* [PATCH 0/12] reducing resource usage of for_each_alternate_ref
From: Jeff King @ 2017-01-24 0:37 UTC (permalink / raw)
To: git
As I've mentioned before, I have some alternates repositories with
absurd numbers of refs, most of which are duplicates of each other[1].
There are a couple of problems I've seen:
1. the way that for_each_alternate_ref() works has very high peak memory
usage for this case
2. the way that receive-pack de-duplicates the refs has high peak memory
usage
3. we access the alternate refs twice in fetch-pack
This fixes all three, along with a few other minor bugfixes, cleanups,
and optimizations. I've tried to order the series to keep bugfixes and
less-contentious changes near the front.
Just to give some numbers, on my worst-case repository (see [1]), this
drops peak RSS for "git clone --reference" from over 25GB to about 40MB.
Sort of, anyway. You still pay a big CPU and RSS cost on the process in
the alternates repo that accesses packed-refs, but the 25GB came on top
of that. So this is a first pass at the low-hanging fruit.
I'll be the first to admit that this setup is insane. And some of the
optimizations are tradeoffs that help particularly the case where your
refs aren't unique. But for the most part should help _every_ case. And
in the cases where your refs are unique, either you don't have many (so
the tradeoffs are OK) or you have so many that you are pretty much
screwed no matter what (if your fetch is looking at 30 million unique
ref tips, the object storage is your problem, not looking at the refs).
A brief overview of the patches:
[01/12]: for_each_alternate_ref: handle failure from real_pathdup()
[02/12]: for_each_alternate_ref: stop trimming trailing slashes
[03/12]: for_each_alternate_ref: use strbuf for path allocation
Bugfixes and cleanups (the first one is actually a recent-ish
regression).
[04/12]: for_each_alternate_ref: pass name/oid instead of ref struct
[05/12]: for_each_alternate_ref: replace transport code with for-each-ref
[06/12]: clone: disable save_commit_buffer
This gives the 25GB->40MB benefit. There are a bunch of caveats in
patch 05, but I _think_ it's the right direction for the reasons I
outlined there.
[07/12]: fetch-pack: cache results of for_each_alternate_ref
Just running for-each-ref in the giant alternates repo is about a
minute of CPU, plus 10GB heap, and fetch-pack wants to do it twice.
This drops it to once.
[08/12]: add oidset API
[09/12]: receive-pack: use oidset to de-duplicate .have lines
These give another ~12GB RSS improvement when receive-pack looks at
the alternates in my worst-case repo.
This is less tested than the earlier ones, as we've disabled
receive-pack looking at alternates in production (see [1] below).
I just did them for completeness upstream.
[10/12]: receive-pack: fix misleading namespace/.have comment
[11/12]: receive-pack: treat namespace .have lines like alternates
[12/12]: receive-pack: avoid duplicates between our refs and alternates
These are optimizations to avoid more duplicate .have lines, and
should benefit even non-insane cases. As with 8-12, not as well
tested by me.
Makefile | 1 +
builtin/clone.c | 1 +
builtin/receive-pack.c | 41 +++++++++++++++-------------
fetch-pack.c | 48 ++++++++++++++++++++++++++++-----
object.h | 2 +-
oidset.c | 49 ++++++++++++++++++++++++++++++++++
oidset.h | 45 +++++++++++++++++++++++++++++++
t/t5400-send-pack.sh | 38 ++++++++++++++++++++++++++
transport.c | 72 +++++++++++++++++++++++++++++++++++---------------
transport.h | 2 +-
10 files changed, 250 insertions(+), 49 deletions(-)
create mode 100644 oidset.c
create mode 100644 oidset.h
-Peff
[1] Background, if you care:
I've mentioned before that GitHub's repository storage uses a
fork-network layout. Each user gets their own "fork" repository, and
it points to "network.git" as shared storage. The network.git
repository has a ref for each fork that is updated periodically via
"git fetch".
So the network.git repo contains O(nr_forks * nr_refs_in_fork) refs.
Quite a few of these point to the same tip sha1s, as each fork has
the same tags. One of the most pathological cases is a popular
public repo that has ~44K tags in it. The network repo has ~80
million refs, of which ~60K are unique. You can imagine that it
takes some time to access the refs. Basically anything that loads
the network.git packed-refs file takes an extra minute of CPU and
needs ~10G RSS to store the internal cache of `packed-refs`.
Most operations don't care about this; they work on the fork repo,
and never look at the network refs. But we _do_ sometimes peek at
alternate refs from receive-pack and fetch-pack to tell the other
side about tips we have.
These optimizations backfire completely in such a setting. Besides
the CPU and memory spikes on the server, even just the unique refs
add over 3MB to the ref advertisement. So for the time being, we've
disabled them. I have patches that add a receive.advertiseAlternates
config option, if anybody is interested.
So why do I care? There is one exception: when we are cloning a
fork, we turn on the fetch-pack side of the optimizations. This is
worth it for a large repository, as it saves us having to transfer
any objects at all (and therefore not process them with index-pack,
which is expensive).
This series is also a first step towards other optimizations, like
asking for just the alternate refs from _one_ of the forks. I hope
one day to turn alternates optimizations back on everywhere.
^ permalink raw reply
* [PATCH 01/12] for_each_alternate_ref: handle failure from real_pathdup()
From: Jeff King @ 2017-01-24 0:38 UTC (permalink / raw)
To: git
In-Reply-To: <20170124003729.j4ygjcgypdq7hceg@sigill.intra.peff.net>
In older versions of git, if real_path() failed to resolve
the alternate object store path, we would die() with an
error. However, since 4ac9006f8 (real_path: have callers use
real_pathdup and strbuf_realpath, 2016-12-12) we use the
real_pathdup() function, which may return NULL. Since we
don't check the return value, we can segfault.
This is hard to trigger in practice, since we check that the
path is accessible before creating the alternate_object_database
struct. But it could be removed racily, or we could see a
transient filesystem error.
We could restore the original behavior by switching back to
xstrdup(real_path()). However, dying is probably not the
best option here. This whole function is best-effort
already; there might not even be a repository around the
shared objects at all. And if the alternate store has gone
away, there are no objects to show.
So let's just quietly return, as we would if we failed to
open "refs/", or if upload-pack failed to start, etc.
Signed-off-by: Jeff King <peff@peff.net>
---
transport.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/transport.c b/transport.c
index c86ba2eb8..74d0e45bd 100644
--- a/transport.c
+++ b/transport.c
@@ -1215,6 +1215,8 @@ static int refs_from_alternate_cb(struct alternate_object_database *e,
struct alternate_refs_data *cb = data;
other = real_pathdup(e->path);
+ if (!other)
+ return 0;
len = strlen(other);
while (other[len-1] == '/')
--
2.11.0.765.g454d2182f
^ permalink raw reply related
* [PATCH 02/12] for_each_alternate_ref: stop trimming trailing slashes
From: Jeff King @ 2017-01-24 0:39 UTC (permalink / raw)
To: git
In-Reply-To: <20170124003729.j4ygjcgypdq7hceg@sigill.intra.peff.net>
The real_pathdup() function will have removed extra slashes
for us already (on top of the normalize_path() done when we
created the alternate_object_database struct in the first
place).
Incidentally, this also fixes the case where the path is
just "/", which would read off the start of the array.
That doesn't seem possible to trigger in practice, though,
as link_alt_odb_entry() blindly eats trailing slashes,
including a bare "/".
Signed-off-by: Jeff King <peff@peff.net>
---
I think the "/" thing in link_alt_odb_entry() is buggy, and it's an easy
one-liner fix. But I notice some of the rest of the code isn't ready to
handle "/" (mostly it just duplicates "/" when appending to the path),
so I left it for now (and I doubt anybody cares anyway).
transport.c | 2 --
1 file changed, 2 deletions(-)
diff --git a/transport.c b/transport.c
index 74d0e45bd..6b4b3ed31 100644
--- a/transport.c
+++ b/transport.c
@@ -1219,8 +1219,6 @@ static int refs_from_alternate_cb(struct alternate_object_database *e,
return 0;
len = strlen(other);
- while (other[len-1] == '/')
- other[--len] = '\0';
if (len < 8 || memcmp(other + len - 8, "/objects", 8))
goto out;
/* Is this a git repository with refs? */
--
2.11.0.765.g454d2182f
^ permalink raw reply related
* [PATCH 03/12] for_each_alternate_ref: use strbuf for path allocation
From: Jeff King @ 2017-01-24 0:40 UTC (permalink / raw)
To: git
In-Reply-To: <20170124003729.j4ygjcgypdq7hceg@sigill.intra.peff.net>
We have a string with ".../objects" pointing to the
alternate object store, and overwrite bits of it to look at
other paths in the (potential) git repository holding it.
This works because the only path we care about is "refs",
which is shorter than "objects".
Using a strbuf to hold the path lets us get rid of some
magic numbers, and makes it more obvious that the memory
operations are safe.
Signed-off-by: Jeff King <peff@peff.net>
---
I had thought I was going to need to generate more paths in the
alternate repo, but I ended up not needing to. So this was originally
done to make that easier, but I think it stands on its own as a cleanup.
transport.c | 28 ++++++++++++++--------------
1 file changed, 14 insertions(+), 14 deletions(-)
diff --git a/transport.c b/transport.c
index 6b4b3ed31..594533d88 100644
--- a/transport.c
+++ b/transport.c
@@ -1207,34 +1207,34 @@ struct alternate_refs_data {
static int refs_from_alternate_cb(struct alternate_object_database *e,
void *data)
{
- char *other;
- size_t len;
+ struct strbuf path = STRBUF_INIT;
+ size_t base_len;
struct remote *remote;
struct transport *transport;
const struct ref *extra;
struct alternate_refs_data *cb = data;
- other = real_pathdup(e->path);
- if (!other)
- return 0;
- len = strlen(other);
-
- if (len < 8 || memcmp(other + len - 8, "/objects", 8))
+ if (!strbuf_realpath(&path, e->path, 0))
+ goto out;
+ if (!strbuf_strip_suffix(&path, "/objects"))
goto out;
+ base_len = path.len;
+
/* Is this a git repository with refs? */
- memcpy(other + len - 8, "/refs", 6);
- if (!is_directory(other))
+ strbuf_addstr(&path, "/refs");
+ if (!is_directory(path.buf))
goto out;
- other[len - 8] = '\0';
- remote = remote_get(other);
- transport = transport_get(remote, other);
+ strbuf_setlen(&path, base_len);
+
+ remote = remote_get(path.buf);
+ transport = transport_get(remote, path.buf);
for (extra = transport_get_remote_refs(transport);
extra;
extra = extra->next)
cb->fn(extra, cb->data);
transport_disconnect(transport);
out:
- free(other);
+ strbuf_release(&path);
return 0;
}
--
2.11.0.765.g454d2182f
^ permalink raw reply related
* [PATCH 04/12] for_each_alternate_ref: pass name/oid instead of ref struct
From: Jeff King @ 2017-01-24 0:40 UTC (permalink / raw)
To: git
In-Reply-To: <20170124003729.j4ygjcgypdq7hceg@sigill.intra.peff.net>
Breaking down the fields in the interface makes it easier to
change the backend of for_each_alternate_ref to something
that doesn't use "struct ref" internally.
The only field that callers actually look at is the oid,
anyway. The refname is kept in the interface as a plausible
thing for future code to want.
Signed-off-by: Jeff King <peff@peff.net>
---
builtin/receive-pack.c | 6 ++++--
fetch-pack.c | 12 ++++++++----
transport.c | 2 +-
transport.h | 2 +-
4 files changed, 14 insertions(+), 8 deletions(-)
diff --git a/builtin/receive-pack.c b/builtin/receive-pack.c
index 6b97cbdbe..b9f2c0cc5 100644
--- a/builtin/receive-pack.c
+++ b/builtin/receive-pack.c
@@ -277,10 +277,12 @@ static int show_one_alternate_sha1(const unsigned char sha1[20], void *unused)
return 0;
}
-static void collect_one_alternate_ref(const struct ref *ref, void *data)
+static void collect_one_alternate_ref(const char *refname,
+ const struct object_id *oid,
+ void *data)
{
struct sha1_array *sa = data;
- sha1_array_append(sa, ref->old_oid.hash);
+ sha1_array_append(sa, oid->hash);
}
static void write_head_info(void)
diff --git a/fetch-pack.c b/fetch-pack.c
index 601f0779a..54f84c573 100644
--- a/fetch-pack.c
+++ b/fetch-pack.c
@@ -253,9 +253,11 @@ static void send_request(struct fetch_pack_args *args,
write_or_die(fd, buf->buf, buf->len);
}
-static void insert_one_alternate_ref(const struct ref *ref, void *unused)
+static void insert_one_alternate_ref(const char *refname,
+ const struct object_id *oid,
+ void *unused)
{
- rev_list_insert_ref(NULL, ref->old_oid.hash);
+ rev_list_insert_ref(NULL, oid->hash);
}
#define INITIAL_FLUSH 16
@@ -619,9 +621,11 @@ static void filter_refs(struct fetch_pack_args *args,
*refs = newlist;
}
-static void mark_alternate_complete(const struct ref *ref, void *unused)
+static void mark_alternate_complete(const char *refname,
+ const struct object_id *oid,
+ void *unused)
{
- mark_complete(ref->old_oid.hash);
+ mark_complete(oid->hash);
}
static int everything_local(struct fetch_pack_args *args,
diff --git a/transport.c b/transport.c
index 594533d88..983d8fec1 100644
--- a/transport.c
+++ b/transport.c
@@ -1231,7 +1231,7 @@ static int refs_from_alternate_cb(struct alternate_object_database *e,
for (extra = transport_get_remote_refs(transport);
extra;
extra = extra->next)
- cb->fn(extra, cb->data);
+ cb->fn(extra->name, &extra->old_oid, cb->data);
transport_disconnect(transport);
out:
strbuf_release(&path);
diff --git a/transport.h b/transport.h
index 9820f10b8..b7bb07d2c 100644
--- a/transport.h
+++ b/transport.h
@@ -254,6 +254,6 @@ int transport_refs_pushed(struct ref *ref);
void transport_print_push_status(const char *dest, struct ref *refs,
int verbose, int porcelain, unsigned int *reject_reasons);
-typedef void alternate_ref_fn(const struct ref *, void *);
+typedef void alternate_ref_fn(const char *refname, const struct object_id *oid, void *);
extern void for_each_alternate_ref(alternate_ref_fn, void *);
#endif
--
2.11.0.765.g454d2182f
^ permalink raw reply related
* [PATCH 05/12] for_each_alternate_ref: replace transport code with for-each-ref
From: Jeff King @ 2017-01-24 0:44 UTC (permalink / raw)
To: git
In-Reply-To: <20170124003729.j4ygjcgypdq7hceg@sigill.intra.peff.net>
The current method for getting the refs from an alternate is
to run upload-pack in the alternate and parse its output
using the normal transport code. This works and is
reasonably short, but it has a very bad memory footprint
when there are a lot of refs in the alternate. There are two
problems:
1. It reads in all of the refs before passing any back to
us. Which means that our peak memory usage has to store
every ref (including duplicates for peeled variants),
even if our callback could determine that some are not
interesting (e.g., because they point to the same sha1
as another ref).
2. It allocates a "struct ref" for each one. Among other
things, this contains 3 separate 20-byte oids, along
with the name and various pointers. That can add up,
especially if the callback is only interested in the
sha1 (which it can store in a sha1_array as just 20
bytes).
On a particularly pathological case, where the alternate had
over 80 million refs pointing to only around 60,000 unique
objects, the peak heap usage of "git clone --reference" grew
to over 25GB.
This patch instead calls git-for-each-ref in the alternate
repository, and passes each line to the callback as we read
it. That drops the peak heap of the same command to 50MB.
I considered and rejected a few alternatives.
We could read all of the refs in the alternate using our own
ref code, just as we do with submodules. However, as memory
footprint is one of the concerns here, we want to avoid
loading those refs into our own memory as a whole.
It's possible that this will be a better technique in the
future when the ref code can more easily iterate without
loading all of packed-refs into memory.
Another option is to keep calling upload-pack, and just
parse its output ourselves in a streaming fashion. Besides
for-each-ref being simpler (we get to define the format
ourselves, and don't have to deal with speaking the git
protocol), it's more flexible for possible future changes.
For instance, it might be useful for the caller to be able
to limit the set of "interesting" alternate refs. The
motivating example is one where many "forks" of a particular
repository share object storage, and the shared storage has
refs for each fork (which is why so many of the refs are
duplicates; each fork has the same tags). A plausible
future optimization would be to ask for the alternate refs
for just _one_ fork (if you had some out-of-band way of
knowing which was the most interesting or important for the
current operation).
Similarly, no callbacks actually care about the symref value
of alternate refs, and as before, this patch ignores them
entirely. However, if we wanted to add them, for-each-ref's
"%(symref)" is going to be more flexible than upload-pack,
because the latter only handles the HEAD symref due to
historical constraints.
There is one potential downside, though: unlike upload-pack,
our for-each-ref command doesn't report the peeled value of
refs. The existing code calls the alternate_ref_fn callback
twice for tags: once for the tag, and once for the peeled
value with the refname set to "ref^{}".
For the callers in fetch-pack, this doesn't matter at all.
We immediately peel each tag down to a commit either way (so
there's a slight improvement, as do not bother passing the
redundant data over the pipe). For the caller in
receive-pack, it means we will not advertise the peeled
values of tags in our alternate. However, we also don't
advertise peeled values for our _own_ tags, so this is
actually making things more consistent.
It's unclear whether receive-pack advertising peeled values
is a win or not. On one hand, giving more information to the
other side may let it omit some objects from the push. On
the other hand, for tags which both sides have, they simply
bloat the advertisement. The upload-pack advertisement of
git.git is about 30% larger than the receive-pack
advertisement due to its peeled information.
This patch omits the peeled information from
for_each_alternate_ref entirely, and leaves it up to the
caller whether they want to dig up the information.
Signed-off-by: Jeff King <peff@peff.net>
---
I also tried adding "%(*objectname)" to for-each-ref to just
grab the peeled information, but the peel implementation in
ref-filter is _really_ slow. It doesn't use the packed-ref
peel information, and it doesn't recognize duplicates (so in
the 80 million case, it really parses 80 million tags).
transport.c | 48 ++++++++++++++++++++++++++++++++++++++----------
1 file changed, 38 insertions(+), 10 deletions(-)
diff --git a/transport.c b/transport.c
index 983d8fec1..ef8e09298 100644
--- a/transport.c
+++ b/transport.c
@@ -1199,6 +1199,42 @@ char *transport_anonymize_url(const char *url)
return xstrdup(url);
}
+static void read_alternate_refs(const char *path,
+ alternate_ref_fn *cb,
+ void *data)
+{
+ struct child_process cmd = CHILD_PROCESS_INIT;
+ struct strbuf line = STRBUF_INIT;
+ FILE *fh;
+
+ cmd.git_cmd = 1;
+ argv_array_pushf(&cmd.args, "--git-dir=%s", path);
+ argv_array_push(&cmd.args, "for-each-ref");
+ argv_array_push(&cmd.args, "--format=%(objectname) %(refname)");
+ cmd.env = local_repo_env;
+ cmd.out = -1;
+
+ if (start_command(&cmd))
+ return;
+
+ fh = xfdopen(cmd.out, "r");
+ while (strbuf_getline_lf(&line, fh) != EOF) {
+ struct object_id oid;
+
+ if (get_oid_hex(line.buf, &oid) ||
+ line.buf[GIT_SHA1_HEXSZ] != ' ') {
+ warning("invalid line while parsing alternate refs: %s",
+ line.buf);
+ break;
+ }
+
+ cb(line.buf + GIT_SHA1_HEXSZ + 1, &oid, data);
+ }
+
+ fclose(fh);
+ finish_command(&cmd);
+}
+
struct alternate_refs_data {
alternate_ref_fn *fn;
void *data;
@@ -1209,9 +1245,6 @@ static int refs_from_alternate_cb(struct alternate_object_database *e,
{
struct strbuf path = STRBUF_INIT;
size_t base_len;
- struct remote *remote;
- struct transport *transport;
- const struct ref *extra;
struct alternate_refs_data *cb = data;
if (!strbuf_realpath(&path, e->path, 0))
@@ -1226,13 +1259,8 @@ static int refs_from_alternate_cb(struct alternate_object_database *e,
goto out;
strbuf_setlen(&path, base_len);
- remote = remote_get(path.buf);
- transport = transport_get(remote, path.buf);
- for (extra = transport_get_remote_refs(transport);
- extra;
- extra = extra->next)
- cb->fn(extra->name, &extra->old_oid, cb->data);
- transport_disconnect(transport);
+ read_alternate_refs(path.buf, cb->fn, cb->data);
+
out:
strbuf_release(&path);
return 0;
--
2.11.0.765.g454d2182f
^ permalink raw reply related
* [PATCH 06/12] clone: disable save_commit_buffer
From: Jeff King @ 2017-01-24 0:45 UTC (permalink / raw)
To: git
In-Reply-To: <20170124003729.j4ygjcgypdq7hceg@sigill.intra.peff.net>
Normally git caches the raw commit object contents in
"struct commit". This makes it fast to run parse_commit()
followed by a pretty-print operation.
For commands which don't actually pretty-print the commits,
the caching is wasteful (and may use quite a lot of memory
if git accesses a large number of commits).
For fetching operations like clone, we already disable
save_commit_buffer in everything_local(), where we may
traverse. But clone also parses commits before it even gets
there; it looks at commits reachable from its refs, and from
alternate refs.
In one real-world case with a large number of tags, this
cut about 10MB off of clone's heap usage. Not spectacular,
but there's really no downside.
Signed-off-by: Jeff King <peff@peff.net>
---
builtin/clone.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/builtin/clone.c b/builtin/clone.c
index 5ef81927a..3fca45e7e 100644
--- a/builtin/clone.c
+++ b/builtin/clone.c
@@ -858,6 +858,7 @@ int cmd_clone(int argc, const char **argv, const char *prefix)
struct refspec *refspec;
const char *fetch_pattern;
+ save_commit_buffer = 0;
packet_trace_identity("clone");
argc = parse_options(argc, argv, prefix, builtin_clone_options,
builtin_clone_usage, 0);
--
2.11.0.765.g454d2182f
^ permalink raw reply related
* [PATCH 07/12] fetch-pack: cache results of for_each_alternate_ref
From: Jeff King @ 2017-01-24 0:45 UTC (permalink / raw)
To: git
In-Reply-To: <20170124003729.j4ygjcgypdq7hceg@sigill.intra.peff.net>
We may run for_each_alternate_ref() twice, once in
find_common() and once in everything_local(). This operation
can be expensive, because it involves running a sub-process
which must freshly load all of the alternate's refs from
disk.
Let's cache and reuse the results between the two calls. We
can make some optimizations based on the particular use
pattern in fetch-pack to keep our memory usage down.
The first is that we only care about the sha1s, not the refs
themselves. So it's OK to store only the sha1s, and to
suppress duplicates. The natural fit would therefore be a
sha1_array.
However, sha1_array's de-duplication happens only after it
has read and sorted all entries. It still stores each
duplicate. For an alternate with a large number of refs
pointing to the same commits, this is a needless expense.
Instead, we'd prefer to eliminate duplicates before putting
them in the cache, which implies using a hash. We can
further note that fetch-pack will call parse_object() on
each alternate sha1. We can therefore keep our cache as a
set of pointers to "struct object". That gives us a place to
put our "already seen" bit with an optimized hash lookup.
And as a bonus, the object stores the sha1 for us, so
pointer-to-object is all we need.
There are two extra optimizations I didn't do here:
- we actually store an array of pointer-to-object.
Technically we could just walk the obj_hash table
looking for entries with the ALTERNATE flag set (because
our use case doesn't care about the order here).
But that hash table may be mostly composed of
non-ALTERNATE entries, so we'd waste time walking over
them. So it would be a slight win in memory use, but a
loss in CPU.
- the items we pull out of the cache are actual "struct
object"s, but then we feed "obj->sha1" to our
sub-functions, which promptly call parse_object().
This second parse is cheap, because it starts with
lookup_object() and will bail immediately when it sees
we've already parsed the object. We could save the extra
hash lookup, but it would involve refactoring the
functions we call. It may or may not be worth the
trouble.
Signed-off-by: Jeff King <peff@peff.net>
---
fetch-pack.c | 52 ++++++++++++++++++++++++++++++++++++++++++----------
object.h | 2 +-
2 files changed, 43 insertions(+), 11 deletions(-)
diff --git a/fetch-pack.c b/fetch-pack.c
index 54f84c573..e0f5d5ce8 100644
--- a/fetch-pack.c
+++ b/fetch-pack.c
@@ -35,6 +35,7 @@ static const char *alternate_shallow_file;
#define COMMON_REF (1U << 2)
#define SEEN (1U << 3)
#define POPPED (1U << 4)
+#define ALTERNATE (1U << 5)
static int marked;
@@ -67,6 +68,41 @@ static inline void print_verbose(const struct fetch_pack_args *args,
fputc('\n', stderr);
}
+struct alternate_object_cache {
+ struct object **items;
+ size_t nr, alloc;
+};
+
+static void cache_one_alternate(const char *refname,
+ const struct object_id *oid,
+ void *vcache)
+{
+ struct alternate_object_cache *cache = vcache;
+ struct object *obj = parse_object(oid->hash);
+
+ if (!obj || (obj->flags & ALTERNATE))
+ return;
+
+ obj->flags |= ALTERNATE;
+ ALLOC_GROW(cache->items, cache->nr + 1, cache->alloc);
+ cache->items[cache->nr++] = obj;
+}
+
+static void for_each_cached_alternate(void (*cb)(struct object *))
+{
+ static int initialized;
+ static struct alternate_object_cache cache;
+ size_t i;
+
+ if (!initialized) {
+ for_each_alternate_ref(cache_one_alternate, &cache);
+ initialized = 1;
+ }
+
+ for (i = 0; i < cache.nr; i++)
+ cb(cache.items[i]);
+}
+
static void rev_list_push(struct commit *commit, int mark)
{
if (!(commit->object.flags & mark)) {
@@ -253,11 +289,9 @@ static void send_request(struct fetch_pack_args *args,
write_or_die(fd, buf->buf, buf->len);
}
-static void insert_one_alternate_ref(const char *refname,
- const struct object_id *oid,
- void *unused)
+static void insert_one_alternate_object(struct object *obj)
{
- rev_list_insert_ref(NULL, oid->hash);
+ rev_list_insert_ref(NULL, obj->oid.hash);
}
#define INITIAL_FLUSH 16
@@ -300,7 +334,7 @@ static int find_common(struct fetch_pack_args *args,
marked = 1;
for_each_ref(rev_list_insert_ref_oid, NULL);
- for_each_alternate_ref(insert_one_alternate_ref, NULL);
+ for_each_cached_alternate(insert_one_alternate_object);
fetching = 0;
for ( ; refs ; refs = refs->next) {
@@ -621,11 +655,9 @@ static void filter_refs(struct fetch_pack_args *args,
*refs = newlist;
}
-static void mark_alternate_complete(const char *refname,
- const struct object_id *oid,
- void *unused)
+static void mark_alternate_complete(struct object *obj)
{
- mark_complete(oid->hash);
+ mark_complete(obj->oid.hash);
}
static int everything_local(struct fetch_pack_args *args,
@@ -661,7 +693,7 @@ static int everything_local(struct fetch_pack_args *args,
if (!args->deepen) {
for_each_ref(mark_complete_oid, NULL);
- for_each_alternate_ref(mark_alternate_complete, NULL);
+ for_each_cached_alternate(mark_alternate_complete);
commit_list_sort_by_date(&complete);
if (cutoff)
mark_recent_complete_commits(args, cutoff);
diff --git a/object.h b/object.h
index 614a00675..f52957dcb 100644
--- a/object.h
+++ b/object.h
@@ -29,7 +29,7 @@ struct object_array {
/*
* object flag allocation:
* revision.h: 0---------10 26
- * fetch-pack.c: 0---4
+ * fetch-pack.c: 0---5
* walker.c: 0-2
* upload-pack.c: 4 11----------------19
* builtin/blame.c: 12-13
--
2.11.0.765.g454d2182f
^ permalink raw reply related
* [PATCH 08/12] add oidset API
From: Jeff King @ 2017-01-24 0:46 UTC (permalink / raw)
To: git
In-Reply-To: <20170124003729.j4ygjcgypdq7hceg@sigill.intra.peff.net>
This is similar to many of our uses of sha1-array, but it
overcomes one limitation of a sha1-array: when you are
de-duplicating a large input with relatively few unique
entries, sha1-array uses 20 bytes per non-unique entry.
Whereas this set will use memory linear in the number of
unique entries (albeit a few more than 20 bytes due to
hashmap overhead).
Signed-off-by: Jeff King <peff@peff.net>
---
This may be overkill. You can get roughly the same thing by making
actual object structs via lookup_unknown_object(). But see the next
patch for some comments on that.
Makefile | 1 +
oidset.c | 49 +++++++++++++++++++++++++++++++++++++++++++++++++
oidset.h | 45 +++++++++++++++++++++++++++++++++++++++++++++
3 files changed, 95 insertions(+)
create mode 100644 oidset.c
create mode 100644 oidset.h
diff --git a/Makefile b/Makefile
index 27afd0f37..e41efc2d8 100644
--- a/Makefile
+++ b/Makefile
@@ -774,6 +774,7 @@ LIB_OBJS += notes-cache.o
LIB_OBJS += notes-merge.o
LIB_OBJS += notes-utils.o
LIB_OBJS += object.o
+LIB_OBJS += oidset.o
LIB_OBJS += pack-bitmap.o
LIB_OBJS += pack-bitmap-write.o
LIB_OBJS += pack-check.o
diff --git a/oidset.c b/oidset.c
new file mode 100644
index 000000000..6094cff8c
--- /dev/null
+++ b/oidset.c
@@ -0,0 +1,49 @@
+#include "cache.h"
+#include "oidset.h"
+
+struct oidset_entry {
+ struct hashmap_entry hash;
+ struct object_id oid;
+};
+
+int oidset_hashcmp(const void *va, const void *vb,
+ const void *vkey)
+{
+ const struct oidset_entry *a = va, *b = vb;
+ const struct object_id *key = vkey;
+ return oidcmp(&a->oid, key ? key : &b->oid);
+}
+
+int oidset_contains(const struct oidset *set, const struct object_id *oid)
+{
+ struct hashmap_entry key;
+
+ if (!set->map.cmpfn)
+ return 0;
+
+ hashmap_entry_init(&key, sha1hash(oid->hash));
+ return !!hashmap_get(&set->map, &key, oid);
+}
+
+int oidset_insert(struct oidset *set, const struct object_id *oid)
+{
+ struct oidset_entry *entry;
+
+ if (!set->map.cmpfn)
+ hashmap_init(&set->map, oidset_hashcmp, 0);
+
+ if (oidset_contains(set, oid))
+ return 1;
+
+ entry = xmalloc(sizeof(*entry));
+ hashmap_entry_init(&entry->hash, sha1hash(oid->hash));
+ oidcpy(&entry->oid, oid);
+
+ hashmap_add(&set->map, entry);
+ return 0;
+}
+
+void oidset_clear(struct oidset *set)
+{
+ hashmap_free(&set->map, 1);
+}
diff --git a/oidset.h b/oidset.h
new file mode 100644
index 000000000..b7eaab5b8
--- /dev/null
+++ b/oidset.h
@@ -0,0 +1,45 @@
+#ifndef OIDSET_H
+#define OIDSET_H
+
+/**
+ * This API is similar to sha1-array, in that it maintains a set of object ids
+ * in a memory-efficient way. The major differences are:
+ *
+ * 1. It uses a hash, so we can do online duplicate removal, rather than
+ * sort-and-uniq at the end. This can reduce memory footprint if you have
+ * a large list of oids with many duplicates.
+ *
+ * 2. The per-unique-oid memory footprint is slightly higher due to hash
+ * table overhead.
+ */
+
+/**
+ * A single oidset; should be zero-initialized (or use OIDSET_INIT).
+ */
+struct oidset {
+ struct hashmap map;
+};
+
+#define OIDSET_INIT { { NULL } }
+
+/**
+ * Returns true iff `set` contains `oid`.
+ */
+int oidset_contains(const struct oidset *set, const struct object_id *oid);
+
+/**
+ * Insert the oid into the set; a copy is made, so "oid" does not need
+ * to persist after this function is called.
+ *
+ * Returns 1 if the oid was already in the set, 0 otherwise. This can be used
+ * to perform an efficient check-and-add.
+ */
+int oidset_insert(struct oidset *set, const struct object_id *oid);
+
+/**
+ * Remove all entries from the oidset, freeing any resources associated with
+ * it.
+ */
+void oidset_clear(struct oidset *set);
+
+#endif /* OIDSET_H */
--
2.11.0.765.g454d2182f
^ permalink raw reply related
* [PATCH 10/12] receive-pack: fix misleading namespace/.have comment
From: Jeff King @ 2017-01-24 0:47 UTC (permalink / raw)
To: git
In-Reply-To: <20170124003729.j4ygjcgypdq7hceg@sigill.intra.peff.net>
The comment claims that we handle alternate ".have" lines
through this function, but that hasn't been the case since
85f251045 (write_head_info(): handle "extra refs" locally,
2012-01-06).
Signed-off-by: Jeff King <peff@peff.net>
---
builtin/receive-pack.c | 5 +----
1 file changed, 1 insertion(+), 4 deletions(-)
diff --git a/builtin/receive-pack.c b/builtin/receive-pack.c
index 27bb52988..8f8762e4a 100644
--- a/builtin/receive-pack.c
+++ b/builtin/receive-pack.c
@@ -261,10 +261,7 @@ static int show_ref_cb(const char *path_full, const struct object_id *oid,
/*
* Advertise refs outside our current namespace as ".have"
* refs, so that the client can use them to minimize data
- * transfer but will otherwise ignore them. This happens to
- * cover ".have" that are thrown in by add_one_alternate_ref()
- * to mark histories that are complete in our alternates as
- * well.
+ * transfer but will otherwise ignore them.
*/
if (!path)
path = ".have";
--
2.11.0.765.g454d2182f
^ permalink raw reply related
* [PATCH 09/12] receive-pack: use oidset to de-duplicate .have lines
From: Jeff King @ 2017-01-24 0:47 UTC (permalink / raw)
To: git
In-Reply-To: <20170124003729.j4ygjcgypdq7hceg@sigill.intra.peff.net>
If you have an alternate object store with a very large
number of refs, the peak memory usage of the sha1_array can
grow high, even if most of them are duplicates that end up
not being printed at all.
The similar for_each_alternate_ref() code-paths in
fetch-pack solve this by using flags in "struct object" to
de-duplicate (and so are relying on obj_hash at the core).
But we don't have a "struct object" at all in this case. We
could call lookup_unknown_object() to get one, but if our
goal is reducing memory footprint, it's not great:
- an unknown object is as large as the largest object type
(a commit), which is bigger than an oidset entry
- we can free the memory after our ref advertisement, but
"struct object" entries persist forever (and the
receive-pack may hang around for a long time, as the
bottleneck is often client upload bandwidth).
So let's use an oidset. Note that unlike a sha1-array it
doesn't sort the output as a side effect. However, our
output is at least stable, because for_each_alternate_ref()
will give us the sha1s in ref-sorted order.
In one particularly pathological case with an alternate that
has 60,000 unique refs out of 80 million total, this reduced
the peak heap usage of "git receive-pack . </dev/null" from
13GB to 14MB.
Signed-off-by: Jeff King <peff@peff.net>
---
builtin/receive-pack.c | 26 ++++++++++++--------------
1 file changed, 12 insertions(+), 14 deletions(-)
diff --git a/builtin/receive-pack.c b/builtin/receive-pack.c
index b9f2c0cc5..27bb52988 100644
--- a/builtin/receive-pack.c
+++ b/builtin/receive-pack.c
@@ -21,6 +21,7 @@
#include "sigchain.h"
#include "fsck.h"
#include "tmp-objdir.h"
+#include "oidset.h"
static const char * const receive_pack_usage[] = {
N_("git receive-pack <git-dir>"),
@@ -271,27 +272,24 @@ static int show_ref_cb(const char *path_full, const struct object_id *oid,
return 0;
}
-static int show_one_alternate_sha1(const unsigned char sha1[20], void *unused)
+static void show_one_alternate_ref(const char *refname,
+ const struct object_id *oid,
+ void *data)
{
- show_ref(".have", sha1);
- return 0;
-}
+ struct oidset *seen = data;
-static void collect_one_alternate_ref(const char *refname,
- const struct object_id *oid,
- void *data)
-{
- struct sha1_array *sa = data;
- sha1_array_append(sa, oid->hash);
+ if (oidset_insert(seen, oid))
+ return;
+
+ show_ref(".have", oid->hash);
}
static void write_head_info(void)
{
- struct sha1_array sa = SHA1_ARRAY_INIT;
+ static struct oidset seen = OIDSET_INIT;
- for_each_alternate_ref(collect_one_alternate_ref, &sa);
- sha1_array_for_each_unique(&sa, show_one_alternate_sha1, NULL);
- sha1_array_clear(&sa);
+ for_each_alternate_ref(show_one_alternate_ref, &seen);
+ oidset_clear(&seen);
for_each_ref(show_ref_cb, NULL);
if (!sent_capabilities)
show_ref("capabilities^{}", null_sha1);
--
2.11.0.765.g454d2182f
^ permalink raw reply related
* [PATCH 12/12] receive-pack: avoid duplicates between our refs and alternates
From: Jeff King @ 2017-01-24 0:48 UTC (permalink / raw)
To: git
In-Reply-To: <20170124003729.j4ygjcgypdq7hceg@sigill.intra.peff.net>
We de-duplicate ".have" refs among themselves, but never
check if they are duplicates of our local refs. It's not
unreasonable that they would be if we are a "--shared" or
"--reference" clone of a similar repository; we'd have all
the same tags.
We can handle this by inserting our local refs into the
oidset, but obviously not suppressing duplicates (since the
refnames are important).
Note that this also switches the order in which we advertise
refs, processing ours first and then any alternates. The
order shouldn't matter (and arguably showing our refs first
makes more sense).
Signed-off-by: Jeff King <peff@peff.net>
---
builtin/receive-pack.c | 4 +++-
t/t5400-send-pack.sh | 38 ++++++++++++++++++++++++++++++++++++++
2 files changed, 41 insertions(+), 1 deletion(-)
diff --git a/builtin/receive-pack.c b/builtin/receive-pack.c
index c55e2f993..bc7ce0ea2 100644
--- a/builtin/receive-pack.c
+++ b/builtin/receive-pack.c
@@ -268,6 +268,8 @@ static int show_ref_cb(const char *path_full, const struct object_id *oid,
if (oidset_insert(seen, oid))
return 0;
path = ".have";
+ } else {
+ oidset_insert(seen, oid);
}
show_ref(path, oid->hash);
return 0;
@@ -289,9 +291,9 @@ static void write_head_info(void)
{
static struct oidset seen = OIDSET_INIT;
+ for_each_ref(show_ref_cb, &seen);
for_each_alternate_ref(show_one_alternate_ref, &seen);
oidset_clear(&seen);
- for_each_ref(show_ref_cb, &seen);
if (!sent_capabilities)
show_ref("capabilities^{}", null_sha1);
diff --git a/t/t5400-send-pack.sh b/t/t5400-send-pack.sh
index 305ca7a93..3331e0f53 100755
--- a/t/t5400-send-pack.sh
+++ b/t/t5400-send-pack.sh
@@ -255,4 +255,42 @@ test_expect_success 'deny pushing to delete current branch' '
)
'
+extract_ref_advertisement () {
+ perl -lne '
+ # \\ is there to skip capabilities after \0
+ /push< ([^\\]+)/ or next;
+ exit 0 if $1 eq "0000";
+ print $1;
+ '
+}
+
+test_expect_success 'receive-pack de-dupes .have lines' '
+ git init shared &&
+ git -C shared commit --allow-empty -m both &&
+ git clone -s shared fork &&
+ (
+ cd shared &&
+ git checkout -b only-shared &&
+ git commit --allow-empty -m only-shared &&
+ git update-ref refs/heads/foo HEAD
+ ) &&
+
+ # Notable things in this expectation:
+ # - local refs are not de-duped
+ # - .have does not duplicate locals
+ # - .have does not duplicate itself
+ local=$(git -C fork rev-parse HEAD) &&
+ shared=$(git -C shared rev-parse only-shared) &&
+ cat >expect <<-EOF &&
+ $local refs/heads/master
+ $local refs/remotes/origin/HEAD
+ $local refs/remotes/origin/master
+ $shared .have
+ EOF
+
+ GIT_TRACE_PACKET=$(pwd)/trace git push fork HEAD:foo &&
+ extract_ref_advertisement <trace >refs &&
+ test_cmp expect refs
+'
+
test_done
--
2.11.0.765.g454d2182f
^ permalink raw reply related
* [PATCH 11/12] receive-pack: treat namespace .have lines like alternates
From: Jeff King @ 2017-01-24 0:48 UTC (permalink / raw)
To: git
In-Reply-To: <20170124003729.j4ygjcgypdq7hceg@sigill.intra.peff.net>
Namely, de-duplicate them. We use the same set as the
alternates, since we call them both ".have" (i.e., there is
no value in showing one versus the other).
Signed-off-by: Jeff King <peff@peff.net>
---
builtin/receive-pack.c | 10 +++++++---
1 file changed, 7 insertions(+), 3 deletions(-)
diff --git a/builtin/receive-pack.c b/builtin/receive-pack.c
index 8f8762e4a..c55e2f993 100644
--- a/builtin/receive-pack.c
+++ b/builtin/receive-pack.c
@@ -251,8 +251,9 @@ static void show_ref(const char *path, const unsigned char *sha1)
}
static int show_ref_cb(const char *path_full, const struct object_id *oid,
- int flag, void *unused)
+ int flag, void *data)
{
+ struct oidset *seen = data;
const char *path = strip_namespace(path_full);
if (ref_is_hidden(path, path_full))
@@ -263,8 +264,11 @@ static int show_ref_cb(const char *path_full, const struct object_id *oid,
* refs, so that the client can use them to minimize data
* transfer but will otherwise ignore them.
*/
- if (!path)
+ if (!path) {
+ if (oidset_insert(seen, oid))
+ return 0;
path = ".have";
+ }
show_ref(path, oid->hash);
return 0;
}
@@ -287,7 +291,7 @@ static void write_head_info(void)
for_each_alternate_ref(show_one_alternate_ref, &seen);
oidset_clear(&seen);
- for_each_ref(show_ref_cb, NULL);
+ for_each_ref(show_ref_cb, &seen);
if (!sent_capabilities)
show_ref("capabilities^{}", null_sha1);
--
2.11.0.765.g454d2182f
^ permalink raw reply related
* Re: [PATCH v2 25/27] attr: store attribute stack in attr_check structure
From: Brandon Williams @ 2017-01-24 1:11 UTC (permalink / raw)
To: Junio C Hamano; +Cc: git, sbeller, pclouds
In-Reply-To: <20170123220614.GA187696@google.com>
On 01/23, Brandon Williams wrote:
> As we discussed off-line I'll also do the rework to break up the
> question and result. That way two threads can be executing using the
> same attr_check structure.
Thinking about this I don't really see what we would gain by breaking
them up.
Right now most callers have a static attr_check struct which holds the
question and answer (and in my series a buffer of all_attrs used during
the collection process). If this struct is broken up into question and
answer then the only part of it that can be shared with multiple threads
is the question, which ends up being an array with 2 maybe 3 entries on
average. The result and the array of all_attrs would then need to be
allocated each time calling into the attribute system since they can't
be shared. Since this allocation is already going to happen wouldn't it
just make sense to drop the static modifier on the check structure (or
have a per-thread check structure) if you really wanted a particular
function thread safe? It seems like breaking the question and answer up
doesn't buy you much in terms of reducing allocation churn and instead
complicates the API with needing to keep track of two structures instead
of a one.
--
Brandon Williams
^ permalink raw reply
* Re: [PATCH 3/3] git-prompt.sh: fix for submodule 'dirty' indicator
From: brian m. carlson @ 2017-01-24 1:17 UTC (permalink / raw)
To: Benjamin Fuchs; +Cc: git, szeder.dev, sbeller
In-Reply-To: <1485113421-22264-1-git-send-email-email@benjaminfuchs.de>
[-- Attachment #1: Type: text/plain, Size: 1924 bytes --]
On Sun, Jan 22, 2017 at 08:30:21PM +0100, Benjamin Fuchs wrote:
> Fixing wrong git diff line.
This patch says 3/3, but I don't see 1 and 2. Also, this description
doesn't tell me what the problem is, or why this fix is useful. Such
information helps us down the line when looking at the history, and it
also helps reviewers determine whether your change makes sense.
Right now I can only guess that there's an issue with spaces or slashes
somehow. You might want to take a look at
Documentation/SubmittingPatches, especially point 2.
> ---
> contrib/completion/git-prompt.sh | 6 +++---
> 1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/contrib/completion/git-prompt.sh b/contrib/completion/git-prompt.sh
> index c44b9a2..43b28e9 100644
> --- a/contrib/completion/git-prompt.sh
> +++ b/contrib/completion/git-prompt.sh
> @@ -306,9 +306,9 @@ __git_ps1_submodule ()
> local submodule_name="$(basename "$git_dir")"
> if [ "$submodule_name" != ".git" ] && [ "$submodule_name" != "." ]; then
> local parent_top="${git_dir%.git*}"
> - local submodule_top="${git_dir#*modules}"
> + local submodule_top="${git_dir#*modules/}"
> local status=""
> - git diff -C "$parent_top" --no-ext-diff --ignore-submodules=dirty --quiet -- "$submodule_top" 2>/dev/null || status="+"
> + git -C "$parent_top" diff --no-ext-diff --ignore-submodules=dirty --quiet -- "$submodule_top" 2>/dev/null || status="+"
> printf "$status$submodule_name:"
> fi
> }
> @@ -544,7 +544,7 @@ __git_ps1 ()
>
> local sub=""
> if [ -n "${GIT_PS1_SHOWSUBMODULE}" ]; then
> - sub="$(__git_ps1_submodule $g)"
> + sub="$(__git_ps1_submodule "$g")"
> fi
>
> local f="$w$i$s$u"
> --
> 2.7.4
>
--
brian m. carlson / brian with sandals: Houston, Texas, US
+1 832 623 2791 | https://www.crustytoothpaste.net/~bmc | My opinion only
OpenPGP: https://keybase.io/bk2204
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 868 bytes --]
^ permalink raw reply
* Re: [PATCH 0/12] reducing resource usage of for_each_alternate_ref
From: Brandon Williams @ 2017-01-24 1:33 UTC (permalink / raw)
To: Jeff King; +Cc: git
In-Reply-To: <20170124003729.j4ygjcgypdq7hceg@sigill.intra.peff.net>
On 01/23, Jeff King wrote:
>
> A brief overview of the patches:
>
> [01/12]: for_each_alternate_ref: handle failure from real_pathdup()
> [02/12]: for_each_alternate_ref: stop trimming trailing slashes
> [03/12]: for_each_alternate_ref: use strbuf for path allocation
>
> Bugfixes and cleanups (the first one is actually a recent-ish
> regression).
Which is most likely my fault, Sorry! :)
I think the old behavior was to die and not return NULL.
--
Brandon Williams
^ permalink raw reply
* [PATCH] [draft]blame: add --aggregate option
From: Edmundo Carmona Antoranz @ 2017-01-24 2:10 UTC (permalink / raw)
To: git; +Cc: Edmundo Carmona Antoranz
---
builtin/blame.c | 78 +++++++++++++++++++++++++++++++++++++--------------------
1 file changed, 51 insertions(+), 27 deletions(-)
diff --git a/builtin/blame.c b/builtin/blame.c
index 126b8c9e5..9e8403303 100644
--- a/builtin/blame.c
+++ b/builtin/blame.c
@@ -1884,6 +1884,7 @@ static const char *format_time(unsigned long time, const char *tz_str,
#define OUTPUT_NO_AUTHOR 0200
#define OUTPUT_SHOW_EMAIL 0400
#define OUTPUT_LINE_PORCELAIN 01000
+#define OUTPUT_AGGREGATE 02000
static void emit_porcelain_details(struct origin *suspect, int repeat)
{
@@ -1931,43 +1932,36 @@ static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent,
putchar('\n');
}
-static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
+/**
+ * Print information about the revision.
+ * This information can be used in either aggregated output
+ * or prepending each line of the content of the file being blamed
+ */
+static void print_revision_info(char* revision_hex, int revision_length, struct blame_entry* ent,
+ struct commit* commit, struct commit_info ci, int opt, int show_raw_time)
{
- int cnt;
- const char *cp;
- struct origin *suspect = ent->suspect;
- struct commit_info ci;
- char hex[GIT_SHA1_HEXSZ + 1];
- int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
-
- get_commit_info(suspect->commit, &ci, 1);
- sha1_to_hex_r(hex, suspect->commit->object.oid.hash);
-
- cp = nth_line(sb, ent->lno);
- for (cnt = 0; cnt < ent->num_lines; cnt++) {
- char ch;
- int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? GIT_SHA1_HEXSZ : abbrev;
-
- if (suspect->commit->object.flags & UNINTERESTING) {
+ if (opt & OUTPUT_AGGREGATE)
+ printf("\t");
+ int length = revision_length;
+ if (commit->object.flags & UNINTERESTING) {
if (blank_boundary)
- memset(hex, ' ', length);
+ memset(revision_hex, ' ', length);
else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
length--;
putchar('^');
}
}
- printf("%.*s", length, hex);
+ printf("%.*s", length, revision_hex);
if (opt & OUTPUT_ANNOTATE_COMPAT) {
const char *name;
if (opt & OUTPUT_SHOW_EMAIL)
name = ci.author_mail.buf;
else
name = ci.author.buf;
- printf("\t(%10s\t%10s\t%d)", name,
+ printf("\t(%10s\t%10s\t", name,
format_time(ci.author_time, ci.author_tz.buf,
- show_raw_time),
- ent->lno + 1 + cnt);
+ show_raw_time));
} else {
if (opt & OUTPUT_SHOW_SCORE)
printf(" %*d %02d",
@@ -1975,11 +1969,7 @@ static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
ent->suspect->refcnt);
if (opt & OUTPUT_SHOW_NAME)
printf(" %-*.*s", longest_file, longest_file,
- suspect->path);
- if (opt & OUTPUT_SHOW_NUMBER)
- printf(" %*d", max_orig_digits,
- ent->s_lno + 1 + cnt);
-
+ ent->suspect->path);
if (!(opt & OUTPUT_NO_AUTHOR)) {
const char *name;
int pad;
@@ -1994,9 +1984,42 @@ static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
ci.author_tz.buf,
show_raw_time));
}
+ }
+ if (opt & OUTPUT_AGGREGATE)
+ printf(")\n");
+}
+
+static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
+{
+ int cnt;
+ const char *cp;
+ struct origin *suspect = ent->suspect;
+ struct commit_info ci;
+ char hex[GIT_SHA1_HEXSZ + 1];
+ int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
+ int revision_length = (opt & OUTPUT_LONG_OBJECT_NAME) ? GIT_SHA1_HEXSZ : abbrev;
+
+ get_commit_info(suspect->commit, &ci, 1);
+ sha1_to_hex_r(hex, suspect->commit->object.oid.hash);
+
+ if (opt & OUTPUT_AGGREGATE)
+ print_revision_info(hex, revision_length, ent, suspect->commit, ci, opt, show_raw_time);
+
+ cp = nth_line(sb, ent->lno);
+ for (cnt = 0; cnt < ent->num_lines; cnt++) {
+ if (!(opt & OUTPUT_AGGREGATE))
+ print_revision_info(hex, revision_length, ent, suspect->commit, ci, opt, show_raw_time);
+ if (opt & OUTPUT_ANNOTATE_COMPAT) {
+ printf("%*d) ",
+ max_digits, ent->lno + 1 + cnt);
+ } else {
+ if (opt & OUTPUT_SHOW_NUMBER)
+ printf(" %*d ", max_orig_digits,
+ ent->s_lno + 1 + cnt);
printf(" %*d) ",
max_digits, ent->lno + 1 + cnt);
}
+ char ch;
do {
ch = *cp++;
putchar(ch);
@@ -2609,6 +2632,7 @@ int cmd_blame(int argc, const char **argv, const char *prefix)
{ OPTION_CALLBACK, 'C', NULL, &opt, N_("score"), N_("Find line copies within and across files"), PARSE_OPT_OPTARG, blame_copy_callback },
{ OPTION_CALLBACK, 'M', NULL, &opt, N_("score"), N_("Find line movements within and across files"), PARSE_OPT_OPTARG, blame_move_callback },
OPT_STRING_LIST('L', NULL, &range_list, N_("n,m"), N_("Process only line range n,m, counting from 1")),
+ OPT_BIT(0, "aggregate", &output_option, N_("Aggregate output"), OUTPUT_AGGREGATE),
OPT__ABBREV(&abbrev),
OPT_END()
};
--
2.11.0.rc1
^ permalink raw reply related
* Re: [PATCH 0/12] reducing resource usage of for_each_alternate_ref
From: Jeff King @ 2017-01-24 2:12 UTC (permalink / raw)
To: Brandon Williams; +Cc: git
In-Reply-To: <20170124013341.GA185930@google.com>
On Mon, Jan 23, 2017 at 05:33:41PM -0800, Brandon Williams wrote:
> On 01/23, Jeff King wrote:
> >
> > A brief overview of the patches:
> >
> > [01/12]: for_each_alternate_ref: handle failure from real_pathdup()
> > [02/12]: for_each_alternate_ref: stop trimming trailing slashes
> > [03/12]: for_each_alternate_ref: use strbuf for path allocation
> >
> > Bugfixes and cleanups (the first one is actually a recent-ish
> > regression).
>
> Which is most likely my fault, Sorry! :)
>
> I think the old behavior was to die and not return NULL.
Yes, it is. :)
But I think it's probably pretty hard to trigger in practice. And on the
plus side, I think the new behavior after my patch is much more sensible
than even the original.
-Peff
^ permalink raw reply
* Re: [PATCH] [draft]blame: add --aggregate option
From: Edmundo Carmona Antoranz @ 2017-01-24 2:18 UTC (permalink / raw)
To: Git List; +Cc: Edmundo Carmona Antoranz
In-Reply-To: <20170124021046.30735-1-eantoranz@gmail.com>
Developers of the world, rejoice! :-)
Junio, Pranit (and whoever is paying attention to the conversation
that was being held about --tips), here's a draft of what I meant when
I was talking about the option of "aggregating" blame output. I'm not
considering _all_ cases yet, just would like for people to give it a
quick test and tell me if they think it's worth "polishing" it for
inclusion into mainline git.
The output would look like this:
$ ./git blame -L 1,19 -t --aggregate builtin/blame.c
Blaming lines: 0% (19/2974), done.
cee7f245dc builtin-pickaxe.c (Junio C Hamano 1161298804 -0700)
1) /*
31653c1abc builtin-blame.c (Eugene Letuchy 1235170271 -0800)
2) * Blame
cee7f245dc builtin-pickaxe.c (Junio C Hamano 1161298804 -0700)
3) *
7e6ac6e439 builtin/blame.c (David Kastrup 1398470209 +0200)
4) * Copyright (c) 2006, 2014 by its authors
5) * See COPYING for licensing conditions
cee7f245dc builtin-pickaxe.c (Junio C Hamano 1161298804 -0700)
6) */
7)
8) #include "cache.h"
fb58c8d507 builtin/blame.c (Michael Haggerty 1434981785 +0200)
9) #include "refs.h"
cee7f245dc builtin-pickaxe.c (Junio C Hamano 1161298804 -0700)
10) #include "builtin.h"
11) #include "blob.h"
12) #include "commit.h"
13) #include "tag.h"
14) #include "tree-walk.h"
15) #include "diff.h"
16) #include "diffcore.h"
17) #include "revision.h"
717d1462ba builtin-blame.c (Linus Torvalds 1169976846 -0800)
18) #include "quote.h"
cee7f245dc builtin-pickaxe.c (Junio C Hamano 1161298804 -0700)
19) #include "xdiff-interface.h"
It can be seen that options like -t still work in aggregation.
In relation to the previous conversation about "tips", I think a
better name could be "hints" and it could be added on top of the
aggregation.
On Mon, Jan 23, 2017 at 8:10 PM, Edmundo Carmona Antoranz
<eantoranz@gmail.com> wrote:
> ---
> builtin/blame.c | 78 +++++++++++++++++++++++++++++++++++++--------------------
> 1 file changed, 51 insertions(+), 27 deletions(-)
>
> diff --git a/builtin/blame.c b/builtin/blame.c
> index 126b8c9e5..9e8403303 100644
> --- a/builtin/blame.c
> +++ b/builtin/blame.c
> @@ -1884,6 +1884,7 @@ static const char *format_time(unsigned long time, const char *tz_str,
> #define OUTPUT_NO_AUTHOR 0200
> #define OUTPUT_SHOW_EMAIL 0400
> #define OUTPUT_LINE_PORCELAIN 01000
> +#define OUTPUT_AGGREGATE 02000
>
> static void emit_porcelain_details(struct origin *suspect, int repeat)
> {
> @@ -1931,43 +1932,36 @@ static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent,
> putchar('\n');
> }
>
> -static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
> +/**
> + * Print information about the revision.
> + * This information can be used in either aggregated output
> + * or prepending each line of the content of the file being blamed
> + */
> +static void print_revision_info(char* revision_hex, int revision_length, struct blame_entry* ent,
> + struct commit* commit, struct commit_info ci, int opt, int show_raw_time)
> {
> - int cnt;
> - const char *cp;
> - struct origin *suspect = ent->suspect;
> - struct commit_info ci;
> - char hex[GIT_SHA1_HEXSZ + 1];
> - int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
> -
> - get_commit_info(suspect->commit, &ci, 1);
> - sha1_to_hex_r(hex, suspect->commit->object.oid.hash);
> -
> - cp = nth_line(sb, ent->lno);
> - for (cnt = 0; cnt < ent->num_lines; cnt++) {
> - char ch;
> - int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? GIT_SHA1_HEXSZ : abbrev;
> -
> - if (suspect->commit->object.flags & UNINTERESTING) {
> + if (opt & OUTPUT_AGGREGATE)
> + printf("\t");
> + int length = revision_length;
> + if (commit->object.flags & UNINTERESTING) {
> if (blank_boundary)
> - memset(hex, ' ', length);
> + memset(revision_hex, ' ', length);
> else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
> length--;
> putchar('^');
> }
> }
>
> - printf("%.*s", length, hex);
> + printf("%.*s", length, revision_hex);
> if (opt & OUTPUT_ANNOTATE_COMPAT) {
> const char *name;
> if (opt & OUTPUT_SHOW_EMAIL)
> name = ci.author_mail.buf;
> else
> name = ci.author.buf;
> - printf("\t(%10s\t%10s\t%d)", name,
> + printf("\t(%10s\t%10s\t", name,
> format_time(ci.author_time, ci.author_tz.buf,
> - show_raw_time),
> - ent->lno + 1 + cnt);
> + show_raw_time));
> } else {
> if (opt & OUTPUT_SHOW_SCORE)
> printf(" %*d %02d",
> @@ -1975,11 +1969,7 @@ static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
> ent->suspect->refcnt);
> if (opt & OUTPUT_SHOW_NAME)
> printf(" %-*.*s", longest_file, longest_file,
> - suspect->path);
> - if (opt & OUTPUT_SHOW_NUMBER)
> - printf(" %*d", max_orig_digits,
> - ent->s_lno + 1 + cnt);
> -
> + ent->suspect->path);
> if (!(opt & OUTPUT_NO_AUTHOR)) {
> const char *name;
> int pad;
> @@ -1994,9 +1984,42 @@ static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
> ci.author_tz.buf,
> show_raw_time));
> }
> + }
> + if (opt & OUTPUT_AGGREGATE)
> + printf(")\n");
> +}
> +
> +static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
> +{
> + int cnt;
> + const char *cp;
> + struct origin *suspect = ent->suspect;
> + struct commit_info ci;
> + char hex[GIT_SHA1_HEXSZ + 1];
> + int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
> + int revision_length = (opt & OUTPUT_LONG_OBJECT_NAME) ? GIT_SHA1_HEXSZ : abbrev;
> +
> + get_commit_info(suspect->commit, &ci, 1);
> + sha1_to_hex_r(hex, suspect->commit->object.oid.hash);
> +
> + if (opt & OUTPUT_AGGREGATE)
> + print_revision_info(hex, revision_length, ent, suspect->commit, ci, opt, show_raw_time);
> +
> + cp = nth_line(sb, ent->lno);
> + for (cnt = 0; cnt < ent->num_lines; cnt++) {
> + if (!(opt & OUTPUT_AGGREGATE))
> + print_revision_info(hex, revision_length, ent, suspect->commit, ci, opt, show_raw_time);
> + if (opt & OUTPUT_ANNOTATE_COMPAT) {
> + printf("%*d) ",
> + max_digits, ent->lno + 1 + cnt);
> + } else {
> + if (opt & OUTPUT_SHOW_NUMBER)
> + printf(" %*d ", max_orig_digits,
> + ent->s_lno + 1 + cnt);
> printf(" %*d) ",
> max_digits, ent->lno + 1 + cnt);
> }
> + char ch;
> do {
> ch = *cp++;
> putchar(ch);
> @@ -2609,6 +2632,7 @@ int cmd_blame(int argc, const char **argv, const char *prefix)
> { OPTION_CALLBACK, 'C', NULL, &opt, N_("score"), N_("Find line copies within and across files"), PARSE_OPT_OPTARG, blame_copy_callback },
> { OPTION_CALLBACK, 'M', NULL, &opt, N_("score"), N_("Find line movements within and across files"), PARSE_OPT_OPTARG, blame_move_callback },
> OPT_STRING_LIST('L', NULL, &range_list, N_("n,m"), N_("Process only line range n,m, counting from 1")),
> + OPT_BIT(0, "aggregate", &output_option, N_("Aggregate output"), OUTPUT_AGGREGATE),
> OPT__ABBREV(&abbrev),
> OPT_END()
> };
> --
> 2.11.0.rc1
>
^ permalink raw reply
* Re: [PATCH v2 25/27] attr: store attribute stack in attr_check structure
From: Junio C Hamano @ 2017-01-24 2:28 UTC (permalink / raw)
To: Brandon Williams; +Cc: git, sbeller, pclouds
In-Reply-To: <20170124011135.GB187696@google.com>
Brandon Williams <bmwill@google.com> writes:
> ... It seems like breaking the question and answer up
> doesn't buy you much in terms of reducing allocation churn and instead
> complicates the API with needing to keep track of two structures instead
> of a one.
In my mind, the value of having a constant check_attr is primarily
that it gives us a stable pointer to serve as a hashmap key,
i.e. the identifier for each call site, in a later iteration.
Of course, in order to populate the "question" array, we'd need the
interning of attribute names to attr objects, which need to be
protected by mutex, and you would probably not want to do that every
time the control hits the codepath.
But all of the above comes from my intuition, and I'll very much
welcome to be proven wrong with an alternative design, or better
yet, a working code based on an alternative design ;-).
^ permalink raw reply
* Re: [PATCH v3 4/5] show-ref: Detect dangling refs under --verify as well
From: Junio C Hamano @ 2017-01-24 2:48 UTC (permalink / raw)
To: Vladimir Panteleev; +Cc: git
In-Reply-To: <20170123180059.4288-5-git@thecybershadow.net>
Vladimir Panteleev <git@thecybershadow.net> writes:
> Move detection of dangling refs into show_one, so that they are
> detected when --verify is present as well as when it is absent.
>
> Signed-off-by: Vladimir Panteleev <git@thecybershadow.net>
> ---
> builtin/show-ref.c | 16 ++++++++--------
> t/t1403-show-ref.sh | 22 ++++++++++++++++++++++
> 2 files changed, 30 insertions(+), 8 deletions(-)
>
> diff --git a/builtin/show-ref.c b/builtin/show-ref.c
> index ab8e0dc41..107d05fe0 100644
> --- a/builtin/show-ref.c
> +++ b/builtin/show-ref.c
> @@ -22,6 +22,14 @@ static void show_one(const char *refname, const struct object_id *oid)
> const char *hex;
> struct object_id peeled;
>
> + /* This changes the semantics slightly that even under quiet we
> + * detect and return error if the repository is corrupt and
> + * ref points at a nonexistent object.
> + */
This is my fault from more than 10 years ago, but I think the
comment shouldn't have been here (or at its original location). It
talks about the behaviour change relative to the previous version
when the comment was added, i.e. cf0adba788 ("Store peeled refs in
packed-refs file.", 2006-11-19).
I'll remove it after the series settles.
^ permalink raw reply
* [PATCH] gitk: use right colour for remote refs in the "Tags and heads" dialog
From: Paul Wise @ 2017-01-24 6:22 UTC (permalink / raw)
To: git; +Cc: Paul Mackerras, Paul Wise
Makes it easier to see which refs are local and which refs are remote.
Adds consistency with the remote background colour in the graph display.
Signed-off-by: Paul Wise <pabs3@bonedaddy.net>
---
gitk-git/gitk | 9 ++++++++-
1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/gitk-git/gitk b/gitk-git/gitk
index a14d7a16b..14aebc23e 100755
--- a/gitk-git/gitk
+++ b/gitk-git/gitk
@@ -3404,6 +3404,8 @@ set rectmask {
}
image create bitmap reficon-H -background black -foreground "#00ff00" \
-data $rectdata -maskdata $rectmask
+image create bitmap reficon-R -background black -foreground "#ffddaa" \
+ -data $rectdata -maskdata $rectmask
image create bitmap reficon-o -background black -foreground "#ddddff" \
-data $rectdata -maskdata $rectmask
@@ -10022,6 +10024,7 @@ proc sel_reflist {w x y} {
set n [lindex $ref 0]
switch -- [lindex $ref 1] {
"H" {selbyid $headids($n)}
+ "R" {selbyid $headids($n)}
"T" {selbyid $tagids($n)}
"o" {selbyid $otherrefids($n)}
}
@@ -10051,7 +10054,11 @@ proc refill_reflist {} {
foreach n [array names headids] {
if {[string match $reflistfilter $n]} {
if {[commitinview $headids($n) $curview]} {
- lappend refs [list $n H]
+ if {[string match "remotes/*" $n]} {
+ lappend refs [list $n R]
+ } else {
+ lappend refs [list $n H]
+ }
} else {
interestedin $headids($n) {run refill_reflist}
}
--
2.11.0
^ permalink raw reply related
* Re: [PATCH] rebase: pass --signoff option to git am
From: Giuseppe Bilotta @ 2017-01-24 7:06 UTC (permalink / raw)
To: Junio C Hamano; +Cc: Git List
In-Reply-To: <xmqqsho9pdbc.fsf@gitster.mtv.corp.google.com>
On Tue, Jan 24, 2017 at 12:27 AM, Junio C Hamano <gitster@pobox.com> wrote:
> Giuseppe Bilotta <giuseppe.bilotta@gmail.com> writes:
>>
>> I'm not sure I follow. If the user doesn't want to signoff during a
>> rebase, they can simply not pass --signoff. If they do, they can not
>> pass it. Am I missing something?
>
> alias.
>
> Which also means that there needs to be --no-signoff option that can
> be given to countermand an earlier --signoff, if a user did
>
> [alias] rb = rebase --signoff
>
> and wants to disable it one time only with
>
> $ git rb --no-signoff
Oh, right, good point. This should be easy, I'll give this a go.
>>> In any case, will queue as-is so that we won't lose the patch while
>>> waiting for people to raise their opinions.
>>
>> Thanks.
>
> Thanks. The final version would also need tests, so it may be a
> good time to start thinking about what aspect of this feature wants
> to be protected against future breakages.
I have troubles thinking how it could go wrong. The most obvious
thing I can think of is it could not be remembered after an
interruption+continue. I'll think about this some more.
--
Giuseppe "Oblomov" Bilotta
^ permalink raw reply
* Re: [PATCH 7/7] completion: recognize more long-options
From: Johannes Sixt @ 2017-01-24 7:15 UTC (permalink / raw)
To: Cornelius Weig
Cc: bitte.keine.werbung.einwerfen, git, thomas.braun, szeder, john
In-Reply-To: <20170122225724.19360-8-cornelius.weig@tngtech.com>
If at all possible, please use your real email address as the From
address. It is pointless to hide behind a fake address because as Git
contributor you will have to reveal your identity anyway.
Please study item (5) "Sign your work" in
Documentation/SubmittingPatches and sign off your work.
I'm by no means a bash completion expert, but I have a comment on this
patch.
Am 22.01.2017 um 23:57 schrieb bitte.keine.werbung.einwerfen@googlemail.com:
> Recognize several new long-options for bash completion in the following
> commands:
AFAIR, it was a deliberate decision that potentially destructive command
line options are not included in command completions. In the list given,
I find these:
> - apply: --unsafe-paths
> - reset: --merge --mixed --hard --soft --patch --keep
> - rm: --force
Additionally, these options are only for internal purposes, but I may be
wrong:
> - archive: --remote= --exec=
-- Hannes
^ permalink raw reply
* Re: [PATCH 7/7] completion: recognize more long-options
From: Cornelius Weig @ 2017-01-24 8:14 UTC (permalink / raw)
To: Johannes Sixt; +Cc: bitte.keine.werbung.einwerfen, git, thomas.braun, john
In-Reply-To: <74ecd09c-55da-3858-5187-52c286a6bf62@kdbg.org>
On 01/24/2017 08:15 AM, Johannes Sixt wrote:
> If at all possible, please use your real email address as the From
> address. It is pointless to hide behind a fake address because as Git
> contributor you will have to reveal your identity anyway.
These are both real addresses, but for send-mail I would not want to use
my work account. I hope this is not a problem.
> Please study item (5) "Sign your work" in
> Documentation/SubmittingPatches and sign off your work.
I followed the recommendations to submitting work, and in the first
round signing is discouraged.
> AFAIR, it was a deliberate decision that potentially destructive command
> line options are not included in command completions. In the list given,
> I find these:
>
>> - reset: --merge --mixed --hard --soft --patch --keep
My bad, I only added --keep, which should be fine. As to these options
>> - apply: --unsafe-paths
>> - rm: --force
let's wait for further comments, but I won't cling to it.
> Additionally, these options are only for internal purposes, but I may be
> wrong:
>
>> - archive: --remote= --exec=
These may in fact be too exotic and just clutter the command line. Best
they are removed.
-- Cornelius
^ permalink raw reply
page: next (older) | prev (newer) | latest
- recent:[subjects (threaded)|topics (new)|topics (active)]
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox