[PATCH] push: respect --no-thin

[PATCH] push: respect --no-thin

[Nguyễn Thái Ngọc Duy]

Over the time the default value for --thin has been switched between
on and off. As of now it's always on, even if --no-thin is given.
Correct the code to respect --no-thin.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
---
 builtin/push.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/builtin/push.c b/builtin/push.c
index 04f0eaf..333a1fb 100644
--- a/builtin/push.c
+++ b/builtin/push.c
@@ -15,7 +15,7 @@ static const char * const push_usage[] = {
 	NULL,
 };
 
-static int thin;
+static int thin = 1;
 static int deleterefs;
 static const char *receivepack;
 static int verbosity;
@@ -313,8 +313,7 @@ static int push_with_options(struct transport *transport, int flags)
 	if (receivepack)
 		transport_set_option(transport,
 				     TRANS_OPT_RECEIVEPACK, receivepack);
-	if (thin)
-		transport_set_option(transport, TRANS_OPT_THIN, "yes");
+	transport_set_option(transport, TRANS_OPT_THIN, thin ? "yes" : NULL);
 
 	if (verbosity > 0)
 		fprintf(stderr, _("Pushing to %s\n"), transport->url);
-- 
1.8.2.83.gc99314b

Re: [PATCH] push: respect --no-thin

[Jeff King]

On Sat, Aug 10, 2013 at 05:10:07PM +0700, Nguyen Thai Ngoc Duy wrote:

> Over the time the default value for --thin has been switched between
> on and off. As of now it's always on, even if --no-thin is given.
> Correct the code to respect --no-thin.
> 
> Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>

This makes sense, and I confirmed with the script I posted earlier that
it does work (and didn't before). Thanks.

It might be nice to have a test, but I am not sure of a good way to
check whether a push was thin. I guess we can grep the output of
GIT_TRACE, though it feels a bit hacky.

-Peff

Re: [PATCH] push: respect --no-thin

[Duy Nguyen]

On Sat, Aug 10, 2013 at 06:31:23AM -0400, Jeff King wrote:
> It might be nice to have a test, but I am not sure of a good way to
> check whether a push was thin. I guess we can grep the output of
> GIT_TRACE, though it feels a bit hacky.

In theory, if we have a way to tell receive-pack not to pass
--fix-thin to index-pack, then it should barf on the thin pack
received from push/send-pack. But it does not work this way, or at
least I set up the test the wrong way, or pack-objects decides not to
create deltas.. anybody sees what's wrong in this test?

-- 8< --
diff --git a/builtin/receive-pack.c b/builtin/receive-pack.c
index e3eb5fc..da60817 100644
--- a/builtin/receive-pack.c
+++ b/builtin/receive-pack.c
@@ -38,6 +38,7 @@ static int quiet;
 static int prefer_ofs_delta = 1;
 static int auto_update_server_info;
 static int auto_gc = 1;
+static int fix_thin = 1;
 static const char *head_name;
 static void *head_name_to_free;
 static int sent_capabilities;
@@ -869,7 +870,8 @@ static const char *unpack(int err_fd)
 		keeper[i++] = "--stdin";
 		if (fsck_objects)
 			keeper[i++] = "--strict";
-		keeper[i++] = "--fix-thin";
+		if (fix_thin)
+			keeper[i++] = "--fix-thin";
 		keeper[i++] = hdr_arg;
 		keeper[i++] = keep_arg;
 		keeper[i++] = NULL;
@@ -975,6 +977,10 @@ int cmd_receive_pack(int argc, const char **argv, const char *prefix)
 				stateless_rpc = 1;
 				continue;
 			}
+			if (!strcmp(arg, "--no-thin")) {
+				fix_thin = 0;
+				continue;
+			}
 
 			usage(receive_pack_usage);
 		}
diff --git a/t/t5516-fetch-push.sh b/t/t5516-fetch-push.sh
index 4691d51..9c9d43c 100755
--- a/t/t5516-fetch-push.sh
+++ b/t/t5516-fetch-push.sh
@@ -126,6 +126,16 @@ test_expect_success setup '
 
 '
 
+test_expect_success 'push --no-thin' '
+	git init no-thin &&
+	git --git-dir=no-thin/.git config receive.unpacklimit 0 &&
+	git push no-thin/.git refs/heads/master:refs/heads/foo &&
+	echo modified >> path1 &&
+	git commit -am modified &&
+	git push --no-thin --receive-pack="git receive-pack --no-thin" no-thin/.git refs/heads/master:refs/heads/foo
+	false
+'
+
 test_expect_success 'fetch without wildcard' '
 	mk_empty testrepo &&
 	(
-- 8< --

[PATCH v2] push: respect --no-thin

[Nguyễn Thái Ngọc Duy]

Over the time the default value for --thin has been switched between
on and off. As of now it's always on, even if --no-thin is given.
Correct the code to respect --no-thin.

receive-pack learns about --no-thin only for testing purposes, hence
no document update.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
---
 v2 adds a test for push --no-thin. I figured out the problem in my
 test. I hit size filtering heuristics in pack-objects.c:try_delta().

 builtin/push.c         |  5 ++---
 builtin/receive-pack.c |  8 +++++++-
 t/t5516-fetch-push.sh  | 12 ++++++++++++
 3 files changed, 21 insertions(+), 4 deletions(-)

diff --git a/builtin/push.c b/builtin/push.c
index 04f0eaf..333a1fb 100644
--- a/builtin/push.c
+++ b/builtin/push.c
@@ -15,7 +15,7 @@ static const char * const push_usage[] = {
 	NULL,
 };
 
-static int thin;
+static int thin = 1;
 static int deleterefs;
 static const char *receivepack;
 static int verbosity;
@@ -313,8 +313,7 @@ static int push_with_options(struct transport *transport, int flags)
 	if (receivepack)
 		transport_set_option(transport,
 				     TRANS_OPT_RECEIVEPACK, receivepack);
-	if (thin)
-		transport_set_option(transport, TRANS_OPT_THIN, "yes");
+	transport_set_option(transport, TRANS_OPT_THIN, thin ? "yes" : NULL);
 
 	if (verbosity > 0)
 		fprintf(stderr, _("Pushing to %s\n"), transport->url);
diff --git a/builtin/receive-pack.c b/builtin/receive-pack.c
index e3eb5fc..da60817 100644
--- a/builtin/receive-pack.c
+++ b/builtin/receive-pack.c
@@ -38,6 +38,7 @@ static int quiet;
 static int prefer_ofs_delta = 1;
 static int auto_update_server_info;
 static int auto_gc = 1;
+static int fix_thin = 1;
 static const char *head_name;
 static void *head_name_to_free;
 static int sent_capabilities;
@@ -869,7 +870,8 @@ static const char *unpack(int err_fd)
 		keeper[i++] = "--stdin";
 		if (fsck_objects)
 			keeper[i++] = "--strict";
-		keeper[i++] = "--fix-thin";
+		if (fix_thin)
+			keeper[i++] = "--fix-thin";
 		keeper[i++] = hdr_arg;
 		keeper[i++] = keep_arg;
 		keeper[i++] = NULL;
@@ -975,6 +977,10 @@ int cmd_receive_pack(int argc, const char **argv, const char *prefix)
 				stateless_rpc = 1;
 				continue;
 			}
+			if (!strcmp(arg, "--no-thin")) {
+				fix_thin = 0;
+				continue;
+			}
 
 			usage(receive_pack_usage);
 		}
diff --git a/t/t5516-fetch-push.sh b/t/t5516-fetch-push.sh
index 4691d51..e80247b 100755
--- a/t/t5516-fetch-push.sh
+++ b/t/t5516-fetch-push.sh
@@ -1172,4 +1172,16 @@ test_expect_success 'push --follow-tag only pushes relevant tags' '
 	test_cmp expect actual
 '
 
+test_expect_success 'push --no-thin must produce non-thin pack' '
+	for i in `seq 100`; do echo modified >> path1; done &&
+	git commit -am "path1 big enough to pass size heuristics in try_delta" &&
+	git init no-thin &&
+	git --git-dir=no-thin/.git config receive.unpacklimit 0 &&
+	git push no-thin/.git refs/heads/master:refs/heads/foo &&
+	echo modified >> path1 &&
+	git commit -am modified &&
+	git repack -adf &&
+	git push --no-thin --receive-pack="git receive-pack --no-thin" no-thin/.git refs/heads/master:refs/heads/foo
+'
+
 test_done
-- 
1.8.2.83.gc99314b

Re: [PATCH v2] push: respect --no-thin

[Eric Sunshine]

On Sat, Aug 10, 2013 at 9:24 PM, Nguyễn Thái Ngọc Duy <pclouds@gmail.com> wrote:
> Over the time the default value for --thin has been switched between
> on and off. As of now it's always on, even if --no-thin is given.
> Correct the code to respect --no-thin.
>
> receive-pack learns about --no-thin only for testing purposes, hence
> no document update.
>
> Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
> ---
> diff --git a/t/t5516-fetch-push.sh b/t/t5516-fetch-push.sh
> index 4691d51..e80247b 100755
> --- a/t/t5516-fetch-push.sh
> +++ b/t/t5516-fetch-push.sh
> @@ -1172,4 +1172,16 @@ test_expect_success 'push --follow-tag only pushes relevant tags' '
>         test_cmp expect actual
>  '
>
> +test_expect_success 'push --no-thin must produce non-thin pack' '
> +       for i in `seq 100`; do echo modified >> path1; done &&

s/seq/test_seq/

d17cf5f3a32f07bf (tests: Introduce test_seq;  2012-08-03)

> +       git commit -am "path1 big enough to pass size heuristics in try_delta" &&
> +       git init no-thin &&
> +       git --git-dir=no-thin/.git config receive.unpacklimit 0 &&
> +       git push no-thin/.git refs/heads/master:refs/heads/foo &&
> +       echo modified >> path1 &&
> +       git commit -am modified &&
> +       git repack -adf &&
> +       git push --no-thin --receive-pack="git receive-pack --no-thin" no-thin/.git refs/heads/master:refs/heads/foo
> +'
> +
>  test_done
> --
> 1.8.2.83.gc99314b

[PATCH v3] push: respect --no-thin

[Nguyễn Thái Ngọc Duy]

Over the time the default value for --thin has been switched between
on and off. As of now it's always on, even if --no-thin is given.
Correct the code to respect --no-thin.

receive-pack learns about --no-thin only for testing purposes, hence
no document update.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
---
 v3 gets rid of seq in the test.

 builtin/push.c         |  5 ++---
 builtin/receive-pack.c |  8 +++++++-
 t/t5516-fetch-push.sh  | 16 ++++++++++++++++
 3 files changed, 25 insertions(+), 4 deletions(-)

diff --git a/builtin/push.c b/builtin/push.c
index 04f0eaf..333a1fb 100644
--- a/builtin/push.c
+++ b/builtin/push.c
@@ -15,7 +15,7 @@ static const char * const push_usage[] = {
 	NULL,
 };
 
-static int thin;
+static int thin = 1;
 static int deleterefs;
 static const char *receivepack;
 static int verbosity;
@@ -313,8 +313,7 @@ static int push_with_options(struct transport *transport, int flags)
 	if (receivepack)
 		transport_set_option(transport,
 				     TRANS_OPT_RECEIVEPACK, receivepack);
-	if (thin)
-		transport_set_option(transport, TRANS_OPT_THIN, "yes");
+	transport_set_option(transport, TRANS_OPT_THIN, thin ? "yes" : NULL);
 
 	if (verbosity > 0)
 		fprintf(stderr, _("Pushing to %s\n"), transport->url);
diff --git a/builtin/receive-pack.c b/builtin/receive-pack.c
index e3eb5fc..da60817 100644
--- a/builtin/receive-pack.c
+++ b/builtin/receive-pack.c
@@ -38,6 +38,7 @@ static int quiet;
 static int prefer_ofs_delta = 1;
 static int auto_update_server_info;
 static int auto_gc = 1;
+static int fix_thin = 1;
 static const char *head_name;
 static void *head_name_to_free;
 static int sent_capabilities;
@@ -869,7 +870,8 @@ static const char *unpack(int err_fd)
 		keeper[i++] = "--stdin";
 		if (fsck_objects)
 			keeper[i++] = "--strict";
-		keeper[i++] = "--fix-thin";
+		if (fix_thin)
+			keeper[i++] = "--fix-thin";
 		keeper[i++] = hdr_arg;
 		keeper[i++] = keep_arg;
 		keeper[i++] = NULL;
@@ -975,6 +977,10 @@ int cmd_receive_pack(int argc, const char **argv, const char *prefix)
 				stateless_rpc = 1;
 				continue;
 			}
+			if (!strcmp(arg, "--no-thin")) {
+				fix_thin = 0;
+				continue;
+			}
 
 			usage(receive_pack_usage);
 		}
diff --git a/t/t5516-fetch-push.sh b/t/t5516-fetch-push.sh
index 4691d51..3cfd1cd 100755
--- a/t/t5516-fetch-push.sh
+++ b/t/t5516-fetch-push.sh
@@ -1172,4 +1172,20 @@ test_expect_success 'push --follow-tag only pushes relevant tags' '
 	test_cmp expect actual
 '
 
+test_expect_success 'push --no-thin must produce non-thin pack' '
+	cat >>path1 <<\EOF &&
+keep base version of path1 big enough, compared to the new changes
+later, in order to pass size heuristics in
+builtin/pack-objects.c:try_delta()
+EOF
+	git commit -am initial &&
+	git init no-thin &&
+	git --git-dir=no-thin/.git config receive.unpacklimit 0 &&
+	git push no-thin/.git refs/heads/master:refs/heads/foo &&
+	echo modified >> path1 &&
+	git commit -am modified &&
+	git repack -adf &&
+	git push --no-thin --receive-pack="git receive-pack --no-thin" no-thin/.git refs/heads/master:refs/heads/foo
+'
+
 test_done
-- 
1.8.2.83.gc99314b

Re: [PATCH v3] push: respect --no-thin

[Junio C Hamano]

Nguyễn Thái Ngọc Duy  <pclouds@gmail.com> writes:

> Over the time the default value for --thin has been switched between
> on and off. As of now it's always on, even if --no-thin is given.
> Correct the code to respect --no-thin.
>
> receive-pack learns about --no-thin only for testing purposes, hence
> no document update.

Please name it "--reject-thin-pack-for-testing" to make it crystal
clear that a documentation patch to "document undocumented option"
is unwanted.

> diff --git a/builtin/push.c b/builtin/push.c
> index 04f0eaf..333a1fb 100644
> --- a/builtin/push.c
> +++ b/builtin/push.c
> @@ -15,7 +15,7 @@ static const char * const push_usage[] = {
>  	NULL,
>  };
>  
> -static int thin;
> +static int thin = 1;
>  static int deleterefs;
>  static const char *receivepack;
>  static int verbosity;
> @@ -313,8 +313,7 @@ static int push_with_options(struct transport *transport, int flags)
>  	if (receivepack)
>  		transport_set_option(transport,
>  				     TRANS_OPT_RECEIVEPACK, receivepack);
> -	if (thin)
> -		transport_set_option(transport, TRANS_OPT_THIN, "yes");
> +	transport_set_option(transport, TRANS_OPT_THIN, thin ? "yes" : NULL);
>  
>  	if (verbosity > 0)
>  		fprintf(stderr, _("Pushing to %s\n"), transport->url);

Hmm, I am utterly confused.

How does the original code have thin as an non-overridable default?
The variable is initialized to 0, parse-options specifies it as
OPT_BOOLEAN, and TRANS_OPT_THIN is set only if "thin" is set.

Updated code flips the default to "1" but unconditionally uses
TRANS_OPT_THIN to set it to either "yes" or NULL.  While it is not
wrong per-se, do_push() (i.e. the caller of this function) grabs the
transport from transport_get() which initializes the transport with
the thin option disabled by default, so it is not immediately
obvious to me why "always call TRANS_OPT_THIN but set it explicitly
to NULL when --no-thin is asked" is necessary or improvement.

Puzzled....

Re: [PATCH v3] push: respect --no-thin

[Duy Nguyen]

On Mon, Aug 12, 2013 at 12:59 PM, Junio C Hamano <gitster@pobox.com> wrote:
>> @@ -15,7 +15,7 @@ static const char * const push_usage[] = {
>>       NULL,
>>  };
>>
>> -static int thin;
>> +static int thin = 1;
>>  static int deleterefs;
>>  static const char *receivepack;
>>  static int verbosity;
>> @@ -313,8 +313,7 @@ static int push_with_options(struct transport *transport, int flags)
>>       if (receivepack)
>>               transport_set_option(transport,
>>                                    TRANS_OPT_RECEIVEPACK, receivepack);
>> -     if (thin)
>> -             transport_set_option(transport, TRANS_OPT_THIN, "yes");
>> +     transport_set_option(transport, TRANS_OPT_THIN, thin ? "yes" : NULL);
>>
>>       if (verbosity > 0)
>>               fprintf(stderr, _("Pushing to %s\n"), transport->url);
>
> Hmm, I am utterly confused.
>
> How does the original code have thin as an non-overridable default?
> The variable is initialized to 0, parse-options specifies it as
> OPT_BOOLEAN, and TRANS_OPT_THIN is set only if "thin" is set.
>
> Updated code flips the default to "1" but unconditionally uses
> TRANS_OPT_THIN to set it to either "yes" or NULL.  While it is not
> wrong per-se, do_push() (i.e. the caller of this function) grabs the
> transport from transport_get() which initializes the transport with
> the thin option disabled by default,

transport_get() actually sets thin option to 1 by default. If I don't
misread the code, the first version of transport.c already flipped
"thin" from 0 (in push.c) to 1 (in transport.c), see 9b28851 (Push
code for transport library - 2007-09-10). The funny thing is that
commit was just one day after Shawn flipped 'thin' from 1 to 0 in
push.c in a4503a1.

> so it is not immediately
> obvious to me why "always call TRANS_OPT_THIN but set it explicitly
> to NULL when --no-thin is asked" is necessary or improvement.
-- 
Duy

[PATCH v4] push: respect --no-thin

[Nguyễn Thái Ngọc Duy]

- From the beginning of push.c in 755225d, 2006-04-29, "thin" option
  was enabled by default but could be turned off with --no-thin.

- Then Shawn changed the default to 0 in favor of saving server
  resources in a4503a1, 2007-09-09. --no-thin worked great.

- One day later, in 9b28851 Daniel extracted some code from push.c to
  create transport.c. He (probably accidentally) flipped the default
  value from 0 to 1 in transport_get().

From then on --no-thin is effectively no-op because git-push still
expects the default value to be false and only calls
transport_set_option() when "thin" variable in push.c is true (which
is unnecessary). Correct the code to respect --no-thin by calling
transport_set_option() in both cases.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
---
 v4 elaborates a bit on the story of "thin" variable in push.c and
 renames receive-pack --no-thin to --reject-thin-pack-for-testing.

 builtin/push.c         |  5 ++---
 builtin/receive-pack.c |  8 +++++++-
 t/t5516-fetch-push.sh  | 17 +++++++++++++++++
 3 files changed, 26 insertions(+), 4 deletions(-)

diff --git a/builtin/push.c b/builtin/push.c
index 04f0eaf..333a1fb 100644
--- a/builtin/push.c
+++ b/builtin/push.c
@@ -15,7 +15,7 @@ static const char * const push_usage[] = {
 	NULL,
 };
 
-static int thin;
+static int thin = 1;
 static int deleterefs;
 static const char *receivepack;
 static int verbosity;
@@ -313,8 +313,7 @@ static int push_with_options(struct transport *transport, int flags)
 	if (receivepack)
 		transport_set_option(transport,
 				     TRANS_OPT_RECEIVEPACK, receivepack);
-	if (thin)
-		transport_set_option(transport, TRANS_OPT_THIN, "yes");
+	transport_set_option(transport, TRANS_OPT_THIN, thin ? "yes" : NULL);
 
 	if (verbosity > 0)
 		fprintf(stderr, _("Pushing to %s\n"), transport->url);
diff --git a/builtin/receive-pack.c b/builtin/receive-pack.c
index e3eb5fc..fc6d53a 100644
--- a/builtin/receive-pack.c
+++ b/builtin/receive-pack.c
@@ -38,6 +38,7 @@ static int quiet;
 static int prefer_ofs_delta = 1;
 static int auto_update_server_info;
 static int auto_gc = 1;
+static int fix_thin = 1;
 static const char *head_name;
 static void *head_name_to_free;
 static int sent_capabilities;
@@ -869,7 +870,8 @@ static const char *unpack(int err_fd)
 		keeper[i++] = "--stdin";
 		if (fsck_objects)
 			keeper[i++] = "--strict";
-		keeper[i++] = "--fix-thin";
+		if (fix_thin)
+			keeper[i++] = "--fix-thin";
 		keeper[i++] = hdr_arg;
 		keeper[i++] = keep_arg;
 		keeper[i++] = NULL;
@@ -975,6 +977,10 @@ int cmd_receive_pack(int argc, const char **argv, const char *prefix)
 				stateless_rpc = 1;
 				continue;
 			}
+			if (!strcmp(arg, "--reject-thin-pack-for-testing")) {
+				fix_thin = 0;
+				continue;
+			}
 
 			usage(receive_pack_usage);
 		}
diff --git a/t/t5516-fetch-push.sh b/t/t5516-fetch-push.sh
index 4691d51..99c32d7 100755
--- a/t/t5516-fetch-push.sh
+++ b/t/t5516-fetch-push.sh
@@ -1172,4 +1172,21 @@ test_expect_success 'push --follow-tag only pushes relevant tags' '
 	test_cmp expect actual
 '
 
+test_expect_success 'push --no-thin must produce non-thin pack' '
+	cat >>path1 <<\EOF &&
+keep base version of path1 big enough, compared to the new changes
+later, in order to pass size heuristics in
+builtin/pack-objects.c:try_delta()
+EOF
+	git commit -am initial &&
+	git init no-thin &&
+	git --git-dir=no-thin/.git config receive.unpacklimit 0 &&
+	git push no-thin/.git refs/heads/master:refs/heads/foo &&
+	echo modified >> path1 &&
+	git commit -am modified &&
+	git repack -adf &&
+	rcvpck="git receive-pack --reject-thin-pack-for-testing" &&
+	git push --no-thin --receive-pack="$rcvpck" no-thin/.git refs/heads/master:refs/heads/foo
+'
+
 test_done
-- 
1.8.2.83.gc99314b

Re: [PATCH v3] push: respect --no-thin

[Junio C Hamano]

Duy Nguyen <pclouds@gmail.com> writes:

> transport_get() actually sets thin option to 1 by default.

Yeah, I missed your message in the nearby thread.  It indeed does.