SEO

I'm using an HP-UX release B.11.11 U 9000/800 969017836.
The iconv(3C) function seems to be returning EINVAL when I
expect it to return EILSEQ.

Specifically, when I run

#include <iconv.h>
#include <stdio.h>
#include <errno.h>
#include <assert.h>

int main(void)
{
char inbuf[3] = "\200\333", outbuf[32], *in = inbuf, *out = outbuf;
size_t n, insize = 2, outsize = sizeof outbuf;
iconv_t context = iconv_open("roman8", "utf8");
assert(context != (iconv_t)-1);

n = iconv(context, &in, &insize, &out, &outsize);
printf("result %d, errno %d: %s (EILSEQ %d, EINVAL %d)\n",
(int)n, errno, strerror(errno), EILSEQ, EINVAL);

iconv_close(context);
return 0;
}

it prints
result -1, errno 22: Invalid argument (EILSEQ 47, EINVAL 22).

Nothing else I've fed it has convinced the UTF-8 converter to
return EILSEQ.

Am I mistaken in my expectations about iconv's errno semantics?
And if this is a known bug, which patchlevel do I need to get to
to fix it?

[Why is the difference at all interesting? I'm converting a
stream of multibyte characters. EINVAL is normally iconv's
way of saying "hey, there's half a multibyte character at
the end of this buffer, give me more input before I can
finish converting this"--not an error at all. In contrast,
EILSEQ requires the character conversion equivalent of a
Heimlich maneuver if one continues at all--remove all or parts
of the offending character, reset the conversion state, emit
a placeholder on the output side.]

Thanks!,

Jutta Degener <jutta@pobox.com>