From: Michael J Gruber <git@drmicha.warpmail.net>
Date: Sat, 21 Nov 2009 17:31:35 GMT
Subject: Re: How to make git diff-* ignore some patterns?
Message-ID: <4B0823F7.7030607@drmicha.warpmail.net>
In-Reply-To: <4B0817EE.1040000@dirk.my1.cc>

Dirk Süsserott venit, vidit, dixit 21.11.2009 17:40:
> Hi list,
> 
> is there a way to tell "git diff-index" to ignore some special patterns, 
> such that /^-- Dump completed on .*$/ is NOT recognized as a difference 
> and "git diff-index" returns 0 if that's the only difference?
> 
>      -- Dirk
> 
> <Background>
> I have a mySQL database which I backup daily using mysqldump (cronjob).
> The result is a text file (*.sql) with all the "create" and "insert"
> statements and some metadata.
> I used to use tar and gzip to backup these files and got a huge
> collection of backups in the last tree years (500+ MB).
> Then I switched to Git and recorded only the diffs between day X and day
> X-1. My repository shrunk to 16 MB for the very same data, which was great!
> 
> My database doesn't change every day, but I backup it anway and store 
> the backup files with Git and a cronjob. It does:
> 
> ---------------
> mysqldump ... -r <backupfile> # that's the output file ;-)
> git add <backupfile>
> if ! git diff-index --quiet HEAD --; then
>      git commit -m "Backup of <database> at <timestamp>"
> fi
> ---------------
> 
> This way, a new commit is only done when the backupfile has changed. So 
> far, so perfect.
> A few days ago my web hoster (where the database actually resides) 
> changed the mySQL version.
> mysqldump now writes "-- Dump completed on <timestamp>" to the file and 
> Git correctly recognizes this as a change and my script creates a new 
> commit. Every day, even if only that line has changed.
> 
> I'd like to skip these commits if only the "Dump completed" line has 
> changed.
> </Background>

Is the dump guaranteed to be in a specific order? If yes then this
procedure makes sense. (pdfs etc. are problematic because of reordering.)

You can either egrep -v through the output of git diff-index, or define
a diff driver: set an attribute, say "dumpdiff", for dump files (see
gitattributes) and define diff driver as
git config diff.dumpdiff.textconv = dumpdiff.sh
where dumpdiff.sh is "egrep -v ...". You may need to call diff-index
with --ext-diff. I haven't tried, though ;)

Cheers,
Michael