On Tue, 12 Sep 2006 17:31:38 -0401
Ray Lai <[EMAIL PROTECTED]> wrote:
> Try to minimize the changes in your diff. Remember, these diffs may be
> sent upstream. No way they will accept a diff that removes __sun__ and
> _WIN32 defines!
>
> There is no need to change the !fstat() and if (ioctl()) to
> 0 == fstat() and 1 != ioctl(). While this may be your preference,
> sneaking in stylistic changes makes it more difficult to judge a diff's
> correctness.
>
> Are you trying to check that the file is a block device with
> S_IFBLK == (S_IFBLK & sb.st_mode)? If so, (S_IFBLK & sb.st_mode) is
> clearer.
Well, S_IFBLK is _two_ bits (check /usr/include/sys/stat.h).
> Why do you only check for block devices in the new code? The original
> code checks for character devices. Do we now ignore character devices?
>
> Should you be checking the ioctl return value against -1? I'm not
> sure, I don't use ioctl.
Yes, that's an error! A revised patch-block_c is attached
for readers' convenience. Now it tries DIOCGDINFO on both
character and block devices, falling back to lseek.
> This diff should be sent to the port maintainer after tree unlocks.
That would be [EMAIL PROTECTED] I guess he, or anyone with commit priv, could
just pick it up from ports@ if they like it. Wrong?
>
> -Ray-
You've improved the diff a great deal. Thank you.
Paul
--- block.c.orig Wed May 3 20:32:58 2006
+++ block.c Tue Sep 12 22:52:56 2006
@@ -30,6 +30,7 @@
#include <sys/ioctl.h>
#include <sys/queue.h>
#include <sys/disk.h>
+#include <sys/disklabel.h>
#endif
#ifdef CONFIG_COCOA
@@ -671,6 +672,7 @@
int64_t size;
#ifdef _BSD
struct stat sb;
+ struct disklabel lab;
#endif
#ifdef __sun__
struct dk_minfo minfo;
@@ -688,6 +690,10 @@
if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
#ifdef DIOCGMEDIASIZE
if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
+#elif defined(DIOCGDINFO)
+ if (ioctl(fd, DIOCGDINFO, &lab) != -1)
+ size = (int64_t)lab.d_secsize * (int64_t)lab.d_secperunit;
+ else
#endif
#ifdef CONFIG_COCOA
size = LONG_LONG_MAX;
