git-archive ignores export-ignore

Using git 1.6.4.2, the manual page prophesies that:

    Files and directories with the attribute export-ignore won't be
    added to archive files. See gitattributes(5) for details.

However, various tools seem to be unhappy about this attribute. In particular, checkout complains about invalid attributes:

    $ cat .gitattributes
    *.py export-subst
    *.html export-subst
    juniperxml.cfg export-ignore

    $ git checkout juniperxml.cfg
    export-ignore is not a valid attribute name: .gitattributes:3

It appears that git-archive doesn't seem to respect the attribute either. Consider the following example:

    cd /tmp
    mkdir foo
    cd foo
    git init
    touch foo.sh
    echo '*sh' > .gitattributes
    git add foo.sh
    git add .gitattributes
    git commit -m test
    git archive HEAD
    pax_global_header00006660000000000000000000000064113261226070014512gustar00rootroot0000000000000052 comment=202d78e6a44c7f10bee2ad96a3b2adc80fc33468
    .gitattributes000066400000000000000000000000041132612260700137430ustar00rootroot00000000000000*sh
    foo.sh000066400000000000000000000000001132612260700121630ustar00rootroot00000000000000

In addition to whatever bugs there might be in reading and applying the attributes correctly, it would seem that the man page could be a bit clearer on how and where to apply the attributes.

Assuming PEBKAC for the moment, how can I get the behavior I'm expecting?

"Todd A. Jacobs" <nospam@codegnome.org> writes:

    $ cat .gitattributes
    * whitespace=!indent,trail,space
    *.[ch] whitespace=indent,trail,space
    *.sh whitespace=indent,trail,space
    COPYING export-ignore
    $ git checkout COPYING
    $ wc -l COPYING
    360 COPYING

There is something that is different in my failed attempt to reproduce it from your problem description, but I cannot tell what it is.

Do you have some garbage character immediately after "e" in "-ignore"?

On Thu, Jan 21, 2010 at 04:57:39PM -0800, Junio C Hamano wrote:

No, but it turns out that juniperxml.cfg is actually a symlink in this repository. I'm not sure that should make a difference, but perhaps that's the missing element in recreating the problem.

I wouldn't think so, because the reason it's a symlink is because I was unable to get git-archive to ignore the actual file on export, so I moved it elsewhere and replaced it with a symlink so that the contents of the file weren't exported.

Hope that helps!