Re: [HACKERS] another locale problem

>>>Tatsuo Ishii said:
[...]
> >From line 5388 of parser/gram.y:
> (strcmp(opname,"~*") == 0 && is
alpha(n->val.val.str[pos])))
>
> I suspect isalpha() does not return true if n->val.val.str[pos] is a
> non ascii char. Probably that's why
[...]

I have tried the following program under BSD/OS 4.0.1:

--- cut here ---
#include <string.h>
#include <locale.h>
#include <rune.h>
#include <errno.h>

main()
{
int i;

printf ("%s\n", setlocale(LC_CTYPE, "CP1251"));
printf ("--- strcoll ---\n");
printf ("%d\n", strcoll("àáâ", "àáã"));
printf ("%d\n", strcoll("àáâ", "àáâ"));
printf ("%d\n", strcoll("àáâ", "àáà"));
printf ("--- isalpha ---\n");
i = 'ë';
printf ("%c, %d\n", i, i);
printf ("%d\n", isalpha('ë'));
printf ("%d\n", isalpha('a'));
}
--- cut here ---

Where the cyrillic letter is the 'L' lowercase in cyrillic, the other are
first characters from the alphabet....

The strcoll work always. This is because BSD/OS does NOT support the
LC_COLLATE and strcoll is effectively strcmp (confirmed by the libc sources).

There are two cases:

1. cc -o test test.c

Returns 0 for isalpha() on cyrillic characers.

2. cc -funsigned-char -o test test.c

Returns 1 for isalpha() on cyrillic characters!

If I substitute the 'ë' character constant with it's code (235), isalpha()
returns 1.

BSD/OS 4.0.1 uses gcc 2.7.2.1. Apparently somewhere in the code 'char' is used
(I did use both --enable-locale and --with-mb=WIN). I am still searching for
it, but the regex code is not very readable... :-(

Daniel