[PATCH] add: add --bulk to index all objects into a pack file

Subsystems: documentation, the rest

STALE3724d

3 messages, 1 author, 2016-06-15 · open the first message on its own page

[PATCH] add: add --bulk to index all objects into a pack file

From: Nguyễn Thái Ngọc Duy <hidden>
Date: 2016-06-15 22:58:56

The use case is

    tar -xzf bigproject.tar.gz
    cd bigproject
    git init
    git add .
    # git grep or something

The first add will generate a bunch of loose objects. With --bulk, all
of them are forced into a single pack instead, less clutter on disk
and maybe faster object access.

This is the equivalent of "git -c core.bigFileThreshold=0 add ." so
it's not really a new functionality. Just some convenient and public
exposure.

Signed-off-by: Nguyễn Thái Ngọc Duy <redacted>
---
 Interestingly index_stream() seems a bit slower than standard
 index_core(). Perhaps mmap() is faster than a series of read() for
 small files. Room for improvement later.

 Documentation/git-add.txt |  6 ++++++
 builtin/add.c             | 10 +++++++++-
 2 files changed, 15 insertions(+), 1 deletion(-)
diff --git a/Documentation/git-add.txt b/Documentation/git-add.txt
index 48754cb..36a77f6 100644
--- a/Documentation/git-add.txt
+++ b/Documentation/git-add.txt
@@ -160,6 +160,12 @@ today's "git add <pathspec>...", ignoring removed files.
 	be ignored, no matter if they are already present in the work
 	tree or not.
 
+--bulk::
+	Normally new objects are indexed and stored in loose format,
+	one file per new object in "$GIT_DIR/objects". This option
+	forces putting all objects into a single new pack. This may
+	be useful when you need to add a lot of files initially.
+
 \--::
 	This option can be used to separate command-line options from
 	the list of files, (useful when filenames might be mistaken
diff --git a/builtin/add.c b/builtin/add.c
index 226f758..40cbb71 100644
--- a/builtin/add.c
+++ b/builtin/add.c
@@ -336,7 +336,7 @@ static struct lock_file lock_file;
 static const char ignore_error[] =
 N_("The following paths are ignored by one of your .gitignore files:\n");
 
-static int verbose, show_only, ignored_too, refresh_only;
+static int verbose, show_only, ignored_too, refresh_only, bulk_index;
 static int ignore_add_errors, intent_to_add, ignore_missing;
 
 #define ADDREMOVE_DEFAULT 0 /* Change to 1 in Git 2.0 */
@@ -368,6 +368,7 @@ static struct option builtin_add_options[] = {
 	OPT_BOOL( 0 , "refresh", &refresh_only, N_("don't add, only refresh the index")),
 	OPT_BOOL( 0 , "ignore-errors", &ignore_add_errors, N_("just skip files which cannot be added because of errors")),
 	OPT_BOOL( 0 , "ignore-missing", &ignore_missing, N_("check if - even missing - files are ignored in dry run")),
+	OPT_BOOL( 0 , "bulk", &bulk_index, N_("pack all objects instead of creating loose ones")),
 	OPT_END(),
 };
 
@@ -560,6 +561,13 @@ int cmd_add(int argc, const char **argv, const char *prefix)
 		free(seen);
 	}
 
+	if (bulk_index)
+		/*
+		 * Pretend all blobs are "large" files, forcing them
+		 * all into a pack
+		 */
+		big_file_threshold = 0;
+
 	plug_bulk_checkin();
 
 	if ((flags & ADD_CACHE_IMPLICIT_DOT) && prefix) {
-- 
1.8.2.82.gc24b958

[PATCH v2] add: add --bulk to index all objects into a pack file

From: Nguyễn Thái Ngọc Duy <hidden>
Date: 2016-06-15 22:58:57

The use case is

    tar -xzf bigproject.tar.gz
    cd bigproject
    git init
    git add .
    # git grep or something

The first add will generate a bunch of loose objects. With --bulk, all
of them are forced into a single pack instead, less clutter on disk
and maybe faster object access.

On gdb-7.5.1 source directory, the loose .git directory takes 66M
according to `du` while the packed one takes 32M. Timing of
"git grep --cached":

          loose     packed
real    0m1.671s   0m1.372s
user    0m1.542s   0m1.313s
sys     0m0.126s   0m0.056s

It's not an all-win situation though. --bulk is slower than --no-bulk
because:

 - Triple hashing: we need to calculate both object SHA-1s _and_ pack
   SHA-1. At the end we have to fix up the pack, which means rehashing
   the entire pack again. --no-bulk only cares about object SHA-1s.

 - We write duplicate objects to the pack then truncate it, because we
   don't know if it's a duplicate until we're done writing, and cannot
   keep it in core because it's potentially big. So extra I/O (but
   hopefully not too much because duplicate objects should not happen
   often).

 - Sort and write .idx file.

 - (For the future) --no-bulk could benefit from multithreading
   because this is CPU bound operation. --bulk could not.

But again this comparison is not fair, --bulk is closer to:

    git add . &&
    git ls-files --stage | awk '{print $2;}'| \
        git pack-objects .git/objects/pack-

except that it does not deltifies nor sort objects.

Signed-off-by: Nguyễn Thái Ngọc Duy <redacted>
---
 v2 examines pros and cons of --bulk and I'm not sure if turning it on
 automatically (with heuristics) is a good idea anymore.

 Oh and it fixes not packing empty files.

 Documentation/git-add.txt | 10 ++++++++++
 builtin/add.c             | 10 +++++++++-
 sha1_file.c               |  3 ++-
 3 files changed, 21 insertions(+), 2 deletions(-)
diff --git a/Documentation/git-add.txt b/Documentation/git-add.txt
index 48754cb..147d191 100644
--- a/Documentation/git-add.txt
+++ b/Documentation/git-add.txt
@@ -160,6 +160,16 @@ today's "git add <pathspec>...", ignoring removed files.
 	be ignored, no matter if they are already present in the work
 	tree or not.
 
+--bulk::
+	Normally new objects are indexed and stored in loose format,
+	one file per new object in "$GIT_DIR/objects". This option
+	forces putting all objects into a single new pack. This may
+	be useful when you need to add a lot of files initially.
++
+This option is equivalent to running `git -c core.bigFileThreshold=0 add`.
+If you want to only pack files larger than a size threshold, use the
+long form.
+
 \--::
 	This option can be used to separate command-line options from
 	the list of files, (useful when filenames might be mistaken
diff --git a/builtin/add.c b/builtin/add.c
index 226f758..40cbb71 100644
--- a/builtin/add.c
+++ b/builtin/add.c
@@ -336,7 +336,7 @@ static struct lock_file lock_file;
 static const char ignore_error[] =
 N_("The following paths are ignored by one of your .gitignore files:\n");
 
-static int verbose, show_only, ignored_too, refresh_only;
+static int verbose, show_only, ignored_too, refresh_only, bulk_index;
 static int ignore_add_errors, intent_to_add, ignore_missing;
 
 #define ADDREMOVE_DEFAULT 0 /* Change to 1 in Git 2.0 */
@@ -368,6 +368,7 @@ static struct option builtin_add_options[] = {
 	OPT_BOOL( 0 , "refresh", &refresh_only, N_("don't add, only refresh the index")),
 	OPT_BOOL( 0 , "ignore-errors", &ignore_add_errors, N_("just skip files which cannot be added because of errors")),
 	OPT_BOOL( 0 , "ignore-missing", &ignore_missing, N_("check if - even missing - files are ignored in dry run")),
+	OPT_BOOL( 0 , "bulk", &bulk_index, N_("pack all objects instead of creating loose ones")),
 	OPT_END(),
 };
 
@@ -560,6 +561,13 @@ int cmd_add(int argc, const char **argv, const char *prefix)
 		free(seen);
 	}
 
+	if (bulk_index)
+		/*
+		 * Pretend all blobs are "large" files, forcing them
+		 * all into a pack
+		 */
+		big_file_threshold = 0;
+
 	plug_bulk_checkin();
 
 	if ((flags & ADD_CACHE_IMPLICIT_DOT) && prefix) {
diff --git a/sha1_file.c b/sha1_file.c
index f80bbe4..8b66840 100644
--- a/sha1_file.c
+++ b/sha1_file.c
@@ -3137,7 +3137,8 @@ int index_fd(unsigned char *sha1, int fd, struct stat *st,
 
 	if (!S_ISREG(st->st_mode))
 		ret = index_pipe(sha1, fd, type, path, flags);
-	else if (size <= big_file_threshold || type != OBJ_BLOB ||
+	else if ((big_file_threshold && size <= big_file_threshold) ||
+		 type != OBJ_BLOB ||
 		 (path && would_convert_to_git(path, NULL, 0, 0)))
 		ret = index_core(sha1, fd, size, type, path, flags);
 	else
-- 
1.8.2.82.gc24b958

Re: [PATCH v2] add: add --bulk to index all objects into a pack file

From: Duy Nguyen <hidden>
Date: 2016-06-15 22:58:57

On Fri, Oct 4, 2013 at 1:57 PM, Nguyễn Thái Ngọc Duy [off-list ref] wrote:
It's not an all-win situation though. --bulk is slower than --no-bulk
because:

 - Triple hashing: we need to calculate both object SHA-1s _and_ pack
   SHA-1. At the end we have to fix up the pack, which means rehashing
   the entire pack again. --no-bulk only cares about object SHA-1s.
Oops, it's quadruple hashing: one for object sha-1 (1), one for pack
sha-1 while writing pack (2), two for pack header fixup: one after the
the header fixup (3), and one repeating the same (2) just to verify
that it matches (2) exactly (4). Do we have to be this paranoid? I
think we could drop (2) and (4) in this case.
-- 
Duy
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help