Ray Butterworth rbutterworth at watmath.UUCP
Sat Oct 18 06:21:29 AEST 1986

How many programs now rely on "while (isdigit(d=*++ptr)) ...",
or "while (isdigit(d=getchar())) ..." ?  Probably a lot,
since these nearly always work correctly.  The X3J11 committee
seems to have tried to make such things reliable and portable,
but they only went half way there.

    ANSI X3J11 4.3  <ctype.h>:
    ... In all cases the argument is an int, the value of which
    shall be representable as an unsigned char or shall equal
    the value of the macro EOF.  If the argument has any other
	value, the behaviour is undefined.

This doubles the size of the ctype table from 129 to 257 bytes,
but at least it fixes one problem with the current <ctype.h>
that blows up if getchar() returns a character with the upper bit
on.  e.g. if the input stream contains '\234', getchar() returns
the int 0234, and isdigit(0234) is sometimes true, sometimes
false, and possibly sometimes a machine fault.

But if they are going the solve the problem of isdigit(getchar()),
why couldn't they have added another 128 bytes and solved the
similar problem of isdigit(*string)?  If *string contains '\234',
*string may be evaluated as -0144, and isdigit(*string) will
still be unreliable.

As it is, the only portable version of this requires something
like "while (isdigit(d=*(unsigned char*)++ptr)) ...".  This
might be portable, but it sure is ugly, especially when one
considers that isdigit() is defined as an X3J11 portable C
function.  I can't imagine many people putting stuff like
this into their code (at least not until they actually run
into the problem).

It is unfortunate that X3J11 didn't either fix their definition to
be truely portable or at least leave these things in the mess that
they already are.

If I'm misreading their definition, and they really do want
both getchar() and *(char*)ptr to be valid arguments, perhaps
they could make this more explicit.

P.S. (I know that the number 128 and even the existence of the
table are implementation specific.)

More information about the Comp.lang.c mailing list