Re: [patch, fortran] Wide character I/O Part 1

[Tobias Burnus <burnus at net-b dot de>]

Jerry DeLisle wrote:
> This patch implements wide character I/O with default encoding for 
> list directed formatted and formatted I/O.
Great! I have to admit, I only loosely tested it.

I have a question: Can one implement writing character(kind=4) with 
encoding="default" such that Latin 1 (ISO-8859-1) characters will not be 
translated into '?' but come out as 8bit characters?

In principle I had expected that the following does so, but my tests fail:

+	  *p++ = (unsigned int) source[i * kind] > 255 ?
+		 '?' : source[i * kind + endian_off];


Test case:
(Note there is some format checking which also goes wrong.)

! Compile with -fbackslash
character(kind=4,len=20) :: str = 4_'X\xF8öABC'
!print '(3a0)', ':',trim(str),':' ! REJECTED, but valid?
print *, ':',trim(str),':'
end


> I don't want to let the patch get too big so if I may, I would like to 
> commit in parts. Part 2 will take care of anything remaining for 
> unformatted wide char I/O and move on to ENCODING="UTF-8".
Great. With the next patch we should have the complete ISO 10646 and 
UTF-8 support implemented as described in the Fortran 2003 standard. 
(Modulo bugs and enhancements beyond the standard.) I think that your 
next patch can then also enable selected_char_kind('ISO 10646')  (see 
gfc_simplify_selected_char_kind).

> I have regression tested this on x86-64 and ppc64.  The patch takes 
> care of endianess.
> OK for trunk?
Looks OK. Thanks for the patch.

Tobias