Re: make git ignore the timestamp embedded in PDFs

From: Johannes Sixt <j6t@kdbg.org>
Date: May 18, 2013, 16:32 UTC
Message-ID: <5197AD3A.5050500@kdbg.org>
In-Reply-To: <87ppwohgyd.fsf@med.uni-goettingen.de>

Am 18.05.2013 09:42, schrieb Andreas Leha:

Show 27 quoted lines

>> Am 14.05.2013 15:17, schrieb Andreas Leha:
>>> Hi all,
>>>
>>> how can I make git ignore the time stamp(s) in a PDF.  Two PDFs that
>>> differ only in these time stamps should be considered identical.
>>> ...
>>> What I tried is a filter:
>>> ,----[ ~/.gitconfig ]
>>> | [filter "pdfresetdate"]
>>> |         clean = pdfresetdate
>>> `----
>>>
>>> This 'works' as far as the committed pdf indeed has the date reset to my
>>> default value.
>>>
>>> However, when I re-checkout the files, they are marked modified by git.
>>
>> I'm using cleaned files every now and then, but not on Linux. I have
>> never observed this behavior recently.
>>
>> If you 'git add' the file, does it keep its modified state? Does 'git
> 
> yes.
> 
>> diff' tell a difference?
> 
> no.

I do not believe you. I'm sure that "Binary files differ" was reported. The reason is that your pdfresetdate script is not idempotent. Look:

$ pdfresetdate < x.pdf > y.pdf $ pdfresetdate < y.pdf > z.pdf $ md5sum x.pdf y.pdf z.pdf c46a7097574a035e89d1a46d93c83528 x.pdf 8e6d942b4cc7d8a4dfe6898867573617 y.pdf e6333bc0f8ab9781d3e1d811a392d516 z.pdf

A file that was already cleaned by the clean filter must not be modified, i.e., the y.pdf and z.pdf should be identical. But they are not.

Fix your clean filter.

-- Hannes