full_read depends on incoming errno

[Andy Wingo]

Hi,

We use full_read in Guile and just got a bug report that full_read was
depending on the incoming errno.  Eli Zaretskii proposed that the fix be
like this:

> +    errno = 0;
>      if (full_read (fd, cookie, sizeof cookie) != sizeof cookie
[...]

To recall, safe_read (called by full_read) looks like this:

    /* Read(write) up to COUNT bytes at BUF from(to) descriptor FD, retrying if
       interrupted.  Return the actual number of bytes read(written), zero for 
EOF,
       or SAFE_READ_ERROR(SAFE_WRITE_ERROR) upon error.  */
    size_t
    safe_rw (int fd, void const *buf, size_t count)
    {
      /* Work around a bug in Tru64 5.1.  Attempting to read more than
         INT_MAX bytes fails with errno == EINVAL.  See
         
<http://lists.gnu.org/archive/html/bug-gnu-utils/2002-04/msg00010.html>.
         When decreasing COUNT, keep it block-aligned.  */
      enum { BUGGY_READ_MAXIMUM = INT_MAX & ~8191 };

      for (;;)
        {
          ssize_t result = rw (fd, buf, count);

          if (0 <= result)
            return result;
          else if (IS_EINTR (errno))
            continue;
          else if (errno == EINVAL && BUGGY_READ_MAXIMUM < count)
            count = BUGGY_READ_MAXIMUM;
          else
            return result;
        }
    }

Here we see that if the system does not set errno on a partial read, as
is apparently the case in MinGW, then we depend on the incoming value of
errno.

WDYT about having safe_rw set errno before entering the loop?

Thanks,

Andy
-- 
http://wingolog.org/