Make git ignore the date in PDF files

Question

First: I am aware of the general comment: Do not track generated files.

Say, I want to track generated PDFs and have git ignore the date written into the PDF. That means, I want git to treat two PDFs as the same if the only difference is the the Date information.

What I tried is a filter that -- in its clean part -- sets the date to some arbitrary value.

(--- comment ----
basically, the filter does sth along:

## dump the pdf metadata to a file and replace the dates
pdftk "$FILENAME" dump_data | sed -e '{N;s/Date\nInfoValue: D:.*/Date\nInfoValue: D:19790101072619/}' > "$TMPFILE"

## update the pdf metadata
pdftk "$FILENAME" update_info "$TMPFILE" output "$TMPFILE2"

) --- end comment ----

The filter works (the committed pdf has the date set to my arbitrary value) but I ran into files re-checked out from git repository with 'clean' filter end up with modified status

So, my filter is apparently not what I want to do here.

My question is:
1) Can I use a clever filter approach to get git ignore the date values in the PDF completely? And how?
or
2) What would be the correct approach if not filters?

Answer 1

Finally solved this with the help from the git mailing list. Wasn't a git issue after all, but more a problem of my filters expectations of pdftk. (Maybe an encoding thing? Did not dig deeper.)

The helpful message on the git mailing list is here: http://permalink.gmane.org/gmane.comp.version-control.git/224797

Basically, the filter script I wrote was not idem-potent, meaning that applying the clean filter a second time to a cleaned file would change the file.

Background: When pdftk is used to update the metadata of a pdf with the metadate it extracted from that exact pdf before, to my surprise it changes the pdf file.

So, I included a safety check into my filter and the issue has gone away.

For reference, here is the full filter:

 #!/bin/bash

 ## use GNU coreutils on OS X explicitely
 ## (install via homebrew, for instance:
 ##  > brew install coreutils
 ##  > brew install gnu-sed
 ## )
 if [ ${OSTYPE:0:6} == "darwin" ]; then
     MKTMP=gmktemp
     SED=gsed
 else
     MKTMP=mktemp
     SED=sed
 fi


 FILEASARG=true
 if [ "$#" == 0 ]; then
     FILEASARG=false
 fi

 if $FILEASARG ; then
     FILENAME="$1"
 else
     FILENAME=`$MKTMP`
     cat /dev/stdin > "${FILENAME}"
 fi

 TMPFILE=`$MKTMP`
 TMPFILE2=`$MKTMP`
 TMPFILE3=`$MKTMP`

 ## dump the pdf metadata to a file and replace the dates
 pdftk "$FILENAME" dump_data > "$TMPFILE3"
 $SED -e '/Date/{ N; s/Date\nInfoValue: D:.*/Date\nInfoValue: D:19790101072619/ }' < "$TMPFILE3" > "$TMPFILE"

 ## if the metadata did not change, do nothing
 if diff "$TMPFILE3" "$TMPFILE"; then
     rm "$TMPFILE3"
     rm "$TMPFILE"
     if [ -n $FILEASARG ] ; then
    cat "$FILENAME"
     fi
     exit 0
 fi

 ## update the pdf metadata
 pdftk "$FILENAME" update_info "$TMPFILE" output "$TMPFILE2"

 ## overwrite the original pdf
 mv -f "$TMPFILE2" "$FILENAME"

 ## clean up
 rm -f "$TMPFILE"
 rm -f "$TMPFILE2"
 if [ -n $FILEASARG ] ; then
     cat "$FILENAME"
 fi