On Mon, 3 Nov 2014 17:12:47 -0800, Charles Mills wrote:
>
>I have seen code that, for example, converted character digits to binary by
>subtracting 0x30. Guess what? Does not work for EBCDIC.
>
It would have been smarter and more likely portable to subtract ('0' - 0).
Or to subtract ( 'a' - 'A' ) for an ad-hoc toupper().
>I don't think any language converter should presume that 0x30 really means
>0xf0, or that \xa0 means anything other than 0xa0.
>
I ran into such a problem trying to compile a terminfo source (with "tic") that
contained octal escape sequences for ASCII command codes. Worked fine
once I translated to EBCDIC by hand. Left me somewhat persuaded that
_Pragma convert() or the like should convert '\x30' assuming it means a char,
and leave 0x30 alone assuming it means an numeric object.
I suspect I'm close to a reason that the POSIX Curses library is not supported
in Enhanced ASCII mode.
Of course, coders would need to be alert to the distinction. They're not.
Cf. Guido van Rossum.
How is _Pragma convert() treated by non-IBM compilers? Could it be
neutralized with macro definitions? More readily if it were a single token:
_Pagma_convert(), so one could:
#if ( ! defined( __IBM__ ) )
#define _Pragma_convert( x ) /* nothing */
#endif
I hate EBCDIC!
-- gil
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN
