Add a command line argument to the git push command to request atomic
pushes.
Signed-off-by: Ronnie Sahlberg <sahlberg@google.com>
Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This adds support to send-pack to negotiate and use atomic pushes
iff the server supports it. Atomic pushes are activated by a new command
line flag --atomic.
In order to do this we also need to change the semantics for send_pack()
slightly. The existing send_pack() function actually doesn't send all the
refs back to the server when multiple refs are involved, for example
when using --all. Several of the failure modes for pushes can already be
detected locally in the send_pack client based on the information from the
initial server side list of all the refs as generated by receive-pack.
Any such refs that we thus know would fail to push are thus pruned from
the list of refs we send to the server to update.
For atomic pushes, we have to deal thus with both failures that are detected
locally as well as failures that are reported back from the server. In order
to do so we treat all local failures as push failures too.
We introduce a new status code REF_STATUS_ATOMIC_PUSH_FAILED so we can
flag all refs that we would normally have tried to push to the server
but we did not due to local failures. This is to improve the error message
back to the end user to flag that "these refs failed to update since the
atomic push operation failed."
Signed-off-by: Ronnie Sahlberg <sahlberg@google.com>
Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This adds the atomic protocol option to allow
receive-pack to inform the client that it has
atomic push capability.
This commit makes the functionality introduced
in the previous commits go live for the serving
side. The changes in documentation reflect the
protocol capabilities of the server.
Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This introduces the new function execute_commands_atomic which will use
one atomic transaction for all updates. The default behavior is still
the old non atomic way, one ref at a time. This is to cause as little
disruption as possible to existing clients. It is unknown if there are
client scripts that depend on the old non-atomic behavior so we make it
opt-in for now.
A later patch will add the possibility to actually use the functionality
added by this patch. For now use_atomic is always 0.
Inspired-by: Ronnie Sahlberg <sahlberg@google.com>
Helped-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This moves all code related to transactions into the
execute_commands_non_atomic function. This includes
beginning and committing the transaction as well as
dealing with the errors which may occur during the
begin and commit phase of a transaction.
No functional changes intended.
Helped-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This commit allows us in a later patch to easily distinguish between
the non atomic way to update the received refs and the atomic way which
is introduced in a later patch.
Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Discussion on the previous patch revealed we rather want to err on the
safe side. To do so we need to stop receive-pack in case of the possible
future bug when connectivity is not checked on a shallow push.
Also while touching that code we considered that removing the reported
refs may be harmful in some situations. Sound the message more like a
"This Cannot Happen, Please Investigate!" instead of giving advice to
remove refs.
Suggested-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Make the main "execute_commands" loop in receive-pack easier to read
by splitting out some steps into helper functions. The new helper
'should_process_cmd' checks if a ref update is unnecessary, whether
due to an error having occurred or for another reason. The helper
'warn_if_skipped_connectivity_check' warns if we have forgotten to
run a connectivity check on a ref which is shallow for the client
which would be a bug.
This will help us to duplicate less code in a later patch when we make
a second copy of the "execute_commands" loop.
No functional change intended.
Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Move expire_reflog() into refs.c and rename it to reflog_expire().
Turn the three policy functions into function pointers that are passed
into reflog_expire(). Add function prototypes and documentation to
refs.h.
[jc: squashed in $gmane/261582, drop "extern" in function definition]
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: Stefan Beller <sbeller@google.com>
Tweaked-by: Ramsay Jones
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Now that expire_reflog() doesn't actually look in the
expire_reflog_policy_cb data structure, we can make it opaque:
* Change the callers of expire_reflog() to pass it a pointer to an
entire "struct expire_reflog_policy_cb" rather than a pointer to a
"struct cmd_reflog_expire_cb".
* Change expire_reflog() to accept the argument as a "void *" and
simply pass it through to the policy functions.
* Change the policy functions, reflog_expiry_prepare(),
reflog_expiry_cleanup(), and should_expire_reflog_ent(), to accept
"void *cb_data" arguments and cast them back to "struct
expire_reflog_policy_cb" internally.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
These members are not needed by the policy functions.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The policy objects don't care about "--rewrite". So move it to
expire_reflog()'s flags parameter.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The policy objects don't care about "--verbose". So move it to
expire_reflog()'s flags parameter.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add a flags field to "struct expire_reflog_cb", and pass the flags
argument through to expire_reflog_ent(). In a moment we will start
using it to pass through flags that expire_reflog_ent() needs.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add a new data type, "struct expire_reflog_cb", for holding the data
that expire_reflog() passes to expire_reflog_ent() via
for_each_reflog_ent(). For now it only holds a pointer to a "struct
expire_reflog_policy_cb", which still contains all of the actual data.
In future commits we will move some fields from the latter to the
former.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is the first step towards separating the data needed by the
policy code from the data needed by the reflog expiration machinery.
(In a moment we will add a *new* "struct expire_reflog_cb" for the use
of expire_reflog() itself, then move fields selectively from
expire_reflog_policy_cb to expire_reflog_cb.)
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The policy objects don't care about "--updateref". So move it to
expire_reflog()'s flags parameter.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The policy objects don't care about "--dry-run". So move it to
expire_reflog()'s flags parameter.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We want to separate the options relevant to the expiry machinery from
the options affecting the expiration policy. So add a "flags" argument
to expire_reflog() to hold the former.
The argument doesn't yet do anything.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Extract two functions, reflog_expiry_prepare() and
reflog_expiry_cleanup(), from expire_reflog(). This is a further step
towards separating the code for deciding on expiration policy from the
code that manages the physical deletion of reflog entries.
This change requires a couple of local variables from expire_reflog()
to be turned into fields of "struct expire_reflog_cb". More
reorganization of the callback data will follow in later commits.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Extract from expire_reflog_ent() a function that is solely responsible
for deciding whether a reflog entry should be expired. By separating
this "business logic" from the mechanics of actually expiring entries,
we are working towards the goal of encapsulating reflog expiry within
the refs API, with policy decided by a callback function passed to it
by its caller.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We don't actually need the locking functionality, because we already
hold the lock on the reference itself, which is how the reflog file is
locked. But the lock_file code can do some of the bookkeeping for us,
and it is more careful than the old code here was. For example:
* It correctly handles the case that the reflog lock file already
exists for some reason or cannot be opened.
* It correctly cleans up the lockfile if the program dies.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There is very little cleanup needed if the reference has no reflog. If
we move the initialization of log_file down a bit, there's even less.
So instead of jumping to the cleanup code at the end of the function,
just do the cleanup and return inline.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is our usual convention.
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Reviewed-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Prior to v1.5.4~14, expire_reflog() had to be an each_ref_fn because
it was passed to for_each_reflog(). Since then, there has been no
reason for it to implement the each_ref_fn interface. So...
* Remove the "unused" parameter (which took the place of "flags", but
was really unused).
* Declare the last parameter to be (struct cmd_reflog_expire_cb *)
rather than (void *).
Helped-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Corner-case bugfixes for "git fetch" around reflog handling.
* jk/fetch-reflog-df-conflict:
ignore stale directories when checking reflog existence
fetch: load all default config at startup
When we start the git-fetch program, we call git_config to
load all config, but our callback only processes the
fetch.prune option; we do not chain to git_default_config at
all.
This means that we may not load some core configuration
which will have an effect. For instance, we do not load
core.logAllRefUpdates, which impacts whether or not we
create reflogs in a bare repository.
Note that I said "may" above. It gets even more exciting. If
we have to transfer actual objects as part of the fetch,
then we call fetch_pack as part of the same process. That
function loads its own config, which does chain to
git_default_config, impacting global variables which are
used by the rest of fetch. But if the fetch is a pure ref
update (e.g., a new ref which is a copy of an old one), we
skip fetch_pack entirely. So we get inconsistent results
depending on whether or not we have actual objects to
transfer or not!
Let's just load the core config at the start of fetch, so we
know we have it (we may also load it again as part of
fetch_pack, but that's OK; it's designed to be idempotent).
Our tests check both cases (with and without a pack). We
also check similar behavior for push for good measure, but
it already works as expected.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Tighten the logic to decide that an unreachable cruft is
sufficiently old by covering corner cases such as an ancient object
becoming reachable and then going unreachable again, in which case
its retention period should be prolonged.
* jk/prune-mtime: (28 commits)
drop add_object_array_with_mode
revision: remove definition of unused 'add_object' function
pack-objects: double-check options before discarding objects
repack: pack objects mentioned by the index
pack-objects: use argv_array
reachable: use revision machinery's --indexed-objects code
rev-list: add --indexed-objects option
rev-list: document --reflog option
t5516: test pushing a tag of an otherwise unreferenced blob
traverse_commit_list: support pending blobs/trees with paths
make add_object_array_with_context interface more sane
write_sha1_file: freshen existing objects
pack-objects: match prune logic for discarding objects
pack-objects: refactor unpack-unreachable expiration check
prune: keep objects reachable from recent objects
sha1_file: add for_each iterators for loose and packed objects
count-objects: use for_each_loose_file_in_objdir
count-objects: do not use xsize_t when counting object size
prune-packed: use for_each_loose_file_in_objdir
reachable: mark index blobs as SEEN
...
If the asynchronous start of copy_to_sideband() fails, then any
env_array entries added to struct child_process proc by
prepare_push_cert_sha1() are leaked. Call the latter function only
after start_async() succeeded so that the allocated entries are
cleaned up automatically by start_command() or finish_command().
Signed-off-by: Rene Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Allow us build with NO_PTHREADS=NoThanks compilation option.
* eb/no-pthreads:
Handle atexit list internaly for unthreaded builds
pack-objects: set number of threads before checking and warning
index-pack: fix compilation with NO_PTHREADS
Add managed "env" array to child_process to clarify the lifetime
rules.
* rs/run-command-env-array:
use env_array member of struct child_process
run-command: add env_array, an optional argv_array for env
Splitting pack-objects output into multiple packs is incompatible
with the use of reachability bitmap.
* jk/pack-objects-no-bitmap-when-splitting:
pack-objects: turn off bitmaps when we split packs
push --signed promises to take user.signingkey as the signing key but
fails to read the config.
Make it do so.
Signed-off-by: Michael J Gruber <git@drmicha.warpmail.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The API to update refs have been restructured to allow introducing
a true transactional updates later. We would even allow storing
refs in backends other than the traditional filesystem-based one.
* rs/ref-transaction: (25 commits)
ref_transaction_commit: bail out on failure to remove a ref
lockfile: remove unable_to_lock_error
refs.c: do not permit err == NULL
remote rm/prune: print a message when writing packed-refs fails
for-each-ref: skip and warn about broken ref names
refs.c: allow listing and deleting badly named refs
test: put tests for handling of bad ref names in one place
packed-ref cache: forbid dot-components in refnames
branch -d: simplify by using RESOLVE_REF_READING
branch -d: avoid repeated symref resolution
reflog test: test interaction with detached HEAD
refs.c: change resolve_ref_unsafe reading argument to be a flags field
refs.c: make write_ref_sha1 static
fetch.c: change s_update_ref to use a ref transaction
refs.c: ref_transaction_commit: distinguish name conflicts from other errors
refs.c: pass a list of names to skip to is_refname_available
refs.c: call lock_ref_sha1_basic directly from commit
refs.c: refuse to lock badly named refs in lock_ref_sha1_basic
rename_ref: don't ask read_ref_full where the ref came from
refs.c: pass the ref log message to _create/delete/update instead of _commit
...
A new filter to programatically edit the tail end of the commit log
messages.
* cc/interpret-trailers:
Documentation: add documentation for 'git interpret-trailers'
trailer: add tests for commands in config file
trailer: execute command from 'trailer.<name>.command'
trailer: add tests for "git interpret-trailers"
trailer: add interpret-trailers command
trailer: put all the processing together and print
trailer: parse trailers from file or stdin
trailer: process command line trailer arguments
trailer: read and process config information
trailer: process trailers from input message and arguments
trailer: add data structures and basic functions
Wrap atexit()s calls on unthreaded builds to handle callback list
internally.
This is needed because on unthreaded builds, asyncs inherits parent's
atexit() list, that gets run as soon as the async exit()s (and again at
the end of async's parent process). That led to remove temporary files
too early.
Also remove a by-atexit-callback guard against this kind of issue in
clone.c, as this patch makes it redundant.
Fixes test 5537 (temporary shallow file vanished before unpack-objects
could open it)
BTW remove an unused variable in shallow.c.
Helped-by: Duy Nguyen <pclouds@gmail.com>
Helped-by: Andreas Schwab <schwab@linux-m68k.org>
Helped-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Etienne Buira <etienne.buira@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Convert users of struct child_process to using the managed env_array for
specifying environment variables instead of supplying an array on the
stack or bringing their own argv_array. This shortens and simplifies
the code and ensures automatically that the allocated memory is freed
after use.
Signed-off-by: Rene Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If a pack.packSizeLimit is set, we may split the pack data
across multiple packfiles. This means we cannot generate
.bitmap files, as they require that all of the reachable
objects are in the same pack. We check that condition when
we are generating the list of objects to pack (and disable
bitmaps if we are not packing everything), but we forgot to
update it when we notice that we needed to split (which
doesn't happen until the actual write phase).
The resulting bitmaps are quite bogus (they mention entries
that do not exist in the pack!) and can cause a fetch or
push to send insufficient objects.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we are given an expiration time like
--unpack-unreachable=2.weeks.ago, we avoid writing out old,
unreachable loose objects entirely, under the assumption
that running "prune" would simply delete them immediately
anyway. However, this is only valid if we computed the same
set of reachable objects as prune would.
In practice, this is the case, because only git-repack uses
the --unpack-unreachable option with an expiration, and it
always feeds as many objects into the pack as possible. But
we can double-check at runtime just to be sure.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we pack all objects, we use only the objects reachable
from references and reflogs. This misses any objects which
are reachable from the index, but not yet referenced.
By itself this isn't a big deal; the objects can remain
loose until they are actually used in a commit. However, it
does create a problem when we drop packed but unreachable
objects. We try to optimize out the writing of objects that
we will immediately prune, which means we must follow the
same rules as prune in determining what is reachable. And
prune uses the index for this purpose.
This is rather uncommon in practice, as objects in the index
would not usually have been packed in the first place. But
it could happen in a sequence like:
1. You make a commit on a branch that references blob X.
2. You repack, moving X into the pack.
3. You delete the branch (and its reflog), so that X is
unreferenced.
4. You "git add" blob X so that it is now referenced only
by the index.
5. You repack again with git-gc. The pack-objects we
invoke will see that X is neither referenced nor
recent and not bother loosening it.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This saves us from having to bump the rp_av count when we
add new traversal options.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git help everyday" to show the Everyday Git document.
* po/everyday-doc:
doc: add 'everyday' to 'git help'
doc: Makefile regularise OBSOLETE_HTML list building
doc: modernise everyday.txt wording and format in man page style
When you resolve a sha1, you can optionally keep any context
found during the resolution, including the path and mode of
a tree entry (e.g., when looking up "HEAD:subdir/file.c").
The add_object_array_with_context function lets you then
attach that context to an entry in a list. Unfortunately,
the interface for doing so is horrible. The object_context
structure is large and most object_array users do not use
it. Therefore we keep a pointer to the structure to avoid
burdening other users too much. But that means when we do
use it that we must allocate the struct ourselves. And the
struct contains a fixed PATH_MAX-sized buffer, which makes
this wholly unsuitable for any large arrays.
We can observe that there is only a single user of the
"with_context" variant: builtin/grep.c. And in that use
case, the only element we care about is the path. We can
therefore store only the path as a pointer (the context's
mode field was redundant with the object_array_entry itself,
and nobody actually cared about the surrounding tree). This
still requires a strdup of the pathname, but at least we are
only consuming the minimum amount of memory for each string.
We can also handle the copying ourselves in
add_object_array_*, and free it as appropriate in
object_array_release_entry.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A recent commit taught git-prune to keep non-recent objects
that are reachable from recent ones. However, pack-objects,
when loosening unreachable objects, tries to optimize out
the write in the case that the object will be immediately
pruned. It now gets this wrong, since its rule does not
reflect the new prune code (and this can be seen by running
t6501 with a strategically placed repack).
Let's teach pack-objects similar logic.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we are loosening unreachable packed objects, we do not
bother to process objects that would simply be pruned
immediately anyway. The "would be pruned" check is a simple
comparison, but is about to get more complicated. Let's pull
it out into a separate function.
Note that this is slightly less efficient than the original,
which avoided even opening old packs, since no object in
them could pass the current check, which cares only about
the pack mtime. But the new rules will depend on the exact
object, so we need to perform the check even for old packs.
Note also that we fix a minor buglet when the pack mtime is
exactly the same as the expiration time. The prune code
considers that worth pruning, whereas our check here
considered it worth keeping. This wasn't a big deal. Besides
being unlikely to happen, the result was simply that the
object was loosened and then pruned, missing the
optimization. Still, we can easily fix it while we are here.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Our current strategy with prune is that an object falls into
one of three categories:
1. Reachable (from ref tips, reflogs, index, etc).
2. Not reachable, but recent (based on the --expire time).
3. Not reachable and not recent.
We keep objects from (1) and (2), but prune objects in (3).
The point of (2) is that these objects may be part of an
in-progress operation that has not yet updated any refs.
However, it is not always the case that objects for an
in-progress operation will have a recent mtime. For example,
the object database may have an old copy of a blob (from an
abandoned operation, a branch that was deleted, etc). If we
create a new tree that points to it, a simultaneous prune
will leave our tree, but delete the blob. Referencing that
tree with a commit will then work (we check that the tree is
in the object database, but not that all of its referred
objects are), as will mentioning the commit in a ref. But
the resulting repo is corrupt; we are missing the blob
reachable from a ref.
One way to solve this is to be more thorough when
referencing a sha1: make sure that not only do we have that
sha1, but that we have objects it refers to, and so forth
recursively. The problem is that this is very expensive.
Creating a parent link would require traversing the entire
object graph!
Instead, this patch pushes the extra work onto prune, which
runs less frequently (and has to look at the whole object
graph anyway). It creates a new category of objects: objects
which are not recent, but which are reachable from a recent
object. We do not prune these objects, just like the
reachable and recent ones.
This lets us avoid the recursive check above, because if we
have an object, even if it is unreachable, we should have
its referent. We can make a simple inductive argument that
with this patch, this property holds (that there are no
objects with missing referents in the repository):
0. When we have no objects, we have nothing to refer or be
referred to, so the property holds.
1. If we add objects to the repository, their direct
referents must generally exist (e.g., if you create a
tree, the blobs it references must exist; if you create
a commit to point at the tree, the tree must exist).
This is already the case before this patch. And it is
not 100% foolproof (you can make bogus objects using
`git hash-object`, for example), but it should be the
case for normal usage.
Therefore for any sequence of object additions, the
property will continue to hold.
2. If we remove objects from the repository, then we will
not remove a child object (like a blob) if an object
that refers to it is being kept. That is the part
implemented by this patch.
Note, however, that our reachability check and the
actual pruning are not atomic. So it _is_ still
possible to violate the property (e.g., an object
becomes referenced just as we are deleting it). This
patch is shooting for eliminating problems where the
mtimes of dependent objects differ by hours or days,
and one is dropped without the other. It does nothing
to help with short races.
Naively, the simplest way to implement this would be to add
all recent objects as tips to the reachability traversal.
However, this does not perform well. In a recently-packed
repository, all reachable objects will also be recent, and
therefore we have to look at each object twice. This patch
instead performs the reachability traversal, then follows up
with a second traversal for recent objects, skipping any
that have already been marked.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
