git-diff-files and fakeroot

I've been trying to track down a strange issue with building kernels (and scripts/setlocalversion) and finally realized the problem was the when run under fakeroot, git-diff-files thinks everything is changed (deleted, I believe)

Oddly, running "git status" seems to correct things.

Running "fakeroot git-diff-files" gives me lines like this:

:100644 100644 f866059f24bacd314fa4a979334a9893dbfc19ba 0000000000000000000000000000000000000000 M      write-tree.c

Looking at strace output, I can't see a difference that appears meaningful,

I can probably work around this, some other way, but knowing if this is something that is fixable in git itself or not would be nice.

Thanks!

Ryan Anderson <ryan@michonline.com> writes:

How does fakeroot keep track of fake ownerships?

That is, I suspect:

	$ date >foo
	$ ls -l foo
	-rw-rw-r--  1 junio junio 29 Jan 16 18:29 foo
	$ fakeroot sh -i
        # ls -l foo
	-rw-rw-r--  1 root root 29 Jan 16 18:29 foo

which means that under fakeroot, stat would give file ownerships for *my* files as if they are owned by root.

And of course .git/index records the as me and fakeroot has no way of knowing me maps to root in that fake environment.

"git status" includes "git update-index --refresh", which reads from the stat bits and ownerships, notices that file contents have not changed, and writes a new index file that records these stat bits (so git thinks they are now owned by root).

If you come out of the fakeroot environment, I am reasonably sure that you would find all the index entries are dirty again, for exactly the same reason. index thinks things are owned by root, the files are owned by you, so without looking at file contents, it says "these things might have changed".

Ryan Anderson <ryan@michonline.com> writes:

BTW, Ryan, I suspect this is where you try to append "-dirty" to the version number. But I wonder why you are doing the build under fakeroot to begin with? Wasn't the SOP "build as yourself, install as root"?

On Mon, Jan 16, 2006 at 06:36:39PM -0800, Junio C Hamano wrote:

That's exactly what started this search, because I was running "make deb-pkg". (Effectively.) dpkg-buildpackage wants to think it is running as root, either via sudo or via fakeroot. I had my build environment switched over entirely to fakeroot, as it just seems to be a better practice, but I've temporarily switched back to sudo.

However, your explanation has pointed out to me how I can solve this - run "fakeroot -u" instead of "fakeroot", and I think it will be fixed.

lkml cc:ed to hopefully stick this in an archive where someone else will find it.

On Jan 17, 2006, at 00:27, Ryan Anderson wrote:

You should run "make" first, then after that completes run "fakeroot make deb-pkg". I think this is similar to what the Debian package "kernel-package" does, except it substitutes an alternate "debian/" directory. IIRC, it just runs "make install" as a normal user to a staging directory, then runs "$(ROOTCMD) dpkg-deb -b [...]" to build the package. IMHO it's somewhat of a cleaner solution, and I've used it for several years now with no issues.

Cheers, Kyle Moffett

--
I have yet to see any problem, however complicated, which, when you  
looked at it in the right way, did not become still more complicated.
   -- Poul Anderson

Kyle Moffett wrote:

Right "make all && fakeroot make deb-pkg" was failing, because with CONFIG_LOCALVERSION_AUTO set, I was both getting a "-dirty" string appended to the version, as well as something awfully close to a full rebuild due to the version number changing, and on top of all that, the dpkg-buildpackage would fail, as "-dirty" doesn't have a number in it (hence why you see dfsg1 or ubuntu0 in version strings.)

I think I might take your suggestion, and fix up the builddeb script to do the "run as root" part itself, rather than needing to do it outside. It would make it possible to just run "make oldconfig deb-pkg" which would make things a little bit simpler.

On Tue, Jan 17, 2006 at 01:08:48AM -0500, Ryan Anderson wrote:

If we do something it must be consistent for all *-pkg targets. So fixing up builddeb is not enough, we must fix it for rpm etc also.

Not that I have looked into what is needed, but we shall not have inconsistent behavious between the different *-pkg targets.

	Sam