patcht1304: Set LOGNAME even if USER is unset or null

Avoid:

  # ./t1304-default-acl.sh
  ok 1 - checking for a working acl setup
  ok 2 - Setup test repo
  not ok 3 - Objects creation does not break ACLs with restrictive umask
  #
  #               # SHA1 for empty blob
  #               check_perms_and_acl .git/objects/e6/9de29bb2d1d6434b8b29ae775ad8c2e48c5391
  #
  not ok 4 - git gc does not break ACLs with restrictive umask
  #
  #               git gc &&
  #               check_perms_and_acl .git/objects/pack/*.pack
  #
  # failed 2 among 4 test(s)
  1..4

on systems where USER isn't set. It's usually set by the login process, but it isn't set when launching some Docker images. For example:

  $ docker run --rm debian env
  HOME=/
  PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
  HOSTNAME=b2dfdfe797ed

'id -u -n' has been in POSIX from Issue 2 through 2013 [1], so I don't expect compatibility issues.

[1]: http://pubs.opengroup.org/onlinepubs/9699919799/utilities/id.html

Signed-off-by: W. Trevor King <wking@tremily.us>
---
The patch is based on the current maint branch.

Previous LOGNAME discussion:

* Michael Gruber on 2011-05-06 suggesting a discussing a whoami
  fallback [1] (but whoami isn't POSIX).
* René Scharfe on 2011-10-14 suggesting USER as a fallback for
  LOGNAME [2].
* Matthieu Moy on 2012-09-17 suggesting dropping $LOGNAME in
  favor of numerical user IDs 'id -u' for a system with multiple
  usernames sharing the same user ID [3].

Obviously, you can work around the problem with:

  # USER=$(id -u -n) ./t1304-default-acl.sh

so the question is really "Are empty-USER systems worth supporting out of the box?".

Cheers, Trevor

[1]: http://thread.gmane.org/gmane.comp.version-control.git/172883/focus=172961 [2]: http://thread.gmane.org/gmane.comp.version-control.git/183586 [3]: http://thread.gmane.org/gmane.comp.version-control.git/205690/focus=205703

 t/t1304-default-acl.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

"W. Trevor King" <wking@tremily.us> writes:

I'll queue this as-is, but it makes me wonder if we want to do this without if/then/fi, e.g.

	: ${LOGNAME:=${USER:-$(id -u -n)}

Spelling everything out with if/then/fi is obviously at the other extreme, i.e.

	if test -z "$LOGNAME"
	then
		if test -n "$USER"
                then
			LOGNAME=$USER
		else
			LOGNAME=$(id -u -n)
		fi
	fi

but it probably is a very bad idea.

More importantly, what if none of the alternatives work? I personally feel it is OK to punt and declare test_done early, instead of giving false positive breakages like you saw without this patch.

On Sun, Oct 19, 2014 at 03:49:36PM -0700, Junio C Hamano wrote:

I'm fine with that too.

And I'm fine with this ;).

I can put this into a v2 if you like. Which conditional syntax do you prefer?

Cheers, Trevor

"W. Trevor King" <wking@tremily.us> writes:

Probably

    if test -z "$LOGNAME"
    then
            LOGNAME="${USER:-$(id -u -n)}"
    else
            : cannot test acl operations without a usable user name
            test_punt!
    fi