lkml.org 
[lkml]   [2010]   [Nov]   [4]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    SubjectRe: [PATCH 1/1] fs: Small refactoring of the code in ext4 2.6.37-rc1
      Hello.

    This is true.
    Despite having gone through all compilation and I have not noticed apparent problems, my mistake is evident in this case.

    I had not realized that the side effect would really be changing variables incremented and decremented.

    Thank you.

    I'll think more about this type of side effect. :)


    On Thu, 4 Nov 2010, André Luis Pereira dos Santos - BSRSoft wrote:

    > From: Andre Luis Pereira dos Santos <andre@bsrsoft.com.br>
    >
    > Hi.
    > Small refactoring of the code in order to make minor enhancements to critical areas.
    > The notation x + 1 has been replaced by more efficient notation x + +.
    >
    > Signed-off-by: Andre Luis Pereira dos Santos <andre@bsrsoft.com.br>
    > ---
    > Signed-off-by: Andre Luis Pereira dos Santos <andre@bsrsoft.com.br>
    > --- linux-2.6.37-rc1/fs/ext4/extents.c 2010-11-01 09:54:12.000000000 -0200
    > +++ linux-2.6.37-rc1-patched/fs/ext4/extents.c 2010-11-04 19:54:26.000000000 -0200
    > @@ -555,9 +555,9 @@ ext4_ext_binsearch(struct inode *inode,
    > while (l <= r) {
    > m = l + (r - l) / 2;
    > if (block < le32_to_cpu(m->ee_block))
    > - r = m - 1;
    > + r = m--;
    > else
    > - l = m + 1;
    > + l = m++;

    These do not give identical results.

    foo = bar + 1; assigns (bar + 1) to foo.
    foo = bar--; assigns bar to foo then decrements bar.
    foo = --bar; decrements bar then assigns bar to foo.

    So your change both change the value that will be assigned to 'r' and 'l'
    and also modify 'm' which was not previously modified.


    > ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
    > m, le32_to_cpu(m->ee_block),
    > r, le32_to_cpu(r->ee_block));
    > @@ -1557,7 +1557,7 @@ static int ext4_ext_try_to_merge(struct
    > if (ext4_ext_is_uninitialized(ex))
    > uninitialized = 1;
    > ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
    > - + ext4_ext_get_actual_len(ex + 1));
    > + + ext4_ext_get_actual_len(ex++));

    After your change gcc complains:

    fs/ext4/extents.c:1559:16: warning: operation on ‘ex’ may be undefined
    fs/ext4/extents.c:1559:16: warning: operation on ‘ex’ may be undefined

    which it is correct in doing since you are now modifying the value of the
    pointer which is dereferenced in the assignment. Previously the value of
    (ex+1) was simply passed to ext4_ext_get_actual_len(), but now you are
    passing the value of (ex) to ext4_ext_get_actual_len() and then
    subsequently incrementing 'ex' itself.


    > if (uninitialized)
    > ext4_ext_mark_uninitialized(ex);
    >
    >


    Was this patch even compile tested?



    --
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    More majordomo info at http://vger.kernel.org/majordomo-info.html
    Please read the FAQ at http://www.tux.org/lkml/

    \
     
     \ /
      Last update: 2010-11-05 00:37    [W:0.025 / U:32.224 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site