[PATCH] ref namespaces: tests

[PATCH] ref namespaces: tests

[Josh Triplett]

Test pushing, pulling, and mirroring of repositories with ref
namespaces.

Signed-off-by: Josh Triplett <josh@joshtriplett.org>
Signed-off-by: Jamey Sharp <jamey@minilop.net>
---

The most recent "What's cooking" suggested that the ref namespaces
patches needed tests.  This test works with PATCHv10, currently in pu.
Please append this patch to the js/ref-namespaces branch.

 t/t5502-fetch-push-namespaces.sh |   77 ++++++++++++++++++++++++++++++++++++++
 1 files changed, 77 insertions(+), 0 deletions(-)
 create mode 100755 t/t5502-fetch-push-namespaces.sh

diff --git a/t/t5502-fetch-push-namespaces.sh b/t/t5502-fetch-push-namespaces.sh
new file mode 100755
index 0000000..85720b6
--- /dev/null
+++ b/t/t5502-fetch-push-namespaces.sh
@@ -0,0 +1,77 @@
+#!/bin/sh
+
+test_description='fetch/push involving ref namespaces'
+. ./test-lib.sh
+
+test_expect_success setup '
+	test_tick &&
+	git init original &&
+	(
+		cd original &&
+		i=0 &&
+		while [ "$i" -lt 2 ]
+		do
+			echo "$i" > count &&
+			git add count &&
+			test_commit "$i" &&
+			i=$(($i + 1))
+		done &&
+		git remote add pushee-namespaced "ext::git --namespace=namespace %s ../pushee" &&
+		git remote add pushee-unnamespaced ../pushee
+	) &&
+	git init pushee &&
+	git init puller
+'
+
+test_expect_success 'pushing into a repository using a ref namespace' '
+	(
+		cd original &&
+		git push pushee-namespaced master &&
+		git ls-remote pushee-namespaced > actual &&
+		printf "dc65a2e0f299dcc7efddbbe01641a28ee84329ba\trefs/heads/master\n" > expected &&
+		test_cmp expected actual &&
+		git push pushee-namespaced --tags &&
+		git ls-remote pushee-namespaced > actual &&
+		printf "fbdf4310c71b916568f04753f603fb24a0544227\trefs/tags/0\n" >> expected &&
+		printf "dc65a2e0f299dcc7efddbbe01641a28ee84329ba\trefs/tags/1\n" >> expected &&
+		test_cmp expected actual &&
+		# Verify that the GIT_NAMESPACE environment variable works as well
+		GIT_NAMESPACE=namespace git ls-remote "ext::git %s ../pushee" > actual &&
+		test_cmp expected actual &&
+		# Verify that --namespace overrides GIT_NAMESPACE
+		GIT_NAMESPACE=garbage git ls-remote pushee-namespaced > actual &&
+		test_cmp expected actual &&
+		# Try a namespace with no content
+		git ls-remote "ext::git --namespace=garbage %s ../pushee" > actual &&
+		test_cmp /dev/null actual &&
+		git ls-remote pushee-unnamespaced > actual &&
+		sed -e "s|refs/|refs/namespaces/namespace/refs/|" expected > expected.unnamespaced &&
+		test_cmp expected.unnamespaced actual
+	)
+'
+
+test_expect_success 'pulling from a repository using a ref namespace' '
+	(
+		cd puller &&
+		git remote add -f pushee-namespaced "ext::git --namespace=namespace %s ../pushee" &&
+		git for-each-ref refs/ > actual &&
+		printf "dc65a2e0f299dcc7efddbbe01641a28ee84329ba commit\trefs/remotes/pushee-namespaced/master\n" > expected &&
+		printf "fbdf4310c71b916568f04753f603fb24a0544227 commit\trefs/tags/0\n" >> expected &&
+		printf "dc65a2e0f299dcc7efddbbe01641a28ee84329ba commit\trefs/tags/1\n" >> expected &&
+		test_cmp expected actual
+	)
+'
+
+test_expect_success 'mirroring a repository using a ref namespace' '
+	git clone --mirror pushee mirror &&
+	(
+		cd mirror &&
+		git for-each-ref refs/ > actual &&
+		printf "dc65a2e0f299dcc7efddbbe01641a28ee84329ba commit\trefs/namespaces/namespace/refs/heads/master\n" > expected &&
+		printf "fbdf4310c71b916568f04753f603fb24a0544227 commit\trefs/namespaces/namespace/refs/tags/0\n" >> expected &&
+		printf "dc65a2e0f299dcc7efddbbe01641a28ee84329ba commit\trefs/namespaces/namespace/refs/tags/1\n" >> expected &&
+		test_cmp expected actual
+	)
+'
+
+test_done
-- 
1.7.5.4

Re: [PATCH] ref namespaces: tests

[Junio C Hamano]

Josh Triplett <josh@joshtriplett.org> writes:

> Test pushing, pulling, and mirroring of repositories with ref
> namespaces.
>
> Signed-off-by: Josh Triplett <josh@joshtriplett.org>
> Signed-off-by: Jamey Sharp <jamey@minilop.net>

Thanks.

>  create mode 100755 t/t5502-fetch-push-namespaces.sh

Isn't 5502 used already?

> +test_expect_success setup '
> +	test_tick &&
> +	git init original &&
> +	(
> +		cd original &&
> +		i=0 &&
> +		while [ "$i" -lt 2 ]
> +		do
> +			echo "$i" > count &&

This is just style, but the test scripts prefer to spell these like this:

	while test "$i" -lt 2
	do
        	echo "$i" >count &&
                ...

to favor "test" over "[ ... ]", and omit SP between ">" redirection (or
"<" for that matter) and the filename.

> +		git remote add pushee-namespaced "ext::git --namespace=namespace %s ../pushee" &&

Nice ;-).

> +test_expect_success 'pushing into a repository using a ref namespace' '
> +	(
> +		cd original &&
> +		git push pushee-namespaced master &&
> +		git ls-remote pushee-namespaced > actual &&
> +		printf "dc65a2e0f299dcc7efddbbe01641a28ee84329ba\trefs/heads/master\n" > expected &&

Could you avoid hardcoding the exact object names here?  Your script knows
what object should appear at refs/heads/master at "pushee-namespaced" (as
you have pushed from the repository "original" you are in), so it may be
something like:

	printf "%s\trefs/heads/mater\n" $(git rev-parse master) >expect

Same comment applies for all the other hardcoded object names.

> +test_expect_success 'mirroring a repository using a ref namespace' '
> +	git clone --mirror pushee mirror &&
> +	(
> +		cd mirror &&
> +		git for-each-ref refs/ > actual &&
> +		printf "dc65a2e0f299dcc7efddbbe01641a28ee84329ba commit\trefs/namespaces/namespace/refs/heads/master\n" > expected &&
> +		printf "fbdf4310c71b916568f04753f603fb24a0544227 commit\trefs/namespaces/namespace/refs/tags/0\n" >> expected &&
> +		printf "dc65a2e0f299dcc7efddbbe01641a28ee84329ba commit\trefs/namespaces/namespace/refs/tags/1\n" >> expected &&
> +		test_cmp expected actual
> +	)
> +'

I am not sure what you are trying to test. "pushee" is pretending to be a
hosting site that uses the namespace feature to house refs pushed from
original in refs/namespaces/namespace/ so it is expected to have these
refs under there.  You didn't make any "git remote" configuration in
either mirror nor pushee, so it is natural with or without the namespace
feature that "git clone --mirror" would find them at the same place.

What hasn't been tested in the above is to see what actual refs pushee has
with (cd pushee && git for-each-ref), and you could argue that this test
is a proxy for that, but then you are assuming that "clone --mirror" is
not broken, which means it would make debugging harder when this test does
start failing---is it the basic namespace feature, or is it mirror cloning
that acquired a bug to break this test?

Also I would have expected that some configuration that affects the
central repository side (i.e. "pushee") to cause unsuspecting clients
(e.g. "original" that pushes into it) to see only a subset of refs
"pushee" has, which would be the real motivation for this feature. To show
that off, we may also need tests that house two or more namespaces hosted
in the same repository.

Re: [PATCH] ref namespaces: tests

[Josh Triplett]

On Thu, Jul 14, 2011 at 04:13:48PM -0700, Junio C Hamano wrote:
> Josh Triplett <josh@joshtriplett.org> writes:
> >  create mode 100755 t/t5502-fetch-push-namespaces.sh
> 
> Isn't 5502 used already?

Argh.  Yes; I put it right after t5501-fetch-push-alternates.sh without
noticing that 5502 already existed.  I'll fix it to use t5507.  Thanks
for catching that.

> > +test_expect_success setup '
> > +	test_tick &&
> > +	git init original &&
> > +	(
> > +		cd original &&
> > +		i=0 &&
> > +		while [ "$i" -lt 2 ]
> > +		do
> > +			echo "$i" > count &&
> 
> This is just style, but the test scripts prefer to spell these like this:
> 
> 	while test "$i" -lt 2
> 	do
>         	echo "$i" >count &&
>                 ...
> 
> to favor "test" over "[ ... ]", and omit SP between ">" redirection (or
> "<" for that matter) and the filename.

Will do.  I had done a quick grep-survey of the tests to check usage of
test versus [, and saw enough of both to assume it didn't matter, but it
hadn't occurred to me to check CodingGuidelines for shell scripts; I now
see that it has a section specifically on shell scripting.  I'll fix
this in the next version.

Actually, I plan to unroll this two-iteration loop in the next version,
so that I can capture the two object hashes I need for use later in the
script.

Out of curiosity, what's the rationale for the use of test rather than
'['?  Just uniformity, or does test have some particular advantage over
'['?

> > +		git remote add pushee-namespaced "ext::git --namespace=namespace %s ../pushee" &&
> 
> Nice ;-).

Thanks. :)

> > +test_expect_success 'pushing into a repository using a ref namespace' '
> > +	(
> > +		cd original &&
> > +		git push pushee-namespaced master &&
> > +		git ls-remote pushee-namespaced > actual &&
> > +		printf "dc65a2e0f299dcc7efddbbe01641a28ee84329ba\trefs/heads/master\n" > expected &&
> 
> Could you avoid hardcoding the exact object names here?  Your script knows
> what object should appear at refs/heads/master at "pushee-namespaced" (as
> you have pushed from the repository "original" you are in), so it may be
> something like:
> 
> 	printf "%s\trefs/heads/mater\n" $(git rev-parse master) >expect
> 
> Same comment applies for all the other hardcoded object names.

I can do that; since the same two object hashes recur throughout the
script, I'll record them in shell variables up at the top.

> > +test_expect_success 'mirroring a repository using a ref namespace' '
> > +	git clone --mirror pushee mirror &&
> > +	(
> > +		cd mirror &&
> > +		git for-each-ref refs/ > actual &&
> > +		printf "dc65a2e0f299dcc7efddbbe01641a28ee84329ba commit\trefs/namespaces/namespace/refs/heads/master\n" > expected &&
> > +		printf "fbdf4310c71b916568f04753f603fb24a0544227 commit\trefs/namespaces/namespace/refs/tags/0\n" >> expected &&
> > +		printf "dc65a2e0f299dcc7efddbbe01641a28ee84329ba commit\trefs/namespaces/namespace/refs/tags/1\n" >> expected &&
> > +		test_cmp expected actual
> > +	)
> > +'
> 
> I am not sure what you are trying to test. "pushee" is pretending to be a
> hosting site that uses the namespace feature to house refs pushed from
> original in refs/namespaces/namespace/ so it is expected to have these
> refs under there.  You didn't make any "git remote" configuration in
> either mirror nor pushee, so it is natural with or without the namespace
> feature that "git clone --mirror" would find them at the same place.
> 
> What hasn't been tested in the above is to see what actual refs pushee has
> with (cd pushee && git for-each-ref), and you could argue that this test
> is a proxy for that, but then you are assuming that "clone --mirror" is
> not broken, which means it would make debugging harder when this test does
> start failing---is it the basic namespace feature, or is it mirror cloning
> that acquired a bug to break this test?

I wrote this test specifically to check for possible regressions in
clone or the machinery underneath it.  I wanted to ensure that no future
change caused clone to ignore refs in refs/namespaces/*.  In particular,
I want to protect against a regression caused by any future change to
the refs machinery that might cause it to ignore refs outside of
refs/heads/* or refs/tags/*, which might otherwise go un-noticed (as
they almost did during the development of this patchset, if not for an
incidental side effect of t5501).

If this test failed, I would expect that it would fail because clone
--mirror produced a mirrored repository which didn't actually contain
any refs, even though pushee contained the correctly namespaced refs;
thus, for-each-ref doesn't seem like the right test.

More generally, I also added this test because it tests a specific
high-level feature I care about: the ability to mirror a repository
containing namespaces using clone --mirror, and preserve those
namespaces.  I plan to use that as a backup mechanism, and I want it to
continue working. :)

- Josh Triplett

[PATCHv2] ref namespaces: tests

[Josh Triplett]

Test pushing, pulling, and mirroring of repositories with ref
namespaces.

Signed-off-by: Josh Triplett <josh@joshtriplett.org>
Signed-off-by: Jamey Sharp <jamey@minilop.net>
---

v2: Incorporate feedback from Junio: change test number to 5507 to avoid
conflicting with existing tests, make style consistent with other test
scripts, avoid hardcoding commit hashes, add rationale for mirror test.

 t/t5507-fetch-push-namespaces.sh |   85 ++++++++++++++++++++++++++++++++++++++
 1 files changed, 85 insertions(+), 0 deletions(-)
 create mode 100755 t/t5507-fetch-push-namespaces.sh

diff --git a/t/t5507-fetch-push-namespaces.sh b/t/t5507-fetch-push-namespaces.sh
new file mode 100755
index 0000000..cc0b31f
--- /dev/null
+++ b/t/t5507-fetch-push-namespaces.sh
@@ -0,0 +1,85 @@
+#!/bin/sh
+
+test_description='fetch/push involving ref namespaces'
+. ./test-lib.sh
+
+test_expect_success setup '
+	test_tick &&
+	git init original &&
+	(
+		cd original &&
+		echo 0 >count &&
+		git add count &&
+		test_commit 0 &&
+		echo 1 >count &&
+		git add count &&
+		test_commit 1 &&
+		git remote add pushee-namespaced "ext::git --namespace=namespace %s ../pushee" &&
+		git remote add pushee-unnamespaced ../pushee
+	) &&
+	commit0=$(cd original && git rev-parse HEAD^) &&
+	commit1=$(cd original && git rev-parse HEAD) &&
+	git init pushee &&
+	git init puller
+'
+
+test_expect_success 'pushing into a repository using a ref namespace' '
+	(
+		cd original &&
+		git push pushee-namespaced master &&
+		git ls-remote pushee-namespaced >actual &&
+		printf "$commit1\trefs/heads/master\n" >expected &&
+		test_cmp expected actual &&
+		git push pushee-namespaced --tags &&
+		git ls-remote pushee-namespaced >actual &&
+		printf "$commit0\trefs/tags/0\n" >>expected &&
+		printf "$commit1\trefs/tags/1\n" >>expected &&
+		test_cmp expected actual &&
+		# Verify that the GIT_NAMESPACE environment variable works as well
+		GIT_NAMESPACE=namespace git ls-remote "ext::git %s ../pushee" >actual &&
+		test_cmp expected actual &&
+		# Verify that --namespace overrides GIT_NAMESPACE
+		GIT_NAMESPACE=garbage git ls-remote pushee-namespaced >actual &&
+		test_cmp expected actual &&
+		# Try a namespace with no content
+		git ls-remote "ext::git --namespace=garbage %s ../pushee" >actual &&
+		test_cmp /dev/null actual &&
+		git ls-remote pushee-unnamespaced >actual &&
+		sed -e "s|refs/|refs/namespaces/namespace/refs/|" expected >expected.unnamespaced &&
+		test_cmp expected.unnamespaced actual
+	)
+'
+
+test_expect_success 'pulling from a repository using a ref namespace' '
+	(
+		cd puller &&
+		git remote add -f pushee-namespaced "ext::git --namespace=namespace %s ../pushee" &&
+		git for-each-ref refs/ >actual &&
+		printf "$commit1 commit\trefs/remotes/pushee-namespaced/master\n" >expected &&
+		printf "$commit0 commit\trefs/tags/0\n" >>expected &&
+		printf "$commit1 commit\trefs/tags/1\n" >>expected &&
+		test_cmp expected actual
+	)
+'
+
+# This test with clone --mirror checks for possible regressions in clone
+# or the machinery underneath it. It ensures that no future change
+# causes clone to ignore refs in refs/namespaces/*. In particular, it
+# protects against a regression caused by any future change to the refs
+# machinery that might cause it to ignore refs outside of refs/heads/*
+# or refs/tags/*. More generally, this test also checks the high-level
+# functionality of using clone --mirror to back up a set of repos hosted
+# in the namespaces of a single repo.
+test_expect_success 'mirroring a repository using a ref namespace' '
+	git clone --mirror pushee mirror &&
+	(
+		cd mirror &&
+		git for-each-ref refs/ >actual &&
+		printf "$commit1 commit\trefs/namespaces/namespace/refs/heads/master\n" >expected &&
+		printf "$commit0 commit\trefs/namespaces/namespace/refs/tags/0\n" >>expected &&
+		printf "$commit1 commit\trefs/namespaces/namespace/refs/tags/1\n" >>expected &&
+		test_cmp expected actual
+	)
+'
+
+test_done
-- 
1.7.5.4

Re: [PATCH] ref namespaces: tests

[Junio C Hamano]

Junio C Hamano <gitster@pobox.com> writes:

>> +test_expect_success 'pushing into a repository using a ref namespace' '
>> +	(
>> +		cd original &&
>> +		git push pushee-namespaced master &&
>> +		git ls-remote pushee-namespaced > actual &&
>> +		printf "dc65a2e0f299dcc7efddbbe01641a28ee84329ba\trefs/heads/master\n" > expected &&
>
> Could you avoid hardcoding the exact object names here?  Your script knows
> what object should appear at refs/heads/master at "pushee-namespaced" (as
> you have pushed from the repository "original" you are in), so it may be
> something like:
>
> 	printf "%s\trefs/heads/mater\n" $(git rev-parse master) >expect
>
> Same comment applies for all the other hardcoded object names.

Just FYI, I did a sample fix-up for earlier tests in this patch and pushed
it out in 'pu'; hopefully it may save your time.

Thanks.

[PATCHv3] ref namespaces: tests

[Josh Triplett]

Test pushing, pulling, and mirroring of repositories with ref
namespaces.

Signed-off-by: Josh Triplett <josh@joshtriplett.org>
Signed-off-by: Jamey Sharp <jamey@minilop.net>
---

v3: Change test number to 5509 to avoid conflicts with new tests 5507
and 5508 in pu.

At this point I think we've incorporated all the outstanding feedback.
Does this test need any further changes to allow the ref-namespaces
branch to graduate to next?

(Also, for future reference, do you prefer to see later versions of
patches as replies to the previous version, as I've made this mail a
reply to PATCHv2, or do you prefer to see them as new threads?)

v2: Incorporate feedback from Junio: change test number to 5507 to avoid
conflicting with existing tests, make style consistent with other test
scripts, avoid hardcoding commit hashes, add rationale for mirror test.

 t/t5509-fetch-push-namespaces.sh |   85 ++++++++++++++++++++++++++++++++++++++
 1 files changed, 85 insertions(+), 0 deletions(-)
 create mode 100755 t/t5509-fetch-push-namespaces.sh

diff --git a/t/t5509-fetch-push-namespaces.sh b/t/t5509-fetch-push-namespaces.sh
new file mode 100755
index 0000000..cc0b31f
--- /dev/null
+++ b/t/t5509-fetch-push-namespaces.sh
@@ -0,0 +1,85 @@
+#!/bin/sh
+
+test_description='fetch/push involving ref namespaces'
+. ./test-lib.sh
+
+test_expect_success setup '
+	test_tick &&
+	git init original &&
+	(
+		cd original &&
+		echo 0 >count &&
+		git add count &&
+		test_commit 0 &&
+		echo 1 >count &&
+		git add count &&
+		test_commit 1 &&
+		git remote add pushee-namespaced "ext::git --namespace=namespace %s ../pushee" &&
+		git remote add pushee-unnamespaced ../pushee
+	) &&
+	commit0=$(cd original && git rev-parse HEAD^) &&
+	commit1=$(cd original && git rev-parse HEAD) &&
+	git init pushee &&
+	git init puller
+'
+
+test_expect_success 'pushing into a repository using a ref namespace' '
+	(
+		cd original &&
+		git push pushee-namespaced master &&
+		git ls-remote pushee-namespaced >actual &&
+		printf "$commit1\trefs/heads/master\n" >expected &&
+		test_cmp expected actual &&
+		git push pushee-namespaced --tags &&
+		git ls-remote pushee-namespaced >actual &&
+		printf "$commit0\trefs/tags/0\n" >>expected &&
+		printf "$commit1\trefs/tags/1\n" >>expected &&
+		test_cmp expected actual &&
+		# Verify that the GIT_NAMESPACE environment variable works as well
+		GIT_NAMESPACE=namespace git ls-remote "ext::git %s ../pushee" >actual &&
+		test_cmp expected actual &&
+		# Verify that --namespace overrides GIT_NAMESPACE
+		GIT_NAMESPACE=garbage git ls-remote pushee-namespaced >actual &&
+		test_cmp expected actual &&
+		# Try a namespace with no content
+		git ls-remote "ext::git --namespace=garbage %s ../pushee" >actual &&
+		test_cmp /dev/null actual &&
+		git ls-remote pushee-unnamespaced >actual &&
+		sed -e "s|refs/|refs/namespaces/namespace/refs/|" expected >expected.unnamespaced &&
+		test_cmp expected.unnamespaced actual
+	)
+'
+
+test_expect_success 'pulling from a repository using a ref namespace' '
+	(
+		cd puller &&
+		git remote add -f pushee-namespaced "ext::git --namespace=namespace %s ../pushee" &&
+		git for-each-ref refs/ >actual &&
+		printf "$commit1 commit\trefs/remotes/pushee-namespaced/master\n" >expected &&
+		printf "$commit0 commit\trefs/tags/0\n" >>expected &&
+		printf "$commit1 commit\trefs/tags/1\n" >>expected &&
+		test_cmp expected actual
+	)
+'
+
+# This test with clone --mirror checks for possible regressions in clone
+# or the machinery underneath it. It ensures that no future change
+# causes clone to ignore refs in refs/namespaces/*. In particular, it
+# protects against a regression caused by any future change to the refs
+# machinery that might cause it to ignore refs outside of refs/heads/*
+# or refs/tags/*. More generally, this test also checks the high-level
+# functionality of using clone --mirror to back up a set of repos hosted
+# in the namespaces of a single repo.
+test_expect_success 'mirroring a repository using a ref namespace' '
+	git clone --mirror pushee mirror &&
+	(
+		cd mirror &&
+		git for-each-ref refs/ >actual &&
+		printf "$commit1 commit\trefs/namespaces/namespace/refs/heads/master\n" >expected &&
+		printf "$commit0 commit\trefs/namespaces/namespace/refs/tags/0\n" >>expected &&
+		printf "$commit1 commit\trefs/namespaces/namespace/refs/tags/1\n" >>expected &&
+		test_cmp expected actual
+	)
+'
+
+test_done
-- 
1.7.5.4

Re: [PATCHv3] ref namespaces: tests

[Junio C Hamano]

Josh Triplett <josh@joshtriplett.org> writes:

> At this point I think we've incorporated all the outstanding feedback.
> Does this test need any further changes to allow the ref-namespaces
> branch to graduate to next?

No more nitpicks from me on this patch at least for now.

Are people who expressed concern during the review on the previous round
of the series happy with the second round? I recall there was a strong
sentiment that it is regrettable that the series specifically changes
fetch and push and is not a more general mechanism. Personally I am OK
with the approach taken by this series, as I do not offhand think of other
ways to serve a modified namespace. You have to view the unaltered reality
when interacting with your own refs to enumerate the objects you have,
while giving the altered view to your clients that is limited to the
"virtual" space.

> (Also, for future reference, do you prefer to see later versions of
> patches as replies to the previous version, as I've made this mail a
> reply to PATCHv2, or do you prefer to see them as new threads?)

Often it is very convenient to be able to go "up" in the thread to re-read
the discussions in the previous round. On the other hand it sometimes gets
inconvenient when viewing tons of threads to have a beginning of a new
round buried deep in other threads. Referring to the messages in the
previous round by their message-id (or thread.gmane.org/ URL) in the body
of the patch message below "---" lines may solve both issues, but I can go
either way.

Re: [PATCHv3] ref namespaces: tests

[Jeff King]

On Thu, Jul 21, 2011 at 02:56:16PM -0700, Junio C Hamano wrote:

> Josh Triplett <josh@joshtriplett.org> writes:
> 
> > At this point I think we've incorporated all the outstanding feedback.
> > Does this test need any further changes to allow the ref-namespaces
> > branch to graduate to next?
> 
> No more nitpicks from me on this patch at least for now.
> 
> Are people who expressed concern during the review on the previous round
> of the series happy with the second round? I recall there was a strong
> sentiment that it is regrettable that the series specifically changes
> fetch and push and is not a more general mechanism. Personally I am OK
> with the approach taken by this series, as I do not offhand think of other
> ways to serve a modified namespace. You have to view the unaltered reality
> when interacting with your own refs to enumerate the objects you have,
> while giving the altered view to your clients that is limited to the
> "virtual" space.

I was one of the people who wanted to have namespaces or virtual repos
at a more fundamental level. However, I tried to do a relatively simple
patch when the discussion started, and ended up getting mired in corner
cases. And it sounds like Josh and Jamey made a good faith effort in
that direction, but still ended up where they are now. So I'm willing to
accept that it is not as simple as we hoped, and the more practical
approach from their series is acceptable.

As for the code itself, I admit I haven't been paying all that close
attention. I can try to give a more careful review if we want another
set of eyes.

-Peff