Re: [PATCH] Show binary file size change in diff --stat

[Rogan Dawes <discard@dawes.za.net>]

Johannes Schindelin wrote:
> Hi,
> 
> On Wed, 28 Feb 2007, Rogan Dawes wrote:
> 
>> Johannes Schindelin wrote:
>>
>>> On Wed, 28 Feb 2007, Rogan Dawes wrote:
>>>
>>>> How about showing the size of the changes between the 2 files via 
>>>> the libxdiff binary patch function?
>>> I briefly considered this, too. But what would it tell you in the case 
>>> of a jpg? I think it has more disadvantages than advantages...
>> It would still tell you the extent of the changes. i.e. Did we change 
>> only 10 bytes of the file, or is it a dramatic change?
> 
> I was not explicit enough, okay. I was not so worried about the case where 
> only 10 bytes changed. If you insert a single dot in a jpg image, chances 
> are that your binary content will change _a lot_.
> 
> So, no problem deducing from 10 bytes changed that it was a minor change. 
> But you cannot deduce the opposite of a 1MB change!
> 
> Hth,
> Dscho
> 

Yeah, I did understand that.

However, if we were to generalise, making a change to an XML document 
(wrapping the whole thing in an additional tag, with the associated 
indentation adjustments), we'd still see all the lines in the file 
change, even though the actual (semantic?) change is small. However, we 
would still report the full numbers of lines added and deleted regardless.

Why should a binary file be any different? It would be up to the 
observer to make a conclusion as to the significance of the numbers, 
based on their intrinsic knowledge of the file type, and its properties.

I just believe that showing bytes changed for binary files is 
_consistent_ with what we do for text.

Rogan