Re: [Bug-gnulib] vasprintf proposed fix for int overflow check

[Bruno Haible]

Paul Eggert wrote:
> Shouldn't asprintf and vasprintf be consistent with snprintf?
> snprintf is defined by POSIX to set errno to EOVERFLOW in this case.

It's just that the other implementations of asprintf() and vasprintf()
currently don't do it like this.

- glibc's libio/vasprintf.c has this code:

  ret = INTUSE(_IO_vfprintf) (&sf._sbf._f, format, args);
  if (ret < 0)
    {
      free (sf._sbf._f._IO_buf_base);
      return ret;
    }

  i.e. it does not guarantee an errno (since free() can kill errno).

- *BSD's documentation http://www.chedong.com/phpMan.php/man/asprintf/3
  doesn't mention errno, and its code
    
http://www.freebsd.org/cgi/cvsweb.cgi/src/lib/libc/stdio/asprintf.c?rev=1.13&content-type=text/plain
    
http://www.freebsd.org/cgi/cvsweb.cgi/src/lib/libc/stdio/vasprintf.c?rev=1.18&content-type=text/plain
  always sets errno to ENOMEM in case of any failure.

> Perhaps glibc's documentation and behavior need to be changed to
> reflect this; but let's start with gnulib.

Well, I think it's better to start with glibc, since it's harder to convince
Ulrich than me :-)

> I looked into EOVERFLOW a bit more, and the usual practice in gnulib
> (e.g. poll.c, c-stack.c) is to substitute EINVAL for EOVERFLOW if
> EOVERFLOW is not available.  So, how about this further patch?

Looks good. I propose to commit it when we get green light on the glibc
side as well.

Bruno