Re: [bug-diffutils] [BUG] Incorrect diff output (diff -d is correct)

[Jean Delvare]

Hallo Andreas,

On Thu, 30 Sep 2010 01:46:48 +0200, Andreas Gruenbacher wrote:
> On Wednesday 29 September 2010 18:45:29 Jean Delvare wrote:
> > Hi there,
> > 
> > I have attached two versions of a C source GNU diff has a hard time
> > diff'ing. The only difference between the two files is that several
> > functions have been moved around to avoid forward declarations. But the
> > output of "diff -u" is more complex than it should be. While I can
> > understand that the algorithm is not perfect and can lead to results
> > different from what a human would have done, the following excerpt:
> > 
> > -   if (!i2c_check_functionality(adapter, I2C_FUNC_SMBUS_BYTE_DATA))
> > -           return -ENODEV;
> > +   if (!i2c_check_functionality(adapter, I2C_FUNC_SMBUS_BYTE_DATA))
> > +           return -ENODEV;
> > 
> > is obviously wrong, no matter how you look at it.
> >
> > "diff -ud" gets the correct output.
> 
> But doesn't this already answer your question?  Diff doesn't find the minimal 
> diff when using the fast algorithm which doesn't guarantee optimal results.

No, this doesn't answer my question. While I understand that the fast
algorithm isn't guaranteed to find the minimal diff, the specific
excerpt I quoted above is not something I consider acceptable. Removing
2 lines to add exactly the same 2 lines immediately doesn't make any
sense. The fact that stripping context earlier in both source files
causes diff to no longer generate this strange output is also a clear
hint that this isn't a fundamental limitation of the fast algorithm, but
more likely a corner case which may be fixable.

> > Apparently git does as well (I originally received the difference between
> > the two files as a git patch.)
> 
> Git 1.7.3.rc2 produces the same, non-minimal result as GNU diff 2.8.7-cvs 
> without the -d option for me.

Odd, as the patch I received originally came from git and looked good.
I asked the sender if he had a special configuration setting for patch
generation and here is his answer:

"Nothing special, really. Currently I am using git verson 1.7.0.87, and 
format-patch 
to produce patch files. I had another brief look into the branch with the 
change;
git show also produces clean output."

I can't explain why your results are different (remember I'm not so
familiar with git.)

-- 
Jean Delvare