RE: DBCS to Shift-JIS -- MIDRANGE-L

Joe,

If you open the file is a "binary" file it won't do translation, and
therefore the data that ends up in the file will be bit-for-bit the same
as the data you're writing.

If you're doing the translation manually, that's probably want you want.

You do that by specifying a "b" in the mode parameter to fopen().  For
example:

in ILE C:
       f = fopen("/path/to/myfile.txt", "wb");

or in ILE RPG:
       eval      f = fopen('/path/to/myfile.txt': 'wb')

For the open() API, binary is the default if you don't specify O_TEXTDATA.

More info is in the ILE C/C++ Programmers Guide:
http://publib.boulder.ibm.com/iseries/v5r2/ic2924/books/c092712317.htm#HDRHBSF

Good Luck


On Sun, 25 Jan 2004, Joe Pluta wrote:

> Okay, I'm about to lose my mind here.
>
> I've done all the research I can to address the issue.  The business
> problem is to take a small hunk of EBCDIC DBCS data and write it to a
> file in Shift-JIS encoding, along with normal ASCII information.  The
> second part is the killer, as you'll soon see.
>
>
> Okay, I tried the simplest way - just opening a file using the IFS APIs
> (_C_IFS_fopen) with a CCSID of 943.  No go.  I keep getting conversion
> errors on the file open.  Maybe because I'm a single-byte machine?  I
> cannot create files of CCSID 5026 or 943 using the fopen API.
>
> Okay, that means I'll have to do the conversion myself.  After a little
> clonking around, I came up with some rather nice wrapper routines for
> the iconv family of APIs.  I can now convert anything to anything in RPG
> with no problems.
>
> So, that's the HARD part, right?  The easy part should be writing this
> raw data to the file, right?  Well, I'm missing something, because it
> ain't easy.
>
> If I open the file with an ASCII codepage, all of the routines (this
> includes fputs, fputc and fwrite) convert the bytes from EBCDIC to
> ASCII.  Since I've already converted to ASCII, this results in a double
> conversion and unusable output.  So I have to somehow stop the
> conversion.
>
> I tried opening a file with CCSID 65535, and while I can create it, I
> can't then reopen it for write.  I get a null pointer from fopen,
> although there is no error when I check the strerror function (the error
> code is zero, and the text you get back says "There is no error."  How
> helpful...
>
> So, my last resort was to use EBCDIC, with the idea that I would later
> change the CCSID using the CHGATR.  IT WORKED!  Except for one little
> problem:
>
> In ASCII, the percent sign (%) is x'25'.  In EBCDIC, x'25' is LF
> (linefeed).  All of the Unix-like routines have this neat little quirk -
> when they notice a LF, they magically prepend a CR (carriage return).
> So now, wherever there is a percent sign in my output, I see a newline.
>
> What should be:
>
> <HTML>
> <% bean.method(); %>
>
> Is instead:
>
> <HTML>
> <
> % bean.method();
> %>
>
> I tried every IFS write technique I could think of.  I used fputs,
> fputc, fwrite and putc, and all of them acted the same way.  The only
> thing I can figure is that I somehow have to set the CCSID of my job to
> an ASCII value and open the file with an ASCII code.  But I can't for
> the life of me figure out how to do that.
>