The commit-graph learned to use corrected commit dates instead of
the generation number to help topological revision traversal.
* ak/corrected-commit-date:
doc: add corrected commit date info
commit-reach: use corrected commit dates in paint_down_to_common()
commit-graph: use generation v2 only if entire chain does
commit-graph: implement generation data chunk
commit-graph: implement corrected commit date
commit-graph: return 64-bit generation number
commit-graph: add a slab to store topological levels
t6600-test-reach: generalize *_three_modes
commit-graph: consolidate fill_commit_graph_info
revision: parse parent in indegree_walk_step()
commit-graph: fix regression when computing Bloom filters
When `gc.writeCommitGraph = true`, it is possible that the commit-graph
is _still_ not written: replace objects, grafts and shallow repositories
are incompatible with the commit-graph feature.
Under such circumstances, we need to indicate to the user why the
commit-graph was not written instead of staying silent about it.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Acked-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Code clean-up to ensure our use of hashtables using object names as
keys use the "struct object_id" objects, not the raw hash values.
* jk/use-oid-pos:
oid_pos(): access table through const pointers
hash_pos(): convert to oid_pos()
rerere: use strmap to store rerere directories
rerere: tighten rr-cache dirname check
rerere: check dirname format while iterating rr_cache directory
commit_graft_pos(): take an oid instead of a bare hash
In preparation for creating an API around file formats using chunks and
tables of contents, prepare the commit-graph write code to use
prototypes that will match this new API.
Specifically, convert chunk_write_fn to take a "void *data" parameter
instead of the commit-graph-specific "struct write_commit_graph_context"
pointer.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Before checking if the repository has a commit-graph loaded, be sure
to run prepare_commit_graph(). This is necessary because otherwise
the topo_levels slab is not initialized. As we compute topo_levels for
the new commits, we iterate further into the lower layers since the
first visit to each commit looks as though the topo_level is not
populated.
By properly initializing the topo_slab, we fix the previously broken
case of a split commit graph where a base layer has the
generation_data_overflow chunk.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Reviewed-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When upgrading to a commit-graph with corrected commit dates from
one without, there are a few things that need to be considered.
When computing generation numbers for the new commit-graph file that
expects to add the generation_data chunk with corrected commit
dates, we need to ensure that the 'generation' member of the
commit_graph_data struct is set to zero for these commits.
Unfortunately, the fallback to use topological level for generation
number when corrected commit dates are not available are causing us
harm here: parsing commits notices that read_generation_data is
false and populates 'generation' with the topological level.
The solution is to iterate through the commits, parse the commits
to populate initial values, then reset the generation values to
zero to trigger recalculation. This loop only occurs when the
existing commit-graph data has no corrected commit dates.
While this improves our situation somewhat, we have not completely
solved the issue for correctly computing generation numbers for mixed
layers. That follows in the next change.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Reviewed-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The compute_generation_numbers() method was introduced by 3258c663
(commit-graph: compute generation numbers, 2018-05-01) to compute what
is now known as "topological levels". These are still stored in the
commit-graph file for compatibility sake while c1a09119 (commit-graph:
implement corrected commit date, 2021-01-16) updated the method to also
compute the new version of generation numbers: corrected commit date.
It makes sense why these are grouped. They perform very similar walks of
the necessary commits and compute similar maximums over each parent.
However, having these two together conflates them in subtle ways that is
hard to separate.
In particular, the topo_level slab is used to store the topological
levels in all cases, but the commit_graph_data_at(c)->generation member
stores different values depending on the state of the existing
commit-graph file.
* If the existing commit-graph file has a "GDAT" chunk, then these
values represent corrected commit dates.
* If the existing commit-graph file doesn't have a "GDAT" chunk, then
these values are actually the topological levels.
This issue only occurs only when upgrading an existing commit-graph file
into one that has the "GDAT" chunk. The current change does not resolve
this upgrade problem, but splitting the implementation into two pieces
here helps with that process, which will follow in the next change.
The important thing this helps with is the case where the
num_generation_data_overflows was being incremented incorrectly,
triggering a write of the overflow chunk.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Reviewed-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We need to be extra careful that we don't use corrected
commit dates from any layer of a commit-graph chain if there is a
single commit-graph file that is missing the generation_data chunk.
Update validate_mixed_generation_chain() to correctly update each
layer to ignore the generation_data chunk in this case. It now also
returns 1 if all layers have a generation_data chunk. This return
value will be used in the next change.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Reviewed-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There is a subtle failure happening when computing corrected commit
dates with --split enabled. It requires a base layer needing the
generation_data_overflow chunk. Then, the next layer on top
erroneously thinks it needs an overflow chunk due to a bug leading
to recalculating all reachable generation numbers. The output of
the failure is
BUG: commit-graph.c:1912: expected to write 8 bytes to
chunk 47444f56, but wrote 0 instead
These "expected" 8 bytes are due to re-computing the corrected
commit date for the lower layer but the new layer does not need
any overflow.
Add a test to t5318-commit-graph.sh that demonstrates this bug. However,
it does not trigger consistently with the existing code.
The generation number data is stored in a slab and accessed by
commit_graph_data_at(). This data is initialized when parsing a commit,
but is otherwise used assuming it has been populated. The loop in
compute_generation_numbers() did not enforce that all reachable
commits were parsed and had correct values. This could lead to some
problems when writing a commit-graph with corrected commit dates based
on a commit-graph without them.
It has been difficult to identify the issue here because it was so hard
to reproduce. It relies on this uninitialized data having a non-zero
value, but also on specifically in a way that overwrites the existing
data.
This patch adds the extra parse to ensure the data is filled before we
compute the generation number of a commit. This triggers the new test
to fail because the generation number overflow count does not match
between this computation and the write for that chunk.
The actual fix will follow as the next few changes.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Reviewed-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The write_commit_graph_context has a repository pointer, so use it.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Reviewed-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we are looking up an oid in an array, we obviously don't need to
write to the array. Let's mark it as const in the function interfaces,
as well as in the local variables we use to derference the void pointer
(note a few cases use pointers-to-pointers, so we mark everything
const).
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
All of our callers are actually looking up an object_id, not a bare
hash. Likewise, the arrays they are looking in are actual arrays of
object_id (not just raw bytes of hashes, as we might find in a pack
.idx; those are handled by bsearch_hash()).
Using an object_id gives us more type safety, and makes the callers
slightly shorter. It also gets rid of the word "sha1" from several
access functions, though we could obviously also rename those with
s/sha1/hash/.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The peel_ref() interface is confusing and error-prone:
- it's typically used by ref iteration callbacks that have both a
refname and oid. But since they pass only the refname, we may load
the ref value from the filesystem again. This is inefficient, but
also means we are open to a race if somebody simultaneously updates
the ref. E.g., this:
int some_ref_cb(const char *refname, const struct object_id *oid, ...)
{
if (!peel_ref(refname, &peeled))
printf("%s peels to %s",
oid_to_hex(oid), oid_to_hex(&peeled);
}
could print nonsense. It is correct to say "refname peels to..."
(you may see the "before" value or the "after" value, either of
which is consistent), but mentioning both oids may be mixing
before/after values.
Worse, whether this is possible depends on whether the optimization
to read from the current iterator value kicks in. So it is actually
not possible with:
for_each_ref(some_ref_cb);
but it _is_ possible with:
head_ref(some_ref_cb);
which does not use the iterator mechanism (though in practice, HEAD
should never peel to anything, so this may not be triggerable).
- it must take a fully-qualified refname for the read_ref_full() code
path to work. Yet we routinely pass it partial refnames from
callbacks to for_each_tag_ref(), etc. This happens to work when
iterating because there we do not call read_ref_full() at all, and
only use the passed refname to check if it is the same as the
iterator. But the requirements for the function parameters are quite
unclear.
Instead of taking a refname, let's instead take an oid. That fixes both
problems. It's a little funny for a "ref" function not to involve refs
at all. The key thing is that it's optimizing under the hood based on
having access to the ref iterator. So let's change the name to make it
clear why you'd want this function versus just peel_object().
There are two other directions I considered but rejected:
- we could pass the peel information into the each_ref_fn callback.
However, we don't know if the caller actually wants it or not. For
packed-refs, providing it is essentially free. But for loose refs,
we actually have to peel the object, which would be wasteful in most
cases. We could likewise pass in a flag to the callback indicating
whether the peeled information is known, but that complicates those
callbacks, as they then have to decide whether to manually peel
themselves. Plus it requires changing the interface of every
callback, whether they care about peeling or not, and there are many
of them.
- we could make a function to return the peeled value of the current
iterated ref (computing it if necessary), and BUG() otherwise. I.e.:
int peel_current_iterated_ref(struct object_id *out);
Each of the current callers is an each_ref_fn callback, so they'd
mostly be happy. But:
- we use those callbacks with functions like head_ref(), which do
not use the iteration code. So we'd need to handle the fallback
case there, anyway.
- it's possible that a caller would want to call into generic code
that sometimes is used during iteration and sometimes not. This
encapsulates the logic to do the fast thing when possible, and
fallback when necessary.
The implementation is mostly obvious, but I want to call out a few
things in the patch:
- the test-tool coverage for peel_ref() is now meaningless, as it all
collapses to a single peel_object() call (arguably they were pretty
uninteresting before; the tricky part of that function is the
fast-path we see during iteration, but these calls didn't trigger
that). I've just dropped it entirely, though note that some other
tests relied on the tags we created; I've moved that creation to the
tests where it matters.
- we no longer need to take a ref_store parameter, since we'd never
look up a ref now. We do still rely on a global "current iterator"
variable which _could_ be kept per-ref-store. But in practice this
is only useful if there are multiple recursive iterations, at which
point the more appropriate solution is probably a stack of
iterators. No caller used the actual ref-store parameter anyway
(they all call the wrapper that passes the_repository).
- the original only kicked in the optimization when the "refname"
pointer matched (i.e., not string comparison). We do likewise with
the "oid" parameter here, but fall back to doing an actual oideq()
call. This in theory lets us kick in the optimization more often,
though in practice no current caller cares. It should never be
wrong, though (peeling is a property of an object, so two refs
pointing to the same object would peel identically).
- the original took care not to touch the peeled out-parameter unless
we found something to put in it. But no caller cares about this, and
anyway, it is enforced by peel_object() itself (and even in the
optimized iterator case, that's where we eventually end up). We can
shorten the code and avoid an extra copy by just passing the
out-parameter through the stack.
Signed-off-by: Jeff King <peff@peff.net>
Reviewed-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
091f4cf (commit: don't use generation numbers if not needed,
2018-08-30) changed paint_down_to_common() to use commit dates instead
of generation numbers v1 (topological levels) as the performance
regressed on certain topologies. With generation number v2 (corrected
commit dates) implemented, we no longer have to rely on commit dates and
can use generation numbers.
For example, the command `git merge-base v4.8 v4.9` on the Linux
repository walks 167468 commits, taking 0.135s for committer date and
167496 commits, taking 0.157s for corrected committer date respectively.
While using corrected commit dates, Git walks nearly the same number of
commits as commit date, the process is slower as for each comparision we
have to access a commit-slab (for corrected committer date) instead of
accessing struct member (for committer date).
This change incidentally broke the fragile t6404-recursive-merge test.
t6404-recursive-merge sets up a unique repository where all commits have
the same committer date without a well-defined merge-base.
While running tests with GIT_TEST_COMMIT_GRAPH unset, we use committer
date as a heuristic in paint_down_to_common(). 6404.1 'combined merge
conflicts' merges commits in the order:
- Merge C with B to form an intermediate commit.
- Merge the intermediate commit with A.
With GIT_TEST_COMMIT_GRAPH=1, we write a commit-graph and subsequently
use the corrected committer date, which changes the order in which
commits are merged:
- Merge A with B to form an intermediate commit.
- Merge the intermediate commit with C.
While resulting repositories are equivalent, 6404.4 'virtual trees were
processed' fails with GIT_TEST_COMMIT_GRAPH=1 as we are selecting
different merge-bases and thus have different object ids for the
intermediate commits.
As this has already causes problems (as noted in 859fdc0 (commit-graph:
define GIT_TEST_COMMIT_GRAPH, 2018-08-29)), we disable commit graph
within t6404-recursive-merge.
Signed-off-by: Abhishek Kumar <abhishekkumar8222@gmail.com>
Reviewed-by: Taylor Blau <me@ttaylorr.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since there are released versions of Git that understand generation
numbers in the commit-graph's CDAT chunk but do not understand the GDAT
chunk, the following scenario is possible:
1. "New" Git writes a commit-graph with the GDAT chunk.
2. "Old" Git writes a split commit-graph on top without a GDAT chunk.
If each layer of split commit-graph is treated independently, as it was
the case before this commit, with Git inspecting only the current layer
for chunk_generation_data pointer, commits in the lower layer (one with
GDAT) whould have corrected commit date as their generation number,
while commits in the upper layer would have topological levels as their
generation. Corrected commit dates usually have much larger values than
topological levels. This means that if we take two commits, one from the
upper layer, and one reachable from it in the lower layer, then the
expectation that the generation of a parent is smaller than the
generation of a child would be violated.
It is difficult to expose this issue in a test. Since we _start_ with
artificially low generation numbers, any commit walk that prioritizes
generation numbers will walk all of the commits with high generation
number before walking the commits with low generation number. In all the
cases I tried, the commit-graph layers themselves "protect" any
incorrect behavior since none of the commits in the lower layer can
reach the commits in the upper layer.
This issue would manifest itself as a performance problem in this case,
especially with something like "git log --graph" since the low
generation numbers would cause the in-degree queue to walk all of the
commits in the lower layer before allowing the topo-order queue to write
anything to output (depending on the size of the upper layer).
Therefore, When writing the new layer in split commit-graph, we write a
GDAT chunk only if the topmost layer has a GDAT chunk. This guarantees
that if a layer has GDAT chunk, all lower layers must have a GDAT chunk
as well.
Rewriting layers follows similar approach: if the topmost layer below
the set of layers being rewritten (in the split commit-graph chain)
exists, and it does not contain GDAT chunk, then the result of rewrite
does not have GDAT chunks either.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Abhishek Kumar <abhishekkumar8222@gmail.com>
Reviewed-by: Taylor Blau <me@ttaylorr.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
As discovered by Ævar, we cannot increment graph version to
distinguish between generation numbers v1 and v2 [1]. Thus, one of
pre-requistes before implementing generation number v2 was to
distinguish between graph versions in a backwards compatible manner.
We are going to introduce a new chunk called Generation DATa chunk (or
GDAT). GDAT will store corrected committer date offsets whereas CDAT
will still store topological level.
Old Git does not understand GDAT chunk and would ignore it, reading
topological levels from CDAT. New Git can parse GDAT and take advantage
of newer generation numbers, falling back to topological levels when
GDAT chunk is missing (as it would happen with a commit-graph written
by old Git).
We introduce a test environment variable 'GIT_TEST_COMMIT_GRAPH_NO_GDAT'
which forces commit-graph file to be written without generation data
chunk to emulate a commit-graph file written by old Git.
To minimize the space required to store corrrected commit date, Git
stores corrected commit date offsets into the commit-graph file, instea
of corrected commit dates. This saves us 4 bytes per commit, decreasing
the GDAT chunk size by half, but it's possible for the offset to
overflow the 4-bytes allocated for storage. As such overflows are and
should be exceedingly rare, we use the following overflow management
scheme:
We introduce a new commit-graph chunk, Generation Data OVerflow ('GDOV')
to store corrected commit dates for commits with offsets greater than
GENERATION_NUMBER_V2_OFFSET_MAX.
If the offset is greater than GENERATION_NUMBER_V2_OFFSET_MAX, we set
the MSB of the offset and the other bits store the position of corrected
commit date in GDOV chunk, similar to how Extra Edge List is maintained.
We test the overflow-related code with the following repo history:
F - N - U
/ \
U - N - U N
\ /
N - F - N
Where the commits denoted by U have committer date of zero seconds
since Unix epoch, the commits denoted by N have committer date of
1112354055 (default committer date for the test suite) seconds since
Unix epoch and the commits denoted by F have committer date of
(2 ^ 31 - 2) seconds since Unix epoch.
The largest offset observed is 2 ^ 31, just large enough to overflow.
[1]: https://lore.kernel.org/git/87a7gdspo4.fsf@evledraar.gmail.com/
Signed-off-by: Abhishek Kumar <abhishekkumar8222@gmail.com>
Reviewed-by: Taylor Blau <me@ttaylorr.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
With most of preparations done, let's implement corrected commit date.
The corrected commit date for a commit is defined as:
* A commit with no parents (a root commit) has corrected commit date
equal to its committer date.
* A commit with at least one parent has corrected commit date equal to
the maximum of its commit date and one more than the largest corrected
commit date among its parents.
As a special case, a root commit with timestamp of zero (01.01.1970
00:00:00Z) has corrected commit date of one, to be able to distinguish
from GENERATION_NUMBER_ZERO (that is, an uncomputed corrected commit
date).
To minimize the space required to store corrected commit date, Git
stores corrected commit date offsets into the commit-graph file. The
corrected commit date offset for a commit is defined as the difference
between its corrected commit date and actual commit date.
Storing corrected commit date requires sizeof(timestamp_t) bytes, which
in most cases is 64 bits (uintmax_t). However, corrected commit date
offsets can be safely stored using only 32-bits. This halves the size
of GDAT chunk, which is a reduction of around 6% in the size of
commit-graph file.
However, using offsets be problematic if a commit is malformed but valid
and has committer date of 0 Unix time, as the offset would be the same
as corrected commit date and thus require 64-bits to be stored properly.
While Git does not write out offsets at this stage, Git stores the
corrected commit dates in member generation of struct commit_graph_data.
It will begin writing commit date offsets with the introduction of
generation data chunk.
Signed-off-by: Abhishek Kumar <abhishekkumar8222@gmail.com>
Reviewed-by: Taylor Blau <me@ttaylorr.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In a preparatory step for introducing corrected commit dates, let's
return timestamp_t values from commit_graph_generation(), use
timestamp_t for local variables and define GENERATION_NUMBER_INFINITY
as (2 ^ 63 - 1) instead.
We rename GENERATION_NUMBER_MAX to GENERATION_NUMBER_V1_MAX to
represent the largest topological level we can store in the commit data
chunk.
With corrected commit dates implemented, we will have two such *_MAX
variables to denote the largest offset and largest topological level
that can be stored.
Signed-off-by: Abhishek Kumar <abhishekkumar8222@gmail.com>
Reviewed-by: Taylor Blau <me@ttaylorr.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In a later commit we will introduce corrected commit date as the
generation number v2. Corrected commit dates will be stored in the new
seperate Generation Data chunk. However, to ensure backwards
compatibility with "Old" Git we need to continue to write generation
number v1 (topological levels) to the commit data chunk. Thus, we need
to compute and store both versions of generation numbers to write the
commit-graph file.
Therefore, let's introduce a commit-slab `topo_level_slab` to store
topological levels; corrected commit date will be stored in the member
`generation` of struct commit_graph_data.
The macros `GENERATION_NUMBER_INFINITY` and `GENERATION_NUMBER_ZERO`
mark commits not in the commit-graph file and commits written by a
version of Git that did not compute generation numbers respectively.
Generation numbers are computed identically for both kinds of commits.
A "slab-miss" should return `GENERATION_NUMBER_INFINITY` as the commit
is not in the commit-graph file. However, since the slab is
zero-initialized, it returns 0 (or rather `GENERATION_NUMBER_ZERO`).
Thus, we no longer need to check if the topological level of a commit is
`GENERATION_NUMBER_INFINITY`.
We will add a pointer to the slab in `struct write_commit_graph_context`
and `struct commit_graph` to populate the slab in
`fill_commit_graph_info` if the commit has a pre-computed topological
level as in case of split commit-graphs.
Signed-off-by: Abhishek Kumar <abhishekkumar8222@gmail.com>
Reviewed-by: Taylor Blau <me@ttaylorr.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Both fill_commit_graph_info() and fill_commit_in_graph() parse
information present in commit data chunk. Let's simplify the
implementation by calling fill_commit_graph_info() within
fill_commit_in_graph().
fill_commit_graph_info() used to not load committer data from commit data
chunk. However, with the upcoming switch to using corrected committer
date as generation number v2, we will have to load committer date to
compute generation number value anyway.
e51217e15 (t5000: test tar files that overflow ustar headers,
30-06-2016) introduced a test 'generate tar with future mtime' that
creates a commit with committer date of (2^36 + 1) seconds since
EPOCH. The CDAT chunk provides 34-bits for storing committer date, thus
committer time overflows into generation number (within CDAT chunk) and
has undefined behavior.
The test used to pass as fill_commit_graph_info() would not set struct
member `date` of struct commit and load committer date from the object
database, generating a tar file with the expected mtime.
However, with corrected commit date, we will load the committer date
from CDAT chunk (truncated to lower 34-bits to populate the generation
number. Thus, Git sets date and generates tar file with the truncated
mtime.
The ustar format (the header format used by most modern tar programs)
only has room for 11 (or 12, depending on some implementations) octal
digits for the size and mtime of each file.
As the CDAT chunk is overflow by 12-octal digits but not 11-octal
digits, we split the existing tests to test both implementations
separately and add a new explicit test for 11-digit implementation.
To test the 11-octal digit implementation, we create a future commit
with committer date of 2^34 - 1, which overflows 11-octal digits without
overflowing 34-bits of the Commit Date chunks.
To test the 12-octal digit implementation, the smallest committer date
possible is 2^36 + 1, which overflows the CDAT chunk and thus
commit-graph must be disabled for the test.
Signed-off-by: Abhishek Kumar <abhishekkumar8222@gmail.com>
Reviewed-by: Taylor Blau <me@ttaylorr.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Before computing Bloom filters, the commit-graph machinery uses
commit_gen_cmp to sort commits by generation order for improved diff
performance. 3d11275505 (commit-graph: examine commits by generation
number, 2020-03-30) claims that this sort can reduce the time spent to
compute Bloom filters by nearly half.
But since c49c82aa4c (commit: move members graph_pos, generation to a
slab, 2020-06-17), this optimization is broken, since asking for a
'commit_graph_generation()' directly returns GENERATION_NUMBER_INFINITY
while writing.
Not all hope is lost, though: 'commit_gen_cmp()' falls back to
comparing commits by their date when they have equal generation number,
and so since c49c82aa4c is purely a date comparison function. This
heuristic is good enough that we don't seem to loose appreciable
performance while computing Bloom filters.
Applying this patch (compared with v2.30.0) speeds up computing Bloom
filters by factors ranging from 0.40% to 5.19% on various repositories [1].
So, avoid the useless 'commit_graph_generation()' while writing by
instead accessing the slab directly. This returns the newly-computed
generation numbers, and allows us to avoid the heuristic by directly
comparing generation numbers.
[1]: https://lore.kernel.org/git/20210105094535.GN8396@szeder.dev/
Signed-off-by: Abhishek Kumar <abhishekkumar8222@gmail.com>
Reviewed-by: Taylor Blau <me@ttaylorr.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Similar to the previous commit, avoid peeking into the `struct
lock_file`. Use the lock file API instead.
Signed-off-by: Martin Ågren <martin.agren@gmail.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Change all remnants of "sha1" in hash-lookup.c and .h and rename them to
reflect that we're not just able to handle SHA-1 these days.
Signed-off-by: Martin Ågren <martin.agren@gmail.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Rename this function to reflect that we're not just able to handle SHA-1
these days. There are a few instances of "sha1" left in sha1-lookup.[ch]
after this, but those will be addressed in the next commit.
Signed-off-by: Martin Ågren <martin.agren@gmail.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Our packed_commit_list is an array of pointers to commit structs. We use
"int" for the allocation, which is 32-bit even on 64-bit platforms. This
isn't likely to overflow in practice (we're writing commit graphs, so
you'd need to actually have billions of unique commits in the
repository). But it's good practice to use size_t for allocations.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Our custom packed_oid_list data structure is really just an oid_array in
disguise. Let's switch to using the generic structure, which shortens
and simplifies the code slightly.
There's one slightly awkward part: in the old code we copied a hash
straight from the mmap'd on-disk data into the final object_id. And now
we'll copy to a temporary oid, which we'll then pass to
oid_array_append(). But this is an operation we have to do all over the
commit-graph code already, since it mostly uses object_id structs
internally. I also measured "git commit-graph --append", which triggers
this code path, and it showed no difference.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When writing a commit graph, we collect a list of object ids in an
array, which we'll eventually copy into an array of "struct commit"
pointers. Before we do that, though, we count the number of distinct
commit entries. There's a subtle bug in this step, though.
We eliminate not only duplicate oids, but also in split mode, any oids
which are not commits or which are already in a graph file. However, the
loop starts at index 1, always counting index 0 as distinct. And indeed
it can't be a duplicate, since we check for those by comparing against
the previous entry, and there isn't one for index 0. But it could be a
commit that's already in a graph file, and we'd overcount the number of
commits by 1 in that case.
That turns out not to be a problem, though. The only things we do with
the count are:
- check if our count will overflow our data structures. But the limit
there is 2^31 commits, so while this is a useful check, the
off-by-one is not likely to matter.
- pre-allocate the array of commit pointers. But over-allocating by
one isn't a problem; we'll just waste a few extra bytes.
The bug would be easy enough to fix, but we can observe that neither of
those steps is necessary.
After building the actual commit array, we'll likewise check its count
for overflow. So the extra check of the distinct commit count here is
redundant.
And likewise we use ALLOC_GROW() when building the commit array, so
there's no need to preallocate it (it's possible that doing so is
slightly more efficient, but if we care we can just optimistically
allocate one slot for each oid; I didn't bother here).
So count_distinct_commits() isn't doing anything useful. Let's just get
rid of that step.
Note that a side effect of the function was that we sorted the list of
oids, which we do rely on in copy_oids_to_commits(), since it must also
skip the duplicates. So we'll move the qsort there. I didn't copy the
"TODO" about adding more progress meters. It's actually quite hard to
make a repository large enough for this qsort would take an appreciable
amount of time, so this doesn't seem like a useful note.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When "git commit-graph" detects the same commit recorded more than
once while it is merging the layers, it used to die. The code now
ignores all but one of them and continues.
* ds/commit-graph-merging-fix:
commit-graph: don't write commit-graph when disabled
commit-graph: ignore duplicates when merging layers
The core.commitGraph config setting can be set to 'false' to prevent
parsing commits from the commit-graph file(s). This causes an issue when
trying to write with "--split" which needs to distinguish between
commits that are in the existing commit-graph layers and commits that
are not. The existing mechanism uses parse_commit() and follows by
checking if there is a 'graph_pos' that shows the commit was parsed from
the commit-graph file.
When core.commitGraph=false, we do not parse the commits from the
commit-graph and 'graph_pos' indicates that no commits are in the
existing file. The --split logic moves forward creating a new layer on
top that holds all reachable commits, then possibly merges down into
those layers, resulting in duplicate commits. The previous change makes
that merging process more robust to such a situation in case it happens
in the written commit-graph data.
The easy answer here is to avoid writing a commit-graph if reading the
commit-graph is disabled. Since the resulting commit-graph will would not
be read by subsequent Git processes. This is more natural than forcing
core.commitGraph to be true for the 'write' process.
Reported-by: Thomas Braun <thomas.braun@virtuell-zuhause.de>
Helped-by: Jeff King <peff@peff.net>
Helped-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Thomas reported [1] that a "git fetch" command was failing with an error
saying "unexpected duplicate commit id". The root cause is that they had
fetch.writeCommitGraph enabled which generates commit-graph chains, and
this instance was merging two layers that both contained the same commit
ID.
[1] https://lore.kernel.org/git/55f8f00c-a61c-67d4-889e-a9501c596c39@virtuell-zuhause.de/
The initial assumption is that Git would not write a commit ID into a
commit-graph layer if it already exists in a lower commit-graph layer.
Somehow, this specific case did get into that situation, leading to this
error.
While unexpected, this isn't actually invalid (as long as the two layers
agree on the metadata for the commit). When we parse a commit that does
not have a graph_pos in the commit_graph_data_slab, we use binary search
in the commit-graph layers to find the commit and set graph_pos. That
position is never used again in this case. However, when we parse a
commit from the commit-graph file, we load its parents from the
commit-graph and assign graph_pos at that point. If those parents were
already parsed from the commit-graph, then nothing needs to be done.
Otherwise, this graph_pos is a valid position in the commit-graph so we
can parse the parents, when necessary.
Thus, this die() is too aggressive. The easiest thing to do would be to
ignore the duplicates.
If we only ignore the duplicates, then we will produce a commit-graph
that has identical commit IDs listed in adjacent positions. This excess
data will never be removed from the commit-graph, which could cascade
into significantly bloated file sizes.
Thankfully, we can collapse the list to erase the duplicate commit
pointers. This allows us to get the end result we want without extra
memory costs and minimal CPU time.
The root cause is due to disabling core.commitGraph, which prevents
parsing commits from the lower layers during a 'git commit-graph write
--split' command. Since we use the 'graph_pos' value to determine
whether a commit is in a lower layer, we never discover that those
commits are already in the commit-graph chain and add them to the top
layer. This layer is then merged down, creating duplicates.
The test added in t5324-split-commit-graph.sh fails without this change.
However, we still have not completely removed the need for this
duplicate check. That will come in a follow-up change.
Reported-by: Thomas Braun <thomas.braun@virtuell-zuhause.de>
Helped-by: Taylor Blau <me@ttaylorr.com>
Co-authored-by: Jeff King <peff@peff.net>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git commit-graph write" learned to limit the number of bloom
filters that are computed from scratch with the --max-new-filters
option.
* tb/bloom-improvements:
commit-graph: introduce 'commitGraph.maxNewFilters'
builtin/commit-graph.c: introduce '--max-new-filters=<n>'
commit-graph: rename 'split_commit_graph_opts'
bloom: encode out-of-bounds filters as non-empty
bloom/diff: properly short-circuit on max_changes
bloom: use provided 'struct bloom_filter_settings'
bloom: split 'get_bloom_filter()' in two
commit-graph.c: store maximum changed paths
commit-graph: respect 'commitGraph.readChangedPaths'
t/helper/test-read-graph.c: prepare repo settings
commit-graph: pass a 'struct repository *' in more places
t4216: use an '&&'-chain
commit-graph: introduce 'get_bloom_filter_settings()'
A "git gc"'s big brother has been introduced to take care of more
repository maintenance tasks, not limited to the object database
cleaning.
* ds/maintenance-part-1:
maintenance: add trace2 regions for task execution
maintenance: add auto condition for commit-graph task
maintenance: use pointers to check --auto
maintenance: create maintenance.<task>.enabled config
maintenance: take a lock on the objects directory
maintenance: add --task option
maintenance: add commit-graph task
maintenance: initialize task array
maintenance: replace run_auto_gc()
maintenance: add --quiet option
maintenance: create basic maintenance runner
Introduce a command-line flag to specify the maximum number of new Bloom
filters that a 'git commit-graph write' is willing to compute from
scratch.
Prior to this patch, a commit-graph write with '--changed-paths' would
compute Bloom filters for all selected commits which haven't already
been computed (i.e., by a previous commit-graph write with '--split'
such that a roll-up or replacement is performed).
This behavior can cause prohibitively-long commit-graph writes for a
variety of reasons:
* There may be lots of filters whose diffs take a long time to
generate (for example, they have close to the maximum number of
changes, diffing itself takes a long time, etc).
* Old-style commit-graphs (which encode filters with too many entries
as not having been computed at all) cause us to waste time
recomputing filters that appear to have not been computed only to
discover that they are too-large.
This can make the upper-bound of the time it takes for 'git commit-graph
write --changed-paths' to be rather unpredictable.
To make this command behave more predictably, introduce
'--max-new-filters=<n>' to allow computing at most '<n>' Bloom filters
from scratch. This lets "computing" already-known filters proceed
quickly, while bounding the number of slow tasks that Git is willing to
do.
Helped-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In the subsequent commit, additional options will be added to the
commit-graph API which have nothing to do with splitting.
Rename the 'split_commit_graph_opts' structure to the more-generic
'commit_graph_opts' to encompass both. Likewise, rename the 'flags'
member to instead be 'split_flags' to clarify that it only has to do
with the behavior implied by '--split'.
Suggested-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When a changed-path Bloom filter has either zero, or more than a
certain number (commonly 512) of entries, the commit-graph machinery
encodes it as "missing". More specifically, it sets the indices adjacent
in the BIDX chunk as equal to each other to indicate a "length 0"
filter; that is, that the filter occupies zero bytes on disk.
This has heretofore been fine, since the commit-graph machinery has no
need to care about these filters with too few or too many changed paths.
Both cases act like no filter has been generated at all, and so there is
no need to store them.
In a subsequent commit, however, the commit-graph machinery will learn
to only compute Bloom filters for some commits in the current
commit-graph layer. This is a change from the current implementation
which computes Bloom filters for all commits that are in the layer being
written. Critically for this patch, only computing some of the Bloom
filters means adding a third state for length 0 Bloom filters: zero
entries, too many entries, or "hasn't been computed".
It will be important for that future patch to distinguish between "not
representable" (i.e., zero or too-many changed paths), and "hasn't been
computed". In particular, we don't want to waste time recomputing
filters that have already been computed.
To that end, change how we store Bloom filters in the "computed but not
representable" category:
- Bloom filters with no entries are stored as a single byte with all
bits low (i.e., all queries to that Bloom filter will return
"definitely not")
- Bloom filters with too many entries are stored as a single byte with
all bits set high (i.e., all queries to that Bloom filter will
return "maybe").
These rules are sufficient to not incur a behavior change by changing
the on-disk representation of these two classes. Likewise, no
specification changes are necessary for the commit-graph format, either:
- Filters that were previously empty will be recomputed and stored
according to the new rules, and
- old clients reading filters generated by new clients will interpret
the filters correctly and be none the wiser to how they were
generated.
Clients will invoke the Bloom machinery in more cases than before, but
this can be addressed by returning a NULL filter when all bits are set
high. This can be addressed in a future patch.
Note that this does increase the size of on-disk commit-graphs, but far
less than other proposals. In particular, this is generally more
efficient than storing a bitmap for which commits haven't computed their
Bloom filters. Storing a bitmap incurs a penalty of one bit per commit,
whereas storing explicit filters as above incurs a penalty of one byte
per too-large or empty commit.
In practice, these boundary commits likely occupy a small proportion of
the overall number of commits, and so the size penalty is likely smaller
than storing a bitmap for all commits.
See, for example, these relative proportions of such boundary commits
(collected by SZEDER Gábor):
| Percentage of | commit-graph | |
| commits modifying | file size | |
├────────┬──────────────┼───────────────────┤ pct. |
| 0 path | >= 512 paths | before | after | change |
┌────────────────┼────────┼──────────────┼─────────┼─────────┼───────────┤
| android-base | 13.20% | 0.13% | 37.468M | 37.534M | +0.1741 % |
| cmssw | 0.15% | 0.23% | 17.118M | 17.119M | +0.0091 % |
| cpython | 3.07% | 0.01% | 7.967M | 7.971M | +0.0423 % |
| elasticsearch | 0.70% | 1.00% | 8.833M | 8.835M | +0.0128 % |
| gcc | 0.00% | 0.08% | 16.073M | 16.074M | +0.0030 % |
| gecko-dev | 0.14% | 0.64% | 59.868M | 59.874M | +0.0105 % |
| git | 0.11% | 0.02% | 3.895M | 3.895M | +0.0020 % |
| glibc | 0.02% | 0.10% | 3.555M | 3.555M | +0.0021 % |
| go | 0.00% | 0.07% | 3.186M | 3.186M | +0.0018 % |
| homebrew-cask | 0.40% | 0.02% | 7.035M | 7.035M | +0.0065 % |
| homebrew-core | 0.01% | 0.01% | 11.611M | 11.611M | +0.0002 % |
| jdk | 0.26% | 5.64% | 5.537M | 5.540M | +0.0590 % |
| linux | 0.01% | 0.51% | 63.735M | 63.740M | +0.0073 % |
| llvm-project | 0.12% | 0.03% | 25.515M | 25.516M | +0.0050 % |
| rails | 0.10% | 0.10% | 6.252M | 6.252M | +0.0027 % |
| rust | 0.07% | 0.17% | 9.364M | 9.364M | +0.0033 % |
| tensorflow | 0.09% | 1.02% | 7.009M | 7.010M | +0.0158 % |
| webkit | 0.05% | 0.31% | 17.405M | 17.406M | +0.0047 % |
(where the above increase is determined by computing a non-split
commit-graph before and after this patch).
Given that these projects are all "large" by commit count, the storage
cost by writing these filters explicitly is negligible. In the most
extreme example, android-base (which has 494,848 commits at the time of
writing) would have its commit-graph increase by a modest 68.4 KB.
Finally, a test to exercise filters which contain too many changed path
entries will be introduced in a subsequent patch.
Suggested-by: SZEDER Gábor <szeder.dev@gmail.com>
Suggested-by: Jakub Narębski <jnareb@gmail.com>
Helped-by: Derrick Stolee <dstolee@microsoft.com>
Helped-by: SZEDER Gábor <szeder.dev@gmail.com>
Helped-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The first new task in the 'git maintenance' builtin is the
'commit-graph' task. This updates the commit-graph file
incrementally with the command
git commit-graph write --reachable --split
By writing an incremental commit-graph file using the "--split"
option we minimize the disruption from this operation. The default
behavior is to merge layers until the new "top" layer is less than
half the size of the layer below. This provides quick writes most
of the time, with the longer writes following a power law
distribution.
Most importantly, concurrent Git processes only look at the
commit-graph-chain file for a very short amount of time, so they
will verly likely not be holding a handle to the file when we try
to replace it. (This only matters on Windows.)
If a concurrent process reads the old commit-graph-chain file, but
our job expires some of the .graph files before they can be read,
then those processes will see a warning message (but not fail).
This could be avoided by a future update to use the --expire-time
argument when writing the commit-graph.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When 'get_or_compute_bloom_filter()' needs to compute a Bloom filter
from scratch, it looks to the default 'struct bloom_filter_settings' in
order to determine the maximum number of changed paths, number of bits
per entry, and so on.
All of these values have so far been constant, and so there was no need
to pass in a pointer from the caller (eg., the one that is stored in the
'struct write_commit_graph_context').
Start passing in a 'struct bloom_filter_settings *' instead of using the
default values to respect graph-specific settings (eg., in the case of
setting 'GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS').
In order to have an initialized value for these settings, move its
initialization to earlier in the commit-graph write.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
'get_bloom_filter' takes a flag to control whether it will compute a
Bloom filter if the requested one is missing. In the next patch, we'll
add yet another parameter to this method, which would force all but one
caller to specify an extra 'NULL' parameter at the end.
Instead of doing this, split 'get_bloom_filter' into two functions:
'get_bloom_filter' and 'get_or_compute_bloom_filter'. The former only
looks up a Bloom filter (and does not compute one if it's missing,
thus dropping the 'compute_if_not_present' flag). The latter does
compute missing Bloom filters, with an additional parameter to store
whether or not it needed to do so.
This simplifies many call-sites, since the majority of existing callers
to 'get_bloom_filter' do not want missing Bloom filters to be computed
(so they can drop the parameter entirely and use the simpler version of
the function).
While we're at it, instrument the new 'get_or_compute_bloom_filter()'
with counters in the 'write_commit_graph_context' struct which store
the number of filters that we did and didn't compute, as well as filters
that were truncated.
It would be nice to drop the 'compute_if_not_present' flag entirely,
since all remaining callers of 'get_or_compute_bloom_filter' pass it as
'1', but this will change in a future patch and hence cannot be removed.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
For now, we assume that there is a fixed constant describing the
maximum number of changed paths we are willing to store in a Bloom
filter.
Prepare for that to (at least partially) not be the case by making it a
member of the 'struct bloom_filter_settings'. This will be helpful in
the subsequent patches by reducing the size of test cases that exercise
storing too many changed paths, as well as preparing for an eventual
future in which this value might change.
This patch alone does not cause newly generated Bloom filters to use
a custom upper-bound on the maximum number of changed paths a single
Bloom filter can hold, that will occur in a later patch.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Git uses the 'core.commitGraph' configuration value to control whether
or not the commit graph is used when parsing commits or performing a
traversal.
Now that commit-graphs can also contain a section for changed-path Bloom
filters, administrators that already have commit-graphs may find it
convenient to use those graphs without relying on their changed-path
Bloom filters. This can happen, for example, during a staged roll-out,
or in the event of an incident.
Introduce 'commitGraph.readChangedPaths' to control whether or not Bloom
filters are read. Note that this configuration is independent from both:
- 'core.commitGraph', to allow flexibility in using all parts of a
commit-graph _except_ for its Bloom filters.
- The '--changed-paths' option for 'git commit-graph write', to allow
reading and writing Bloom filters to be controlled independently.
When the variable is set, pretend as if no Bloom data was specified at
all. This avoids adding additional special-casing outside of the
commit-graph internals.
Suggested-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In a future commit, some commit-graph internals will want access to
'r->settings', but we only have the 'struct object_directory *'
corresponding to that repository.
Add an additional parameter to pass the repository around in more
places.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Many places in the code often need a pointer to the commit-graph's
'struct bloom_filter_settings', in which case they often take the value
from the top-most commit-graph.
In the non-split case, this works as expected. In the split case,
however, things get a little tricky. Not all layers in a chain of
incremental commit-graphs are required to themselves have Bloom data,
and so whether or not some part of the code uses Bloom filters depends
entirely on whether or not the top-most level of the commit-graph chain
has Bloom filters.
This has been the behavior since Bloom filters were introduced, and has
been codified into the tests since a759bfa9ee (t4216: add end to end
tests for git log with Bloom filters, 2020-04-06). In fact, t4216.130
requires that Bloom filters are not used in exactly the case described
earlier.
There is no reason that this needs to be the case, since it is perfectly
valid for commits in an earlier layer to have Bloom filters when commits
in a newer layer do not.
Since Bloom settings are guaranteed in practice to be the same for any
layer in a chain that has Bloom data, it is sufficient to traverse the
'->base_graph' pointer until either (1) a non-null 'struct
bloom_filter_settings *' is found, or (2) until we are at the root of
the commit-graph chain.
Introduce a 'get_bloom_filter_settings()' function that does just this,
and use it instead of purely dereferencing the top-most graph's
'->bloom_filter_settings' pointer.
While we're at it, add an additional test in t5324 to guard against code
in the commit-graph writing machinery that doesn't correctly handle a
NULL 'struct bloom_filter *'.
Co-authored-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The commit-graph format reserved a byte among the header of the file to
store a "hash version". During the SHA-256 work, this was not modified
because file formats are not necessarily intended to work across hash
versions. If a repository has SHA-256 as its hash algorithm, it
automatically up-shifts the lengths of object names in all necessary
formats.
However, since we have this byte available for adjusting the version, we
can make the file formats more obviously incompatible instead of relying
on other context from the repository.
Update the oid_version() method in commit-graph.c to add a new value, 2,
for sha-256. This automatically writes the new value in a SHA-256
repository _and_ verifies the value is correct. This is a breaking
change relative to the current 'master' branch since 092b677 (Merge
branch 'bc/sha-256-cvs-svn-updates', 2020-08-13) but it is not breaking
relative to any released version of Git.
The test impact is relatively minor: the output of 'test-tool
read-graph' lists the header information, so those instances of '1' need
to be replaced with a variable determined by GIT_TEST_DEFAULT_HASH. A
more careful test is added that specifically creates a repository of
each type then swaps the commit-graph files. The important value here is
that the "git log" command succeeds while writing a message to stderr.
Helped-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Reviewed-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Updates to the changed-paths bloom filter.
* ds/commit-graph-bloom-updates:
commit-graph: check all leading directories in changed path Bloom filters
revision: empty pathspecs should not use Bloom filters
revision.c: fix whitespace
commit-graph: check chunk sizes after writing
commit-graph: simplify chunk writes into loop
commit-graph: unify the signatures of all write_graph_chunk_*() functions
commit-graph: persist existence of changed-paths
bloom: fix logic in get_bloom_filter()
commit-graph: change test to die on parse, not load
commit-graph: place bloom_settings in context
The changed-path Bloom filter is improved using ideas from an
independent implementation.
* sg/commit-graph-cleanups:
commit-graph: simplify write_commit_graph_file() #2
commit-graph: simplify write_commit_graph_file() #1
commit-graph: simplify parse_commit_graph() #2
commit-graph: simplify parse_commit_graph() #1
commit-graph: clean up #includes
diff.h: drop diff_tree_oid() & friends' return value
commit-slab: add a function to deep free entries on the slab
commit-graph-format.txt: all multi-byte numbers are in network byte order
commit-graph: fix parsing the Chunk Lookup table
tree-walk.c: don't match submodule entries for 'submod/anything'
The code to produce progress output from "git commit-graph --write"
had a few breakages, which have been fixed.
* sg/commit-graph-progress-fix:
commit-graph: fix "Writing out commit graph" progress counter
commit-graph: fix progress of reachable commits
When "fetch.writeCommitGraph" configuration is set in a shallow
repository and a fetch moves the shallow boundary, we wrote out
broken commit-graph files that do not match the reality, which has
been corrected.
* tb/fix-persistent-shallow:
commit.c: don't persist substituted parents when unshallowing
76ffbca71a (commit-graph: write Bloom filters to commit graph file,
2020-04-06) added two delayed progress lines to writing the Bloom
filter index and data chunk. This is wrong, because a single common
progress is used while writing all chunks, which is not updated while
writing these two new chunks, resulting in incomplete-looking "done"
lines:
Expanding reachable commits in commit graph: 888679, done.
Computing commit changed paths Bloom filters: 100% (888678/888678), done.
Writing out commit graph in 6 passes: 66% (3554712/5332068), done.
Use the common 'struct progress' instance while writing the Bloom
filter chunks as well.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
To display a progress line while iterating over all refs,
d335ce8f24 (commit-graph.c: show progress of finding reachable
commits, 2020-05-13) should have added a pair of
start_delayed_progress() and stop_progress() calls around a
for_each_ref() invocation. Alas, the stop_progress() call ended up at
the wrong place, after write_commit_graph(), which does all the
commit-graph computation and writing, and has several progress lines
of its own. Consequently, that new
Collecting referenced commits: 123
progress line is overwritten by the first progress line shown by
write_commit_graph(), and its final "done" line is shown last, after
everything is finished:
Expanding reachable commits in commit graph: 344786, done.
Computing commit changed paths Bloom filters: 100% (344786/344786), done.
Collecting referenced commits: 154, done.
Move that stop_progress() call to the right place.
While at it, drop the unnecessary 'if (data.progress)' condition
protecting the stop_progress() call, because that function is prepared
to handle a NULL progress struct.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since 37b9dcabfc (shallow.c: use '{commit,rollback}_shallow_file',
2020-04-22), Git knows how to reset stat-validity checks for the
$GIT_DIR/shallow file, allowing it to change between a shallow and
non-shallow state in the same process (e.g., in the case of 'git fetch
--unshallow').
However, when $GIT_DIR/shallow changes, Git does not alter or remove any
grafts (nor substituted parents) in memory.
This comes up in a "git fetch --unshallow" with fetch.writeCommitGraph
set to true. Ordinarily in a shallow repository (and before 37b9dcabfc,
even in this case), commit_graph_compatible() would return false,
indicating that the repository should not be used to write a
commit-graphs (since commit-graph files cannot represent a shallow
history). But since 37b9dcabfc, in an --unshallow operation that check
succeeds.
Thus even though the repository isn't shallow any longer (that is, we
have all of the objects), the in-core representation of those objects
still has munged parents at the shallow boundaries. When the
commit-graph write proceeds, we use the incorrect parentage, producing
wrong results.
There are two ways for a user to work around this: either (1) set
'fetch.writeCommitGraph' to 'false', or (2) drop the commit-graph after
unshallowing.
One way to fix this would be to reset the parsed object pool entirely
(flushing the cache and thus preventing subsequent reads from modifying
their parents) after unshallowing. That would produce a problem when
callers have a now-stale reference to the old pool, and so this patch
implements a different approach. Instead, attach a new bit to the pool,
'substituted_parent', which indicates if the repository *ever* stored a
commit which had its parents modified (i.e., the shallow boundary
prior to unshallowing).
This bit needs to be sticky because all reads subsequent to modifying a
commit's parents are unreliable when unshallowing. Modify the check in
'commit_graph_compatible' to take this bit into account, and correctly
avoid generating commit-graphs in this case, thus solving the bug.
Helped-by: Derrick Stolee <dstolee@microsoft.com>
Helped-by: Jonathan Nieder <jrnieder@gmail.com>
Reported-by: Jay Conrod <jayconrod@google.com>
Reviewed-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In my experience while experimenting with new commit-graph chunks,
early versions of the corresponding new write_commit_graph_my_chunk()
functions are, sadly but not surprisingly, often buggy, and write more
or less data than they are supposed to, especially if the chunk size
is not directly proportional to the number of commits. This then
causes all kinds of issues when reading such a bogus commit-graph
file, raising the question of whether the writing or the reading part
happens to be buggy this time.
Let's catch such issues early, already when writing the commit-graph
file, and check that each write_graph_chunk_*() function wrote the
amount of data that it was expected to, and what has been encoded in
the Chunk Lookup table. Now that all commit-graph chunks are written
in a loop we can do this check in a single place for all chunks, and
any chunks added in the future will get checked as well.
Helped-by: René Scharfe <l.s.r@web.de>
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In write_commit_graph_file() we now have one block of code filling the
array of 'struct chunk_info' with the IDs and sizes of chunks to be
written, and an other block of code calling the functions responsible
for writing individual chunks. In case of optional chunks like Extra
Edge List an Base Graphs List there is also a condition checking
whether that chunk is necessary/desired, and that same condition is
repeated in both blocks of code. Other, newer chunks have similar
optional conditions.
Eliminate these repeated conditions by storing the function pointers
responsible for writing individual chunks in the 'struct chunk_info'
array as well, and calling them in a loop to write the commit-graph
file. This will open up the possibility for a bit of foolproofing in
the following patch.
Helped-by: René Scharfe <l.s.r@web.de>
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Update the write_graph_chunk_*() helper functions to have the same
signature:
- Return an int error code from all these functions.
write_graph_chunk_base() already has an int error code, now the
others will have one, too, but since they don't indicate any
error, they will always return 0.
- Drop the hash size parameter of write_graph_chunk_oids() and
write_graph_chunk_data(); its value can be read directly from
'the_hash_algo' inside these functions as well.
This opens up the possibility for further cleanups and foolproofing in
the following two patches.
Helped-by: René Scharfe <l.s.r@web.de>
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The changed-path Bloom filters were released in v2.27.0, but have a
significant drawback. A user can opt-in to writing the changed-path
filters using the "--changed-paths" option to "git commit-graph write"
but the next write will drop the filters unless that option is
specified.
This becomes even more important when considering the interaction with
gc.writeCommitGraph (on by default) or fetch.writeCommitGraph (part of
features.experimental). These config options trigger commit-graph writes
that the user did not signal, and hence there is no --changed-paths
option available.
Allow a user that opts-in to the changed-path filters to persist the
property of "my commit-graph has changed-path filters" automatically. A
user can drop filters using the --no-changed-paths option.
In the process, we need to be extremely careful to match the Bloom
filter settings as specified by the commit-graph. This will allow future
versions of Git to customize these settings, and the version with this
change will persist those settings as commit-graphs are rewritten on
top.
Use the trace2 API to signal the settings used during the write, and
check that output in a test after manually adjusting the correct bytes
in the commit-graph file.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The get_bloom_filter() method is a bit complicated in some parts where
it does not need to be. In particular, it needs to return a NULL filter
only when compute_if_not_present is zero AND the filter data cannot be
loaded from a commit-graph file. This currently happens by accident
because the commit-graph does not load changed-path Bloom filters from
an existing commit-graph when writing a new one. This will change in a
later patch.
Also clean up some style issues while we are here.
One side-effect of returning a NULL filter is that the filters that are
reported as "too large" will now be reported as NULL insead of length
zero. This case was not properly covered before, so add a test. Further,
remote the counting of the zero-length filters from revision.c and the
trace2 logs.
Helped-by: René Scharfe <l.s.r@web.de>
Helped-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
43d3561 (commit-graph write: don't die if the existing graph is corrupt,
2019-03-25) introduced the GIT_TEST_COMMIT_GRAPH_DIE_ON_LOAD environment
variable. This was created to verify that commit-graph was not loaded
when writing a new non-incremental commit-graph.
An upcoming change wants to load a commit-graph in some valuable cases,
but we want to maintain that we don't trust the commit-graph data when
writing our new file. Instead of dying on load, instead die if we ever
try to parse a commit from the commit-graph. This functionally verifies
the same intended behavior, but allows a more advanced feature in the
next change.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Place an instance of struct bloom_settings into the struct
write_commit_graph_context. This allows simplifying the function
prototype of write_graph_chunk_bloom_data(). This will allow us
to combine the function prototypes and use function pointers to
simplify write_commit_graph_file().
By using a pointer, we can later replace the settings to match those
that exist in the current commit-graph, in case a future Git version
allows customization of these parameters.
Reported-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In an earlier patch, multiple struct acccesses to `graph_pos` and
`generation` were auto-converted to multiple method calls.
Since the values are fixed and commit-slab access costly, we would be
better off with storing the values as a local variable and reusing it.
Signed-off-by: Abhishek Kumar <abhishekkumar8222@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We remove members `graph_pos` and `generation` from the struct commit.
The default assignments in init_commit_node() are no longer valid,
which is fine as the slab helpers return appropriate default values and
the assignments are removed.
We will replace existing use of commit->generation and commit->graph_pos
by commit_graph_data_slab helpers using
`contrib/coccinelle/commit.cocci'.
Signed-off-by: Abhishek Kumar <abhishekkumar8222@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The struct commit is used in many contexts. However, members
`generation` and `graph_pos` are only used for commit-graph related
operations and otherwise waste memory.
This wastage would have been more pronounced as we transition to
generation number v2, which uses 64-bit generation number instead of
current 32-bits.
As they are often accessed together, let's introduce struct
commit_graph_data and move them to a commit_graph_data slab.
While the overall test suite runs just as fast as master,
(series: 26m48s, master: 27m34s, faster by 2.87%), certain commands
like `git merge-base --is-ancestor` were slowed by 40% as discovered
by Szeder Gábor [1]. After minimizing commit-slab access, the slow down
persists but is closer to 20%.
Derrick Stolee believes the slow down is attributable to the underlying
algorithm rather than the slowness of commit-slab access [2] and we will
follow-up in a later series.
[1]: https://lore.kernel.org/git/20200607195347.GA8232@szeder.dev/
[2]: https://lore.kernel.org/git/13db757a-9412-7f1e-805c-8a028c4ab2b1@gmail.com/
Signed-off-by: Abhishek Kumar <abhishekkumar8222@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Unify the 'chunk_ids' and 'chunk_sizes' arrays into an array of
'struct chunk_info'. This will allow more cleanups in the following
patches.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In write_commit_graph_file() one block of code fills the array of
chunk IDs, another block of code fills the array of chunk offsets,
then the chunk IDs and offsets are written to the Chunk Lookup table,
and finally a third block of code writes the actual chunks. In case
of optional chunks like Extra Edge List and Base Graphs List there is
also a condition checking whether that chunk is necessary/desired, and
that same condition is repeated in all those three blocks of code.
This patch series is about to add more optional chunks, so there would
be even more repeated conditions.
Those chunk offsets are relative to the beginning of the file, so they
inherently depend on the size of the Chunk Lookup table, which in turn
depends on the number of chunks that are to be written to the
commit-graph file. IOW at the time we set the first chunk's ID we
can't yet know its offset, because we don't yet know how many chunks
there are.
Simplify this by initially filling an array of chunk sizes, not
offsets, and calculate the offsets based on the chunk sizes only
later, while we are writing the Chunk Lookup table. This way we can
fill the arrays of chunk IDs and sizes in one go, eliminating one set
of repeated conditions.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The Chunk Lookup table stores the chunks' starting offset in the
commit-graph file, not their sizes. Consequently, the size of a chunk
can only be calculated by subtracting its offset from the offset of
the subsequent chunk (or that of the terminating label). This is
currenly implemented in a bit complicated way: as we iterate over the
entries of the Chunk Lookup table, we check the id of each chunk and
store its starting offset, then we check the id of the last seen chunk
and calculate its size using its previously saved offset. At the
moment there is only one chunk for which we calculate its size, but
this patch series will add more, and the repeated chunk id checks are
not that pretty.
Instead let's read ahead the offset of the next chunk on each
iteration, so we can calculate the size of each chunk right away,
right where we store its starting offset.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
While we iterate over all entries of the Chunk Lookup table we make
sure that we don't attempt to read past the end of the mmap-ed
commit-graph file, and check in each iteration that the chunk ID and
offset we are about to read is still within the mmap-ed memory region.
However, these checks in each iteration are not really necessary,
because the number of chunks in the commit-graph file is already known
before this loop from the just parsed commit-graph header.
So let's check that the commit-graph file is large enough for all
entries in the Chunk Lookup table before we start iterating over those
entries, and drop those per-iteration checks. While at it, take into
account the size of everything that is necessary to have a valid
commit-graph file, i.e. the size of the header, the size of the
mandatory OID Fanout chunk, and the size of the signature in the
trailer as well.
Note that this necessitates the change of the error message as well,
and, consequently, have to update the 'detect incorrect chunk count'
test in 't5318-commit-graph.sh' as well.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Our CodingGuidelines says that it's sufficient to include one of
'git-compat-util.h' and 'cache.h', but both 'commit-graph.c' and
'commit-graph.h' include both. Let's include only 'git-compat-util.h'
to loose a bunch of unnecessary dependencies; but include 'hash.h',
because 'commit-graph.h' does require the definition of 'struct
object_id'.
'commit-graph.h' explicitly includes 'repository.h' and
'string-list.h', but only needs the declaration of a few structs from
them. Drop these includes and forward-declare the necessary structs
instead.
'commit-graph.c' includes 'dir.h', but doesn't actually use anything
from there, so let's drop that #include as well.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The commit-graph file format specifies that the chunks may be in any
order. However, if the OID Lookup chunk happens to be the last one in
the file, then any command attempting to access the commit-graph data
will fail with:
fatal: invalid commit position. commit-graph is likely corrupt
In this case the error is wrong, the commit-graph file does conform to
the specification, but the parsing of the Chunk Lookup table is a bit
buggy, and leaves the field holding the number of commits in the
commit-graph zero-initialized.
The number of commits in the commit-graph is determined while parsing
the Chunk Lookup table, by dividing the size of the OID Lookup chunk
with the hash size. However, the Chunk Lookup table doesn't actually
store the size of the chunks, but it stores their starting offset.
Consequently, the size of a chunk can only be calculated by
subtracting the starting offsets of that chunk from the offset of the
subsequent chunk, or in case of the last chunk from the offset
recorded in the terminating label. This is currenly implemented in a
bit complicated way: as we iterate over the entries of the Chunk
Lookup table, we check the ID of each chunk and store its starting
offset, then we check the ID of the last seen chunk and calculate its
size using its previously saved offset if necessary (at the moment
it's only necessary for the OID Lookup chunk). Alas, while parsing
the Chunk Lookup table we only interate through the "real" chunks, but
never look at the terminating label, thus don't even check whether
it's necessary to calulate the size of the last chunk. Consequently,
if the OID Lookup chunk is the last one, then we don't calculate its
size and turn don't run the piece of code determining the number of
commits in the commit graph, leaving the field holding that number
unchanged (i.e. zero-initialized), eventually triggering the sanity
check in load_oid_from_graph().
Fix this by iterating through all entries in the Chunk Lookup table,
including the terminating label.
Note that this is the minimal fix, suitable for the maintenance track.
A better fix would be to simplify how the chunk sizes are calculated,
but that is a more invasive change, less suitable for 'maint', so that
will be done in later patches.
This additional flexibility of scanning more chunks breaks a test for
"git commit-graph verify" so alter that test to mutate the commit-graph
to have an even lower chunk count.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since 7c5c9b9c57 (commit-graph: error out on invalid commit oids in
'write --stdin-commits', 2019-08-05), the commit-graph builtin dies on
receiving non-commit OIDs as input to '--stdin-commits'.
This behavior can be cumbersome to work around in, say, the case of
piping 'git for-each-ref' to 'git commit-graph write --stdin-commits' if
the caller does not want to cull out non-commits themselves. In this
situation, it would be ideal if 'git commit-graph write' wrote the graph
containing the inputs that did pertain to commits, and silently ignored
the remainder of the input.
Some options have been proposed to the effect of '--[no-]check-oids'
which would allow callers to have the commit-graph builtin do just that.
After some discussion, it is difficult to imagine a caller who wouldn't
want to pass '--no-check-oids', suggesting that we should get rid of the
behavior of complaining about non-commit inputs altogether.
If callers do wish to retain this behavior, they can easily work around
this change by doing the following:
git for-each-ref --format='%(objectname) %(objecttype) %(*objecttype)' |
awk '
!/commit/ { print "not-a-commit:"$1 }
/commit/ { print $1 }
' |
git commit-graph write --stdin-commits
To make it so that valid OIDs that refer to non-existent objects are
indeed an error after loosening the error handling, perform an extra
lookup to make sure that object indeed exists before sending it to the
commit-graph internals.
Helped-by: Jeff King <peff@peff.net>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In the previous handful of commits, both 'git commit-graph write
--reachable' and '--stdin-commits' learned to peel tags down to the
commits which they refer to before passing them into the commit-graph
internals.
This makes the call to 'lookup_commit_reference_gently()' inside of
'fill_oids_from_commits()' a noop, since all OIDs are commits by that
point.
As such, remove the call entirely, as well as the progress meter, which
has been split and moved out to the callers in the aforementioned
earlier commits.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
While iterating references (to discover the set of commits to write to
the commit-graph with 'git commit-graph write --reachable'),
'add_ref_to_set' can save 'fill_oids_from_commits()' some time by
peeling the references beforehand.
Move peeling out of 'fill_oids_from_commits()' and into
'add_ref_to_set()' to use 'peel_ref()' instead of 'deref_tag()'. Doing
so allows the commit-graph machinery to use the peeled value from
'$GIT_DIR/packed-refs' instead of having to load and parse tags.
While we're at it, discard non-commit objects reachable from ref tips.
This would be done automatically by 'fill_oids_from_commits()', but such
functionality will be removed in a subsequent patch after the call to
'lookup_commit_reference_gently' is dropped (at which point a non-commit
object in the commits oidset will become an error).
Suggested-by: Jeff King <peff@peff.net>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When 'git commit-graph write --reachable' is invoked, the commit-graph
machinery calls 'for_each_ref()' to discover the set of reachable
commits.
Right now the 'add_ref_to_set' callback is not doing anything other than
adding an OID to the set of known-reachable OIDs. In a subsequent
commit, 'add_ref_to_set' will presumptively peel references. This
operation should be fast for repositories with an up-to-date
'$GIT_DIR/packed-refs', but may be slow in the general case.
So that it doesn't appear that 'git commit-graph write' is idling with
'--reachable' in the slow case, add a progress meter to provide some
output in the meantime.
In general, we don't expect a progress meter to appear at all, since
peeling references with a 'packed-refs' file is quick. If it's slow and
we do show a progress meter, the subsequent 'fill_oids_from_commits()'
will be fast, since all of the calls to
'lookup_commit_reference_gently()' will be no-ops.
Both progress meters are delayed, so it is unlikely that more than one
will appear. In either case, this intermediate state will go away in a
handful of patches, at which point there will be at most one progress
meter.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Some of the files commit-graph subsystem keeps on disk did not
correctly honor the core.sharedRepository settings and some were
left read-write.
* tb/commit-graph-perm-bits:
commit-graph.c: make 'commit-graph-chain's read-only
commit-graph.c: ensure graph layers respect core.sharedRepository
commit-graph.c: write non-split graphs as read-only
lockfile.c: introduce 'hold_lock_file_for_update_mode'
tempfile.c: introduce 'create_tempfile_mode'
In subsequent patches, we are going to update a progress meter when
'add_ref_to_set()' is called, and need a convenient way to pass a
'struct progress *' in from the caller.
Introduce 'refs_cb_data' as a catch-all for parameters that
'add_ref_to_set' may need, and wrap the existing single parameter in
that struct.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A fuzzer running on the entry point provided by fuzz-commit-graph.c
revealed a memory leak when parse_commit_graph() creates a struct
bloom_filter_settings and then returns early due to error. Fix that
error by always freeing that struct first (if it exists) before
returning early due to error.
While making that change, I also noticed another possible memory leak -
when the BLOOMDATA chunk is provided but not BLOOMINDEXES. Also fix that
error.
Signed-off-by: Jonathan Tan <jonathantanmy@google.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git blame" learns to take advantage of the "changed-paths" Bloom
filter stored in the commit-graph file.
* ds/blame-on-bloom:
test-bloom: check that we have expected arguments
test-bloom: fix some whitespace issues
blame: drop unused parameter from maybe_changed_path
blame: use changed-path Bloom filters
tests: write commit-graph with Bloom filters
revision: complicated pathspecs disable filters
Introduce an extension to the commit-graph to make it efficient to
check for the paths that were modified at each commit using Bloom
filters.
* gs/commit-graph-path-filter:
bloom: ignore renames when computing changed paths
commit-graph: add GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS test flag
t4216: add end to end tests for git log with Bloom filters
revision.c: add trace2 stats around Bloom filter usage
revision.c: use Bloom filters to speed up path based revision walks
commit-graph: add --changed-paths option to write subcommand
commit-graph: reuse existing Bloom filters during write
commit-graph: write Bloom filters to commit graph file
commit-graph: examine commits by generation number
commit-graph: examine changed-path objects in pack order
commit-graph: compute Bloom filters for changed paths
diff: halt tree-diff early after max_changes
bloom.c: core Bloom filter implementation for changed paths.
bloom.c: introduce core Bloom filter constructs
bloom.c: add the murmur3 hash implementation
commit-graph: define and use MAX_NUM_CHUNKS
The commit-graph code exhausted file descriptors easily when it
does not have to.
* tb/commit-graph-fd-exhaustion-fix:
commit-graph: close descriptors after mmap
commit-graph.c: gracefully handle file descriptor exhaustion
t/test-lib.sh: make ULIMIT_FILE_DESCRIPTORS available to tests
commit-graph.c: don't use discarded graph_name in error
There are many functions in commit.h that are more related to shallow
repositories than they are to any sort of generic commit machinery.
Likely this began when there were only a few shallow-related functions,
and commit.h seemed a reasonable enough place to put them.
But, now there are a good number of shallow-related functions, and
placing them all in 'commit.h' doesn't make sense.
This patch extracts a 'shallow.h', which takes all of the declarations
from 'commit.h' for functions which already exist in 'shallow.c'. We
will bring the remaining shallow-related functions defined in 'commit.c'
in a subsequent patch.
For now, move only the ones that already are implemented in 'shallow.c',
and update the necessary includes.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In a previous commit, we made incremental graph layers read-only by
using 'git_mkstemp_mode' with permissions '0444'.
There is no reason that 'commit-graph-chain's should be modifiable by
the user, since they are generated at a temporary location and then
atomically renamed into place.
To ensure that these files are read-only, too, use
'hold_lock_file_for_update_mode' with the same read-only permission
bits, and let the umask and 'adjust_shared_perm' take care of the rest.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Non-layered commit-graphs use 'adjust_shared_perm' to make the
commit-graph file readable (or not) to a combination of the user, group,
and others.
Call 'adjust_shared_perm' for split-graph layers to make sure that these
also respect 'core.sharedRepository'. The 'commit-graph-chain' file
already respects this configuration since it uses
'hold_lock_file_for_update' (which calls 'adjust_shared_perm' eventually
in 'create_tempfile_mode').
Suggested-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In the previous commit, Git learned 'hold_lock_file_for_update_mode' to
allow the caller to specify the permission bits (prior to further
adjustment by the umask and shared repository permissions) used when
acquiring a temporary file.
Use this in the commit-graph machinery for writing a non-split graph to
acquire an opened temporary file with permissions read-only permissions
to match the split behavior. (In the split case, Git uses
git_mkstemp_mode' for each of the commit-graph layers with permission
bits '0444').
One can notice this discrepancy when moving a non-split graph to be part
of a new chain. This causes a commit-graph chain where all layers have
read-only permission bits, except for the base layer, which is writable
for the current user.
Resolve this discrepancy by using the new
'hold_lock_file_for_update_mode' and passing the desired permission
bits.
Doing so causes some test fallout in t5318 and t6600. In t5318, this
occurs in tests that corrupt a commit-graph file by writing into it. For
these, 'chmod u+w'-ing the file beforehand resolves the issue. The
additional spot in 'corrupt_graph_verify' is necessary because of the
extra 'git commit-graph write' beforehand (which *does* rewrite the
commit-graph file). In t6600, this is caused by copying a read-only
commit-graph file into place and then trying to replace it. For these,
make these files writable.
Helped-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git commit-graph write --expire-time=<timestamp>" did not use the
given timestamp correctly, which has been corrected.
* ds/commit-graph-expiry-fix:
commit-graph: fix buggy --expire-time option
We don't ever refer to the descriptor after mmap-ing it. And keeping it
open means we can run out of descriptors in degenerate cases (e.g.,
thousands of split chain files). Let's close it as soon as possible.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When writing a layered commit-graph, the commit-graph machinery uses
'commit_graph_filenames_after' and 'commit_graph_hash_after' to keep
track of the layers in the chain that we are in the process of writing.
When the number of commit-graph layers shrinks, we initialize all
entries in the aforementioned arrays, because we know the structure of
the new commit-graph chain immediately (since there are no new layers,
there are no unknown hash values).
But when the number of commit-graph layers grows (i.e., that
'num_commit_graphs_after > num_commit_graphs_before'), then we leave
some entries in the filenames and hashes arrays as uninitialized,
because we will fill them in later as those values become available.
For instance, we rely on 'write_commit_graph_file's to store the
filename and hash of the last layer in the new chain, which is the one
that it is responsible for writing. But, it's possible that
'write_commit_graph_file' may fail, e.g., from file descriptor
exhaustion. In this case it is possible that 'git_mkstemp_mode' will
fail, and that function will return early *before* setting the values
for the last commit-graph layer's filename and hash.
This causes a number of upleasant side-effects. For instance, trying to
'free()' each entry in 'ctx->commit_graph_filenames_after' (and
similarly for the hashes array) causes us to 'free()' uninitialized
memory, since the area is allocated with 'malloc()' and is therefore
subject to contain garbage (which is left alone when
'write_commit_graph_file' returns early).
This can manifest in other issues, like a general protection fault,
and/or leaving a stray 'commit-graph-chain.lock' around after the
process dies. (The reasoning for this is still a mystery to me, since
we'd otherwise usually expect the kernel to run tempfile.c's 'atexit()'
handlers in the case of a normal death...)
To resolve this, initialize the memory with 'CALLOC_ARRAY' so that
uninitialized entries are filled with zeros, and can thus be 'free()'d
as a noop instead of causing a fault.
Helped-by: Jeff King <peff@peff.net>
Helped-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When writing a commit-graph layer, we do so in a temporary file which is
renamed into place. If we fail to create a temporary file, for e.g.,
because we have too many open files, then 'git_mkstemp_mode' sets the
pattern to the empty string, in which case we get an error something
along the lines of:
error: unable to create ''
It's not useful to show the pattern here at all, since we (1) know the
pattern is well-formed, and (2) would have already shown the dirname
when trying to create the leading directories. So, replace this error
with something friendlier.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The GIT_TEST_COMMIT_GRAPH environment variable updates the commit-
graph file whenever "git commit" is run, ensuring that we always
have an updated commit-graph throughout the test suite. The
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS environment variable was
introduced to write the changed-path Bloom filters whenever "git
commit-graph write" is run. However, the GIT_TEST_COMMIT_GRAPH
trick doesn't launch a separate process and instead writes it
directly.
To expand the number of tests that have commits in the commit-graph
file, add a helper method that computes the commit-graph and place
that helper inside "git commit" and "git merge".
In the helper method, check GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS
to ensure we are writing changed-path Bloom filters whenever
possible.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When operating on a stream of commit OIDs on stdin, 'git commit-graph
write' checks that each OID refers to an object that is indeed a commit.
This is convenient to make sure that the given input is well-formed, but
can sometimes be undesirable.
For example, server operators may wish to feed the refnames that were
updated during a push to 'git commit-graph write --input=stdin-commits',
and silently discard refs that don't point at commits. This can be done
by combing the output of 'git for-each-ref' with '--format
%(*objecttype)', but this requires opening up a potentially large number
of objects. Instead, it is more convenient to feed the updated refs to
the commit-graph machinery, and let it throw out refs that don't point
to commits.
Introduce '--[no-]check-oids' to make such a behavior possible. With
'--check-oids' (the default behavior to retain backwards compatibility),
'git commit-graph write' will barf on a non-commit line in its input.
With 'no-check-oids', such lines will be silently ignored, making the
above possible by specifying this option.
No matter which is supplied, 'git commit-graph write' retains the
behavior from the previous commit of rejecting non-OID inputs like
"HEAD" and "refs/heads/foo" as before.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The 'write_commit_graph()' function takes in either a string list of
pack indices, or a string list of hexadecimal commit OIDs. These
correspond to the '--stdin-packs' and '--stdin-commits' mode(s) from
'git commit-graph write'.
Using a string_list of hexadecimal commit IDs is not the most efficient
use of memory, since we can instead use the 'struct oidset', which is
more well-suited for this case.
This has another benefit which will become apparent in the following
commit. This is that we are about to disambiguate the kinds of errors we
produce with '--stdin-commits' into "non-hex input" and "hex-input, but
referring to a non-commit object". By having 'write_commit_graph' take
in a 'struct oidset *' of commits, we place the burden on the caller (in
this case, the builtin) to handle the first case, and the commit-graph
machinery can handle the second case.
Suggested-by: Jeff King <peff@peff.net>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When using split commit-graphs, it is sometimes useful to completely
replace the commit-graph chain with a new base.
For example, consider a scenario in which a repository builds a new
commit-graph incremental for each push. Occasionally (say, after some
fixed number of pushes), they may wish to rebuild the commit-graph chain
with all reachable commits.
They can do so with
$ git commit-graph write --reachable
but this removes the chain entirely and replaces it with a single
commit-graph in 'objects/info/commit-graph'. Unfortunately, this means
that the next push will have to move this commit-graph into the first
layer of a new chain, and then write its new commits on top.
Avoid such copying entirely by allowing the caller to specify that they
wish to replace the entirety of their commit-graph chain, while also
specifying that the new commit-graph should become the basis of a fresh,
length-one chain.
This addresses the above situation by making it possible for the caller
to instead write:
$ git commit-graph write --reachable --split=replace
which writes a new length-one chain to 'objects/info/commit-graphs',
making the commit-graph incremental generated by the subsequent push
relatively cheap by avoiding the aforementioned copy.
In order to do this, remove an assumption in 'write_commit_graph_file'
that chains are always at least two incrementals long.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In the previous commit, we laid the groundwork for supporting different
splitting strategies. In this commit, we introduce the first splitting
strategy: 'no-merge'.
Passing '--split=no-merge' is useful for callers which wish to write a
new incremental commit-graph, but do not want to spend effort condensing
the incremental chain [1]. Previously, this was possible by passing
'--size-multiple=0', but this no longer the case following 63020f175f
(commit-graph: prefer default size_mult when given zero, 2020-01-02).
When '--split=no-merge' is given, the commit-graph machinery will never
condense an existing chain, and it will always write a new incremental.
[1]: This might occur when, for example, a server administrator running
some program after each push may want to ensure that each job runs
proportional in time to the size of the push, and does not "jump" when
the commit-graph machinery decides to trigger a merge.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add logic to
a) parse Bloom filter information from the commit graph file and,
b) re-use existing Bloom filters.
See Documentation/technical/commit-graph-format for the format in which
the Bloom filter information is written to the commit graph file.
To read Bloom filter for a given commit with lexicographic position
'i' we need to:
1. Read BIDX[i] which essentially gives us the starting index in BDAT for
filter of commit i+1. It is essentially the index past the end
of the filter of commit i. It is called end_index in the code.
2. For i>0, read BIDX[i-1] which will give us the starting index in BDAT
for filter of commit i. It is called the start_index in the code.
For the first commit, where i = 0, Bloom filter data starts at the
beginning, just past the header in the BDAT chunk. Hence, start_index
will be 0.
3. The length of the filter will be end_index - start_index, because
BIDX[i] gives the cumulative 8-byte words including the ith
commit's filter.
We toggle whether Bloom filters should be recomputed based on the
compute_if_not_present flag.
Helped-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Garima Singh <garima.singh@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Update the technical documentation for commit-graph-format with
the formats for the Bloom filter index (BIDX) and Bloom filter
data (BDAT) chunks. Write the computed Bloom filters information
to the commit graph file using this format.
Helped-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Garima Singh <garima.singh@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The commit-graph builtin has an --expire-time option that takes a
datetime using OPT_EXPIRY_DATE(). However, the implementation inside
expire_commit_graphs() was treating a non-zero value as a number of
seconds to subtract from "now".
Update t5323-split-commit-graph.sh to demonstrate the correct value
of the --expire-time option by actually creating a crud .graph file
with mtime earlier than the expire time. Instead of using a super-
early time (1980) we use an explicit, and recent, time. Using
test-tool chmtime to create two files on either end of an exact
second, we create a test that catches this failure no matter the
current time. Using a fixed date is more portable than trying to
format a relative date string into the --expiry-date input.
I noticed this when inspecting some Scalar repos that had an excess
number of commit-graph files. In Scalar, we were using this second
interpretation by using "--expire-time=3600" to mean "delete graphs
older than one hour ago" to avoid deleting a commit-graph that a
foreground process may be trying to load.
Also I noticed that the help text was copied from the --max-commits
option. Fix that help text.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When running 'git commit-graph write --changed-paths', we sort the
commits by pack-order to save time when computing the changed-paths
bloom filters. This does not help when finding the commits via the
'--reachable' flag.
If not using pack-order, then sort by generation number before
examining the diff. Commits with similar generation are more likely
to have many trees in common, making the diff faster.
On the Linux kernel repository, this change reduced the computation
time for 'git commit-graph write --reachable --changed-paths' from
3m00s to 1m37s.
Helped-by: Jeff King <peff@peff.net>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Garima Singh <garima.singh@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
