Re: hmm..

[Valdis . Kletnieks]

On Mon, 22 Dec 2003 13:31:51 PST, Linus Torvalds said:

>  - algorithmically, there aren't that many ways to test whether a 
>    character is a number or not. That's _especially_ true in
>    C, where a macro must not use it's argument more than once. So for 
>    example, the "obvious" implementation of "isdigit()" (which tests for 
>    whether a character is a digit or not) would be
> 
> 	#define isdigit(x) ((x) >= '0' && (x) <= '9')
> 
>    but this is not actually allowed by the C standard (because 'x' is used 
>    twice).

Somebody tell IBM that.  From the AIX 4.3.3 and 5.1 /usr/include/ctype.h:

#define _VALC(__c)              ((__c)>=0&&(__c)<=256)
#define _IS(__c,__m)            (__OBJ_DATA(__lc_ctype)->mask[__c] & __m)
#define isalpha(__a)    (_VALC(__a)?_IS(__a,_ISALPHA):0)
#define isalnum(__a)    (_VALC(__a)?_IS(__a,_ISALNUM):0)
#define iscntrl(__a)    (_VALC(__a)?_IS(__a,_ISCNTRL):0)
#define isdigit(__a)    (_VALC(__a)?_IS(__a,_ISDIGIT):0)
#define isgraph(__a)    (_VALC(__a)?_IS(__a,_ISGRAPH):0)
#define islower(__a)    (_VALC(__a)?_IS(__a,_ISLOWER):0)
#define isprint(__a)    (_VALC(__a)?_IS(__a,_ISPRINT):0)
#define ispunct(__a)    (_VALC(__a)?_IS(__a,_ISPUNCT):0)
#define isspace(__a)    (_VALC(__a)?_IS(__a,_ISSPACE):0)
#define isupper(__a)    (_VALC(__a)?_IS(__a,_ISUPPER):0)
#define isxdigit(__a)   (_VALC(__a)?_IS(__a,_ISXDIGIT):0)
#define isascii(c)      (!((c) & ~0177))

You'd be *amazed* how far through memory a 'while (isalpha(*s++)) {..};' can go
(which in fact is how I discovered this blecherousness).

The AIX 4.3 support I contributed to  Sendmail 8.9.0 back in Feb 98 included a
work-around because IBM refused to fix it on the grounds that the VALC macro
was to protect against a SEGV if the macro was fed an 'int' rather than a
'char' (why they didn't just use 'mask[__c & 255]' is beyond me), and that you
only got hit if you compiled(*) with -D_ILS_MACROS.  At least IBM eventually fixed
isascii(), which was originally broken the same way.... 

Feel free to file this under "Code we can prove that IBM never contributed" :)

(*) The default is to use actual function calls due to locale considerations - building
with _ILS_MACROS provides a measured 30%+ CPU savings for Sendmail, which
doesn't care if it's nailed into a 'LANG=C' environ anyhow...

Attachment: pgp00000.pgp
Description: PGP signature