether you are using C or C++.  In C '89 these
> >>functions accept an 'int'.  Naturally C++ has made things more
> >>correct.
> >
> >No.  It's independent of the standard, except that I forgot C++
> >compatibility, sorry.  At least in C the cast to unsigned char is
> >required, because in C these functions accept ranges within unsigned
> >char only.  So for C++ we need to add another cast.  Sigh.
> 
> I don't think that libltdl needs to compile using C++ (I don't know of 
> any such requirement).  What I meant was that C++ is much more 
> concise/correct than C regarding characters passed to/from functions. 
> The C '89 language uses `int' while C++ uses `char'.

Yes, but: the C++ standard is supposed to be aligned with C89.

> So my system uses the prototypes:
> 
>   int isalnum(int c);
>   int isspace(int c);

..which means, that not only your system uses this, but every system
that adheres to C89.  And every system adhering to C++98 uses
corresponding

  extern "C" int isalnum(int);

or compatible prototypes.  Other than that, C++ is to be treated as a
different language as C.  Character handling is different, not more
correct or less.

> >No.  This issue was discovered by reading the code alone.  If you think
> >my change is wrong, I would like backing from some standard which tells
> >me so.  I know the original code can fail on some systems when given
> >characters with the 8th bit set, whereas the corrected does not.
> >
> >OK, updated patch below.  Does this fix the warnings?
> 
> I didn't mean to imply that there were new warnings.  Libltdl already 
> produces gobs of warnings.  It was just that the cast was to a type 
> other than what the function prototypes specify.

Yes, but the C standard specifies conversion from unsigned char to int:
C99, 6.3#2 says explicitly:

| The following may be used in an expression wherever an int or unsigned
| int may be used:
|    - An object or expression with an integer type whose integer
|      conversion rank is less than the rank of int and unsigned int.

So since there is no reason for casting to int (not even prevention of
warnings, as you mentioned), I am reverting this patch.

Regards,
Ralf