Bug: file named - on git commit

Hi

Found a little issue in git version 1.7.9.5 if a file named "-", causing "git commit" to read from stdin.

(So you must hit ctrl-d or ctrl-c to finish the commit.)

Everything looks ok to me after the commit. Other users reported to be fixed in 1.8.1.1 but haven't it tested myself.

This does not work:

mkdir tmp && cd tmp; echo foo >./-; git init; git add .; git commit -m "is this a bug?"

Kind regards

René

Rene Moser <mail@renemoser.net> writes:

Can't reproduce with Git version 1.8.1.1.440.g1d329bd, this probably has been fixed already.

On Mon, Jan 28, 2013 at 5:38 PM, Rene Moser <mail@renemoser.net> wrote:

Yes, it's fixed in 4682d85 (diff-index.c: "git diff" has no need to read blob from the standard input - 2012-06-27) since v1.7.11.3.

Rene Moser <mail@renemoser.net> writes:

This was fixed by Junio around 4682d85 (diff-index.c: "git diff" has no need to read blob from the standard input, 2012-06-27), which is included starting with v1.7.12 and the v1.7.11.3 maint release. Please upgrade.

On 01/28/2013 12:05 PM, Thomas Rast wrote:

Thanks.

Hi,

Thomas Rast wrote:

[...]

Should upgrade-averse folks stuck on 1.7.10.y (like Debian 7.0, which is currently in the release candidate stage) take this fix? Do you happen to know of any other fixes such people would want?

Thanks, Jonathan

Jonathan Nieder <jrnieder@gmail.com> writes:

FYI, the fix referred to in this thread are three-patch series that forked from 1.7.6.6, so it should be trivial to merge it even to such an old version.

The topic-branch workflow shines ;-)

Jonathan Nieder <jrnieder@gmail.com> writes:

There are files with four dotted decimal numbers in their names in the Documentation/RelNotes/ directory to help distro maintainers like you to figure it want.

This is a tangent, but even with a project like git that is managed with a good use of topic branch workflow, we may want to have a way to reliably identify the tip of an ancient fix like this. People may be able to bisect down to 4682d85, and in this particular case, I happen to know that there wasn't any side-effect breakage introduced by that commit, but there needs to be an easy way (it can be expensive to compute) to make sure there is no follow-up fix to that particular commit.

I can read "git rev-list --parents | grep -C3 $(git rev-parse 4682d85)" and then figure out what the children commits of that fix are, of course, but I suspect most people will view it as primitive ;-)

Jonathan Nieder <jrnieder@gmail.com> writes:

I've been wondering if we can help automating this for backporters.

Because of the way my integration branches are managed, if you run

	git log --first-parent v1.8.0..maint-1.8.0
	git log --first-parent v1.8.1..maint

the output should give us a birds-eye view (because most are merges of one or more patches on a topic) of the changes that are fixes, excluding any feature enhancements.

You can then iterate over the single patches applied directly on top of maint (or maint-1.8.0) and tips of the topics merged to maint (or maint-1.8.0) and see if each of them is applicable to maint-1.7.10 codebase. I think you can mechanically reject the ones that are on 'maint' that merge topics that were forked from v1.8.1 as too new. That hopefully culls the topics that needs manual review and assessment (some may be too minor to be worth backproting, for example).

You should be able to do the same for

        git log --first-parent v1.8.1..master

There will be fixes and features mixed in the output, but if you can mechanically narrow down the ones that may be relevant to your old maintenance track, eyeballing the rest to judge if each of them is worth backporting will become a manageable task.

Junio C Hamano wrote:

Yes, this is the part I was asking for help with. Backporting is easy but convincing the release team and upgrade-averse sysadmins to like the result generally isn't. Occasional nominations of the form "this change is important in my workflow" could help.

Continuing to stick to fixes to very severe bugs that stand out plus a random assortment of problems people have reported can also work fine, though.

Thanks, Jonathan