[PATCH 1/3 v2] Add support for external programs for handling native fetches

transport_get() can call transport_shim_init() to have list and
fetch-ref operations handled by running a separate program as:

 git shim-<something> <remote> [<url>]

This program then accepts, on its stdin, "list" and "fetch <hex>
<name>" commands; the former prints out a list of available refs and
either their hashes or what they are symrefs to, while the latter
fetches them into the local object database and prints a newline when done.

Signed-off-by: Daniel Barkalow <redacted>
---
This version uses the more traditional ordering for listing refs and fixes 
a memory leak.

 Documentation/git-shim.txt |   37 +++++++++++
 Makefile                   |    1 +
 transport-shim.c           |  142 ++++++++++++++++++++++++++++++++++++++++++++
 transport.h                |    3 +
 4 files changed, 183 insertions(+), 0 deletions(-)
 create mode 100644 Documentation/git-shim.txt
 create mode 100644 transport-shim.c

diff --git a/Documentation/git-shim.txt b/Documentation/git-shim.txt
new file mode 100644
index 0000000..dd80c6d
--- /dev/null
+++ b/Documentation/git-shim.txt

@@ -0,0 +1,37 @@
+git-shim(1)
+============
+
+NAME
+----
+git-shim - Helper programs for interoperation with remote git
+
+SYNOPSIS
+--------
+'git shim-<transport>' <remote>
+
+DESCRIPTION
+-----------
+
+These programs are normally not used directly by end users, but are
+invoked by various git programs that interact with remote repositories
+when the repository they would operate on will be accessed using
+transport code not linked into the main git binary.
+
+COMMANDS
+--------
+
+Commands are given by the caller on the helper's standard input, one per line.
+
+'list'::
+	Lists the refs, one per line, if the format "<value>
+	<name>". The value is either a hex sha1 hash or "@<dest>" for
+	symrefs. After the complete list, outputs a blank line.
+
+'fetch' ref::
+	Fetches the given ref, writing the necessary objects to the
+	database. Outputs a blank line when the fetch is complete.
++
+If a fatal error occurs, the program writes the error message to
+stderr and exits. The caller should expect that a suitable error
+message has been printed if the child closes the connection without
+completing a valid response for the current command.

diff --git a/Makefile b/Makefile
index bde27ed..01efc73 100644
--- a/Makefile
+++ b/Makefile

@@ -549,6 +549,7 @@ LIB_OBJS += symlinks.o
 LIB_OBJS += tag.o
 LIB_OBJS += trace.o
 LIB_OBJS += transport.o
+LIB_OBJS += transport-shim.o
 LIB_OBJS += tree-diff.o
 LIB_OBJS += tree.o
 LIB_OBJS += tree-walk.o

diff --git a/transport-shim.c b/transport-shim.c
new file mode 100644
index 0000000..2518aba
--- /dev/null
+++ b/transport-shim.c

@@ -0,0 +1,142 @@
+#include "cache.h"
+#include "transport.h"
+
+#include "run-command.h"
+#include "commit.h"
+#include "diff.h"
+#include "revision.h"
+
+struct shim_data
+{
+	const char *name;
+	struct child_process *shim;
+};
+
+static struct child_process *get_shim(struct transport *transport)
+{
+	struct shim_data *data = transport->data;
+	if (!data->shim) {
+		struct strbuf buf = STRBUF_INIT;
+		struct child_process *shim = xcalloc(1, sizeof(*shim));
+		shim->in = -1;
+		shim->out = -1;
+		shim->err = 0;
+		shim->argv = xcalloc(4, sizeof(*shim->argv));
+		strbuf_addf(&buf, "shim-%s", data->name);
+		shim->argv[0] = buf.buf;
+		shim->argv[1] = transport->remote->name;
+		shim->argv[2] = transport->url;
+		shim->git_cmd = 1;
+		start_command(shim);
+		data->shim = shim;
+	}
+	return data->shim;
+}
+
+static int disconnect_shim(struct transport *transport)
+{
+	struct shim_data *data = transport->data;
+	if (data->shim) {
+		write(data->shim->in, "\n", 1);
+		close(data->shim->in);
+		finish_command(data->shim);
+		free(data->shim->argv);
+		free(data->shim);
+		transport->data = NULL;
+	}
+	return 0;
+}
+
+static int fetch_refs_via_shim(struct transport *transport,
+			       int nr_heads, const struct ref **to_fetch)
+{
+	struct child_process *shim;
+	const struct ref *posn;
+	struct strbuf buf = STRBUF_INIT;
+	int i, count;
+	FILE *file;
+
+	count = 0;
+	for (i = 0; i < nr_heads; i++) {
+		posn = to_fetch[i];
+		if (posn->status & REF_STATUS_UPTODATE)
+			continue;
+		count++;
+	}
+
+	if (count) {
+		shim = get_shim(transport);
+		for (i = 0; i < nr_heads; i++) {
+			posn = to_fetch[i];
+			if (posn->status & REF_STATUS_UPTODATE)
+				continue;
+			write(shim->in, "fetch ", 6);
+			write(shim->in, sha1_to_hex(posn->old_sha1), 40);
+			write(shim->in, " ", 1);
+			write(shim->in, posn->name, strlen(posn->name));
+			write(shim->in, "\n", 1);
+		}
+		file = fdopen(shim->out, "r");
+		while (count) {
+			if (strbuf_getline(&buf, file, '\n') == EOF)
+				exit(128); /* child died, message supplied already */
+
+			count--;
+		}
+	}
+	return 0;
+}
+
+static struct ref *get_refs_via_shim(struct transport *transport, int for_push)
+{
+	struct child_process *shim;
+	struct ref *ret = NULL;
+	struct ref **end = &ret;
+	struct ref *posn;
+	struct strbuf buf = STRBUF_INIT;
+	FILE *file;
+
+	shim = get_shim(transport);
+	write(shim->in, "list\n", 5);
+
+	file = fdopen(shim->out, "r");
+	while (1) {
+		char *eov;
+		if (strbuf_getline(&buf, file, '\n') == EOF)
+			exit(128); /* child died, message supplied already */
+
+		if (!*buf.buf)
+			break;
+
+		eov = strchr(buf.buf, ' ');
+		if (!eov)
+			die("Malformed response in ref list: %s", buf.buf);
+		*eov = '\0';
+		*end = alloc_ref(eov + 1);
+		if (eov) {
+			if (buf.buf[0] == '@')
+				(*end)->symref = xstrdup(buf.buf + 1);
+			else
+				get_sha1_hex(buf.buf, (*end)->old_sha1);
+		}
+		end = &((*end)->next);
+		strbuf_reset(&buf);
+	}
+	strbuf_release(&buf);
+
+	for (posn = ret; posn; posn = posn->next)
+		resolve_remote_symref(posn, ret);
+
+	return ret;
+}
+
+void transport_shim_init(struct transport *transport, const char *name)
+{
+	struct shim_data *data = xmalloc(sizeof(*data));
+	data->shim = NULL;
+	data->name = name;
+	transport->data = data;
+	transport->get_refs_list = get_refs_via_shim;
+	transport->fetch = fetch_refs_via_shim;
+	transport->disconnect = disconnect_shim;
+}

diff --git a/transport.h b/transport.h
index 51b5397..01f650d 100644
--- a/transport.h
+++ b/transport.h

@@ -77,4 +77,7 @@ void transport_unlock_pack(struct transport *transport);
 int transport_disconnect(struct transport *transport);
 char *transport_anonymize_url(const char *url);
 
+/* Transport methods defined outside transport.c */
+void transport_shim_init(struct transport *transport, const char *name);
+
 #endif

-- 
1.6.4.rc2.13.ga9762.dirty

On Tue, Jul 28, 2009 at 08:08, Daniel Barkalow[off-list ref] wrote:

+static struct child_process *get_shim(struct transport *transport)
+{
+       struct shim_data *data = transport->data;
+       if (!data->shim) {
+               struct strbuf buf = STRBUF_INIT;
+               struct child_process *shim = xcalloc(1, sizeof(*shim));
+               shim->in = -1;
+               shim->out = -1;
+               shim->err = 0;
+               shim->argv = xcalloc(4, sizeof(*shim->argv));
+               strbuf_addf(&buf, "shim-%s", data->name);
+               shim->argv[0] = buf.buf;
+               shim->argv[1] = transport->remote->name;
+               shim->argv[2] = transport->url;
+               shim->git_cmd = 1;
+               start_command(shim);
+               data->shim = shim;
+       }
+       return data->shim;
+}
+
+static int disconnect_shim(struct transport *transport)
+{
+       struct shim_data *data = transport->data;
+       if (data->shim) {
+               write(data->shim->in, "\n", 1);
+               close(data->shim->in);
+               finish_command(data->shim);
+               free(data->shim->argv);

Does this leak data->shim->argv[0] (Ie "shim-%s")?

Bert

+               free(data->shim);
+               transport->data = NULL;
+       }
+       return 0;
+}

Hi,

On Tue, 28 Jul 2009, Daniel Barkalow wrote:

transport_get() can call transport_shim_init() to have list and
fetch-ref operations handled by running a separate program as:

As I commented already, "shim" is a meaningless word in the context of 
Git.  At _least_ call it something like "git-remote-<protocol>".  Even 
"git-fetch-<protocol>" would be better than "shim".

Actually, this is just one helper program, no?  Why can't it be integrated 
into transport.c?

At the very least, the text so far is misleading.

+COMMANDS
+--------
+
+Commands are given by the caller on the helper's standard input, one per line.
+
+'list'::
+	Lists the refs, one per line, if the format "<value>
+	<name>". The value is either a hex sha1 hash or "@<dest>" for
+	symrefs. After the complete list, outputs a blank line.
+
+'fetch' ref::
+	Fetches the given ref, writing the necessary objects to the
+	database. Outputs a blank line when the fetch is complete.
++

So you allow only 'list' and 'fetch'.  That is very limiting, and you do 
not even foresee a method to ask for the helper's capabilities.  We 
already saw how much pain that caused in the transport protocol, so please 
do not repeat the mistake here.

Why can't the caller check for this?  Would this not make much more sense 
to begin with?

+static int disconnect_shim(struct transport *transport)
+{
+	struct shim_data *data = transport->data;
+	if (data->shim) {
+		write(data->shim->in, "\n", 1);
+		close(data->shim->in);
+		finish_command(data->shim);
+		free(data->shim->argv);
+		free(data->shim);
+		transport->data = NULL;
+	}
+	return 0;
+}

Why is this function returning anything?

+static int fetch_refs_via_shim(struct transport *transport,
+			       int nr_heads, const struct ref **to_fetch)

Do you fetch only the refs, or also their objects?  If the latter, the 
name needs to be adjusted.

+{
+	struct child_process *shim;
+	const struct ref *posn;
+	struct strbuf buf = STRBUF_INIT;
+	int i, count;
+	FILE *file;
+
+	count = 0;
+	for (i = 0; i < nr_heads; i++) {
+		posn = to_fetch[i];
+		if (posn->status & REF_STATUS_UPTODATE)
+			continue;
+		count++;
+	}

This would be more readable IMO if it read like this:

	for (count = i = 0; i < nr_heads; i++)
		if (!(to_fetch[i]->status & REF_STATUS_UPTODATE))
			count++;

+	if (count) {
+		shim = get_shim(transport);

It would be much better to say "if (!count) return 0;" here rather than 
indenting a whole block of code, with no code after that.

+		for (i = 0; i < nr_heads; i++) {
+			posn = to_fetch[i];
+			if (posn->status & REF_STATUS_UPTODATE)
+				continue;
+			write(shim->in, "fetch ", 6);
+			write(shim->in, sha1_to_hex(posn->old_sha1), 40);
+			write(shim->in, " ", 1);
+			write(shim->in, posn->name, strlen(posn->name));
+			write(shim->in, "\n", 1);
+		}
+		file = fdopen(shim->out, "r");
+		while (count) {
+			if (strbuf_getline(&buf, file, '\n') == EOF)
+				exit(128); /* child died, message supplied already */
+
+			count--;

while (count--)

+		}
+	}
+	return 0;
+}
+
+static struct ref *get_refs_via_shim(struct transport *transport, int for_push)
+{
+	struct child_process *shim;
+	struct ref *ret = NULL;
+	struct ref **end = &ret;

A better name for this is "tail", as is used at least in many parts of the 
Git source code already.

+	struct ref *posn;
+	struct strbuf buf = STRBUF_INIT;
+	FILE *file;
+
+	shim = get_shim(transport);
+	write(shim->in, "list\n", 5);

What about the return value of this write()?  It can indicate error or 
short write.

+
+	file = fdopen(shim->out, "r");

No check for file != NULL?

+	while (1) {
+		char *eov;
+		if (strbuf_getline(&buf, file, '\n') == EOF)
+			exit(128); /* child died, message supplied already */
+
+		if (!*buf.buf)
+			break;
+
+		eov = strchr(buf.buf, ' ');
+		if (!eov)
+			die("Malformed response in ref list: %s", buf.buf);
+		*eov = '\0';
+		*end = alloc_ref(eov + 1);
+		if (eov) {

Did we not die 4 lines earlier if eov == NULL?

+			if (buf.buf[0] == '@')
+				(*end)->symref = xstrdup(buf.buf + 1);
+			else
+				get_sha1_hex(buf.buf, (*end)->old_sha1);

IMHO it is not at all clear what you are doing here.  At least a little 
hint is in order if the code does not explain itself.

+		}
+		end = &((*end)->next);
+		strbuf_reset(&buf);
+	}
+	strbuf_release(&buf);
+
+	for (posn = ret; posn; posn = posn->next)
+		resolve_remote_symref(posn, ret);
+
+	return ret;
+}

Ciao,
Dscho

On Tue, 28 Jul 2009, Bert Wesarg wrote:

On Tue, Jul 28, 2009 at 08:08, Daniel Barkalow[off-list ref] wrote:

quoted

+static struct child_process *get_shim(struct transport *transport)
+{
+       struct shim_data *data = transport->data;
+       if (!data->shim) {
+               struct strbuf buf = STRBUF_INIT;
+               struct child_process *shim = xcalloc(1, sizeof(*shim));
+               shim->in = -1;
+               shim->out = -1;
+               shim->err = 0;
+               shim->argv = xcalloc(4, sizeof(*shim->argv));
+               strbuf_addf(&buf, "shim-%s", data->name);
+               shim->argv[0] = buf.buf;
+               shim->argv[1] = transport->remote->name;
+               shim->argv[2] = transport->url;
+               shim->git_cmd = 1;
+               start_command(shim);
+               data->shim = shim;
+       }
+       return data->shim;
+}
+
+static int disconnect_shim(struct transport *transport)
+{
+       struct shim_data *data = transport->data;
+       if (data->shim) {
+               write(data->shim->in, "\n", 1);
+               close(data->shim->in);
+               finish_command(data->shim);
+               free(data->shim->argv);

Does this leak data->shim->argv[0] (Ie "shim-%s")?

Yes, I guess it would. I think it's time to look up how to run tests with 
valgrind...

Thanks.
	-Daniel
*This .sig left intentionally blank*

On Tue, 28 Jul 2009, Johannes Schindelin wrote:

Hi,

On Tue, 28 Jul 2009, Daniel Barkalow wrote:

quoted

transport_get() can call transport_shim_init() to have list and
fetch-ref operations handled by running a separate program as:

As I commented already, "shim" is a meaningless word in the context of 
Git.  At _least_ call it something like "git-remote-<protocol>".  Even 
"git-fetch-<protocol>" would be better than "shim".

I think git-remote-<protocol> is a good name; I'm not particularly tied to 
"shim", although I think this should be able to support push as well as 
fetch (that is, instead of http-push being a separate program that 
transport.c knows how to run, have that be handled by transport-side code 
that could support pushing with arbitrary protocols, and have the pushing 
code built into the curl handler). I do think that I want to form the 
command names differently depending on whether this is a helper that 
transports git objects to a remote git repository using a protocol that 
the helper is for, or a helper that interacts with a non-git remote 
repository where the helper is particular to the foreign scm.

quoted

diff --git a/Documentation/git-shim.txt b/Documentation/git-shim.txt
new file mode 100644
index 0000000..dd80c6d
--- /dev/null
+++ b/Documentation/git-shim.txt

@@ -0,0 +1,37 @@
+git-shim(1)
+============
+
+NAME
+----
+git-shim - Helper programs for interoperation with remote git

Actually, this is just one helper program, no?  Why can't it be integrated 
into transport.c?

No, this defines a pattern of helper programs, all of which should behave 
as documented by the same man page, so that the same code can run any of 
them.

quoted

+COMMANDS
+--------
+
+Commands are given by the caller on the helper's standard input, one per line.
+
+'list'::
+	Lists the refs, one per line, if the format "<value>
+	<name>". The value is either a hex sha1 hash or "@<dest>" for
+	symrefs. After the complete list, outputs a blank line.
+
+'fetch' ref::
+	Fetches the given ref, writing the necessary objects to the
+	database. Outputs a blank line when the fetch is complete.
++

So you allow only 'list' and 'fetch'.  That is very limiting, and you do 
not even foresee a method to ask for the helper's capabilities.  We 
already saw how much pain that caused in the transport protocol, so please 
do not repeat the mistake here.

I'll put the "capabilities" command back in; I dropped it from this subset 
of the foreign vcs helper protocol because there's not much variation 
possible when the remote repository must be a git repository.

quoted

diff --git a/transport-shim.c b/transport-shim.c
new file mode 100644
index 0000000..2518aba
--- /dev/null
+++ b/transport-shim.c

@@ -0,0 +1,142 @@
+#include "cache.h"
+#include "transport.h"
+
+#include "run-command.h"
+#include "commit.h"
+#include "diff.h"
+#include "revision.h"
+
+struct shim_data
+{
+	const char *name;
+	struct child_process *shim;
+};
+
+static struct child_process *get_shim(struct transport *transport)
+{
+	struct shim_data *data = transport->data;
+	if (!data->shim) {

Why can't the caller check for this?  Would this not make much more sense 
to begin with?

Depending on the order that the user of transport.c calls commands, the 
helper may or may not already be running (and it may or may not have been 
closed by the previous command, in the future extension of also handling 
importers which produce a fast-import stream). The caller could check for 
the connection already being constructed, but all callers would have to 
add the same check.

quoted

+static int disconnect_shim(struct transport *transport)
+{
+	struct shim_data *data = transport->data;
+	if (data->shim) {
+		write(data->shim->in, "\n", 1);
+		close(data->shim->in);
+		finish_command(data->shim);
+		free(data->shim->argv);
+		free(data->shim);
+		transport->data = NULL;
+	}
+	return 0;
+}

Why is this function returning anything?

Good point; we can't really fail to disconnect, and we've already 
determined that things haven't screwed up. It's left over from the pattern 
used by some of the native protocol code, where it may determine that the 
operation didn't work after all when the "I'm done" packet fails to 
produce the "goodbye" response.

quoted

+static int fetch_refs_via_shim(struct transport *transport,
+			       int nr_heads, const struct ref **to_fetch)

Do you fetch only the refs, or also their objects?  If the latter, the 
name needs to be adjusted.

This is inherited from the transport.h naming, which made sense years ago 
(when it contrasted fetching objects by name of ref versus fetching 
objects by hash); it is fetching objects as specified by struct refs.

quoted

+{
+	struct child_process *shim;
+	const struct ref *posn;
+	struct strbuf buf = STRBUF_INIT;
+	int i, count;
+	FILE *file;
+
+	count = 0;
+	for (i = 0; i < nr_heads; i++) {
+		posn = to_fetch[i];
+		if (posn->status & REF_STATUS_UPTODATE)
+			continue;
+		count++;
+	}

This would be more readable IMO if it read like this:

	for (count = i = 0; i < nr_heads; i++)
		if (!(to_fetch[i]->status & REF_STATUS_UPTODATE))
			count++;

I think it's more readable to match the flow control of the later loop. 
This is a dry run of the main loop, counting the number of times we 
don't skip the important part. I think your version is a more readable way 
of counting the number of not-up-to-date items, but it's not nearly so 
readable a way of calculating how many times we'll reach the interesting 
part of the loop further down.

quoted

+	if (count) {
+		shim = get_shim(transport);

It would be much better to say "if (!count) return 0;" here rather than 
indenting a whole block of code, with no code after that.

True.

quoted

+		for (i = 0; i < nr_heads; i++) {
+			posn = to_fetch[i];
+			if (posn->status & REF_STATUS_UPTODATE)
+				continue;
+			write(shim->in, "fetch ", 6);
+			write(shim->in, sha1_to_hex(posn->old_sha1), 40);
+			write(shim->in, " ", 1);
+			write(shim->in, posn->name, strlen(posn->name));
+			write(shim->in, "\n", 1);
+		}
+		file = fdopen(shim->out, "r");
+		while (count) {
+			if (strbuf_getline(&buf, file, '\n') == EOF)
+				exit(128); /* child died, message supplied already */
+
+			count--;

while (count--)

I don't like decrement operators used with values, except for the common 
string idioms. But I like:

	for (; count; count--)
		if ...

quoted

+		}
+	}
+	return 0;
+}
+
+static struct ref *get_refs_via_shim(struct transport *transport, int for_push)
+{
+	struct child_process *shim;
+	struct ref *ret = NULL;
+	struct ref **end = &ret;

A better name for this is "tail", as is used at least in many parts of the 
Git source code already.

True.

quoted

+	struct ref *posn;
+	struct strbuf buf = STRBUF_INIT;
+	FILE *file;
+
+	shim = get_shim(transport);
+	write(shim->in, "list\n", 5);

What about the return value of this write()?  It can indicate error or 
short write.

Yeah, I should use write_in_full(), although we know at this point that 
shim->in is a pipe which has been drained entirely.

quoted

+
+	file = fdopen(shim->out, "r");

No check for file != NULL?

I don't think this can fail, if setting up the child process didn't fail.

quoted

+	while (1) {
+		char *eov;
+		if (strbuf_getline(&buf, file, '\n') == EOF)
+			exit(128); /* child died, message supplied already */
+
+		if (!*buf.buf)
+			break;
+
+		eov = strchr(buf.buf, ' ');
+		if (!eov)
+			die("Malformed response in ref list: %s", buf.buf);
+		*eov = '\0';
+		*end = alloc_ref(eov + 1);
+		if (eov) {

Did we not die 4 lines earlier if eov == NULL?

Yeah, I confused myself while paring down the vcs output parsing code. I 
think I actually want to future-proof this code to allow the helper to add 
space-separated flags after the name, and I don't want this test.

quoted

+			if (buf.buf[0] == '@')
+				(*end)->symref = xstrdup(buf.buf + 1);
+			else
+				get_sha1_hex(buf.buf, (*end)->old_sha1);

IMHO it is not at all clear what you are doing here.  At least a little 
hint is in order if the code does not explain itself.

It's parsing the output of the "list" command, as given in the 
documentation. Item starting with a '@' is a symref, otherwise it is a 
sha1.

quoted

+		}
+		end = &((*end)->next);
+		strbuf_reset(&buf);
+	}
+	strbuf_release(&buf);
+
+	for (posn = ret; posn; posn = posn->next)
+		resolve_remote_symref(posn, ret);
+
+	return ret;
+}

Ciao,
Dscho

Hi,

On Tue, 28 Jul 2009, Daniel Barkalow wrote:

On Tue, 28 Jul 2009, Johannes Schindelin wrote:

quoted

On Tue, 28 Jul 2009, Daniel Barkalow wrote:

quoted

transport_get() can call transport_shim_init() to have list and
fetch-ref operations handled by running a separate program as:

As I commented already, "shim" is a meaningless word in the context of 
Git.  At _least_ call it something like "git-remote-<protocol>".  Even 
"git-fetch-<protocol>" would be better than "shim".

I think git-remote-<protocol> is a good name;

Good.

I think this should be able to support push as well as fetch

So you agree that it should be extensible.  Also good.

I do think that I want to form the command names differently depending 
on whether this is a helper that transports git objects to a remote git 
repository using a protocol that the helper is for, or a helper that 
interacts with a non-git remote repository where the helper is 
particular to the foreign scm.

I think that is a rather bad decision: you should not _need_ to care on 
the transport.c side.  If the helper handles fetching and/or pushing: 
fine.  How it accomplishes that: don't care.

quoted

quoted

diff --git a/Documentation/git-shim.txt b/Documentation/git-shim.txt
new file mode 100644
index 0000000..dd80c6d
--- /dev/null
+++ b/Documentation/git-shim.txt

@@ -0,0 +1,37 @@
+git-shim(1)
+============
+
+NAME
+----
+git-shim - Helper programs for interoperation with remote git

Actually, this is just one helper program, no?  Why can't it be integrated 
into transport.c?

No, this defines a pattern of helper programs, all of which should 
behave as documented by the same man page, so that the same code can run 
any of them.

See?  That description _is_ misleading.

quoted

quoted

+COMMANDS
+--------
+
+Commands are given by the caller on the helper's standard input, one per line.
+
+'list'::
+	Lists the refs, one per line, if the format "<value>
+	<name>". The value is either a hex sha1 hash or "@<dest>" for
+	symrefs. After the complete list, outputs a blank line.
+
+'fetch' ref::
+	Fetches the given ref, writing the necessary objects to the
+	database. Outputs a blank line when the fetch is complete.
++

So you allow only 'list' and 'fetch'.  That is very limiting, and you do 
not even foresee a method to ask for the helper's capabilities.  We 
already saw how much pain that caused in the transport protocol, so please 
do not repeat the mistake here.

I'll put the "capabilities" command back in; I dropped it from this 
subset of the foreign vcs helper protocol because there's not much 
variation possible when the remote repository must be a git repository.

Just because you do not foresee it now does not mean that it won't happen.

As I hinted at: it is rather likely that you do not foresee what kind of 
capability negotiation we'll want in the future.

Let's learn from history.

quoted

quoted

diff --git a/transport-shim.c b/transport-shim.c
new file mode 100644
index 0000000..2518aba
--- /dev/null
+++ b/transport-shim.c

@@ -0,0 +1,142 @@
+#include "cache.h"
+#include "transport.h"
+
+#include "run-command.h"
+#include "commit.h"
+#include "diff.h"
+#include "revision.h"
+
+struct shim_data
+{
+	const char *name;
+	struct child_process *shim;
+};
+
+static struct child_process *get_shim(struct transport *transport)
+{
+	struct shim_data *data = transport->data;
+	if (!data->shim) {

Why can't the caller check for this?  Would this not make much more sense 
to begin with?

Depending on the order that the user of transport.c calls commands, the 
helper may or may not already be running (and it may or may not have been 
closed by the previous command, in the future extension of also handling 
importers which produce a fast-import stream). The caller could check for 
the connection already being constructed, but all callers would have to 
add the same check.

Fair enough.  That large deeply indented part that makes up the most part 
of the function is very ugly, though.

quoted

quoted

+static int fetch_refs_via_shim(struct transport *transport,
+			       int nr_heads, const struct ref **to_fetch)

Do you fetch only the refs, or also their objects?  If the latter, the 
name needs to be adjusted.

This is inherited from the transport.h naming, which made sense years ago 
(when it contrasted fetching objects by name of ref versus fetching 
objects by hash); it is fetching objects as specified by struct refs.

It is actually not inherited from the naming, as the member you assign 
this function to does not contain the string "refs".

quoted

quoted

+{
+	struct child_process *shim;
+	const struct ref *posn;
+	struct strbuf buf = STRBUF_INIT;
+	int i, count;
+	FILE *file;
+
+	count = 0;
+	for (i = 0; i < nr_heads; i++) {
+		posn = to_fetch[i];
+		if (posn->status & REF_STATUS_UPTODATE)
+			continue;
+		count++;
+	}

This would be more readable IMO if it read like this:

	for (count = i = 0; i < nr_heads; i++)
		if (!(to_fetch[i]->status & REF_STATUS_UPTODATE))
			count++;

I think it's more readable to match the flow control of the later loop. 

Yeah, right.  It was so readable that I bothered to comment.
 

quoted

quoted

+		for (i = 0; i < nr_heads; i++) {
+			posn = to_fetch[i];
+			if (posn->status & REF_STATUS_UPTODATE)
+				continue;
+			write(shim->in, "fetch ", 6);
+			write(shim->in, sha1_to_hex(posn->old_sha1), 40);
+			write(shim->in, " ", 1);
+			write(shim->in, posn->name, strlen(posn->name));
+			write(shim->in, "\n", 1);
+		}
+		file = fdopen(shim->out, "r");
+		while (count) {
+			if (strbuf_getline(&buf, file, '\n') == EOF)
+				exit(128); /* child died, message supplied already */
+
+			count--;

while (count--)

I don't like decrement operators used with values, except for the common 
string idioms. But I like:

	for (; count; count--)
		if ...

Which is even more ugly than your current version.  Just my personal 
taste, of course.  Oh, and maybe also a little disagreement with the rest 
of Git's source.

quoted

quoted

+	struct ref *posn;
+	struct strbuf buf = STRBUF_INIT;
+	FILE *file;
+
+	shim = get_shim(transport);
+	write(shim->in, "list\n", 5);

What about the return value of this write()?  It can indicate error or 
short write.

Yeah, I should use write_in_full(), although we know at this point that 
shim->in is a pipe which has been drained entirely.

I am not at all sure that you can guarantee that at this point.

quoted

quoted

+	file = fdopen(shim->out, "r");

No check for file != NULL?

I don't think this can fail, if setting up the child process didn't fail.

Until somebody reports a segmentation fault because her operating system 
ran out of file descriptors.

In any case, I would rather be safe than sorry.  It is not a good habit to 
get into, calling functions that can return NULL and not checking them 
(even if you try to be super-clever).

quoted

quoted

+	while (1) {
+		char *eov;
+		if (strbuf_getline(&buf, file, '\n') == EOF)
+			exit(128); /* child died, message supplied already */
+
+		if (!*buf.buf)
+			break;
+
+		eov = strchr(buf.buf, ' ');
+		if (!eov)
+			die("Malformed response in ref list: %s", buf.buf);
+		*eov = '\0';
+		*end = alloc_ref(eov + 1);
+		if (eov) {

Did we not die 4 lines earlier if eov == NULL?

Yeah, I confused myself while paring down the vcs output parsing code. I 
think I actually want to future-proof this code to allow the helper to 
add space-separated flags after the name, and I don't want this test.

Yeah, why not making it hard on the reviewer for the slight possibility 
that some change will go in in the future, in the exact form you envisaged 
it now?

quoted

quoted

+			if (buf.buf[0] == '@')
+				(*end)->symref = xstrdup(buf.buf + 1);
+			else
+				get_sha1_hex(buf.buf, (*end)->old_sha1);

IMHO it is not at all clear what you are doing here.  At least a little 
hint is in order if the code does not explain itself.

It's parsing the output of the "list" command, as given in the 
documentation. Item starting with a '@' is a symref, otherwise it is a 
sha1.

That's nice.

It would be even nicer if there was a comment in the patch so that future 
puzzled readers do not know that they have to search the mailing list to 
understand this code.

Ciao,
Dscho