git/t/t4216-log-bloom.sh

458 lines
14 KiB
Bash
Raw Normal View History

#!/bin/sh
test_description='git log for a path with Bloom filters'
GIT_TEST_DEFAULT_INITIAL_BRANCH_NAME=main
tests: mark tests relying on the current default for `init.defaultBranch` In addition to the manual adjustment to let the `linux-gcc` CI job run the test suite with `master` and then with `main`, this patch makes sure that GIT_TEST_DEFAULT_INITIAL_BRANCH_NAME is set in all test scripts that currently rely on the initial branch name being `master by default. To determine which test scripts to mark up, the first step was to force-set the default branch name to `master` in - all test scripts that contain the keyword `master`, - t4211, which expects `t/t4211/history.export` with a hard-coded ref to initialize the default branch, - t5560 because it sources `t/t556x_common` which uses `master`, - t8002 and t8012 because both source `t/annotate-tests.sh` which also uses `master`) This trick was performed by this command: $ sed -i '/^ *\. \.\/\(test-lib\|lib-\(bash\|cvs\|git-svn\)\|gitweb-lib\)\.sh$/i\ GIT_TEST_DEFAULT_INITIAL_BRANCH_NAME=master\ export GIT_TEST_DEFAULT_INITIAL_BRANCH_NAME\ ' $(git grep -l master t/t[0-9]*.sh) \ t/t4211*.sh t/t5560*.sh t/t8002*.sh t/t8012*.sh After that, careful, manual inspection revealed that some of the test scripts containing the needle `master` do not actually rely on a specific default branch name: either they mention `master` only in a comment, or they initialize that branch specificially, or they do not actually refer to the current default branch. Therefore, the aforementioned modification was undone in those test scripts thusly: $ git checkout HEAD -- \ t/t0027-auto-crlf.sh t/t0060-path-utils.sh \ t/t1011-read-tree-sparse-checkout.sh \ t/t1305-config-include.sh t/t1309-early-config.sh \ t/t1402-check-ref-format.sh t/t1450-fsck.sh \ t/t2024-checkout-dwim.sh \ t/t2106-update-index-assume-unchanged.sh \ t/t3040-subprojects-basic.sh t/t3301-notes.sh \ t/t3308-notes-merge.sh t/t3423-rebase-reword.sh \ t/t3436-rebase-more-options.sh \ t/t4015-diff-whitespace.sh t/t4257-am-interactive.sh \ t/t5323-pack-redundant.sh t/t5401-update-hooks.sh \ t/t5511-refspec.sh t/t5526-fetch-submodules.sh \ t/t5529-push-errors.sh t/t5530-upload-pack-error.sh \ t/t5548-push-porcelain.sh \ t/t5552-skipping-fetch-negotiator.sh \ t/t5572-pull-submodule.sh t/t5608-clone-2gb.sh \ t/t5614-clone-submodules-shallow.sh \ t/t7508-status.sh t/t7606-merge-custom.sh \ t/t9302-fast-import-unpack-limit.sh We excluded one set of test scripts in these commands, though: the range of `git p4` tests. The reason? `git p4` stores the (foreign) remote branch in the branch called `p4/master`, which is obviously not the default branch. Manual analysis revealed that only five of these tests actually require a specific default branch name to pass; They were modified thusly: $ sed -i '/^ *\. \.\/lib-git-p4\.sh$/i\ GIT_TEST_DEFAULT_INITIAL_BRANCH_NAME=master\ export GIT_TEST_DEFAULT_INITIAL_BRANCH_NAME\ ' t/t980[0167]*.sh t/t9811*.sh Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-11-18 23:44:19 +00:00
export GIT_TEST_DEFAULT_INITIAL_BRANCH_NAME
. ./test-lib.sh
commit-graph: check bounds when accessing BDAT chunk When loading Bloom filters from a commit-graph file, we use the offset values in the BIDX chunk to index into the memory mapped for the BDAT chunk. But since we don't record how big the BDAT chunk is, we just trust that the BIDX offsets won't cause us to read outside of the chunk memory. A corrupted or malicious commit-graph file will cause us to segfault (in practice this isn't a very interesting attack, since commit-graph files are local-only, and the worst case is an out-of-bounds read). We can't fix this by checking the chunk size during parsing, since the data in the BDAT chunk doesn't have a fixed size (that's why we need the BIDX in the first place). So we'll fix it in two parts: 1. Record the BDAT chunk size during parsing, and then later check that the BIDX offsets we look up are within bounds. 2. Because the offsets are relative to the end of the BDAT header, we must also make sure that the BDAT chunk is at least as large as the expected header size. Otherwise, we overflow when trying to move past the header, even for an offset of "0". We can check this early, during the parsing stage. The error messages are rather verbose, but since this is not something you'd expect to see outside of severe bugs or corruption, it makes sense to err on the side of too many details. Sadly we can't mention the filename during the chunk-parsing stage, as we haven't set g->filename at this point, nor passed it down through the stack. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-10-09 21:05:50 +00:00
. "$TEST_DIRECTORY"/lib-chunk.sh
GIT_TEST_COMMIT_GRAPH=0
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS=0
test_expect_success 'setup test - repo, commits, commit graph, log outputs' '
git init &&
mkdir A A/B A/B/C &&
test_commit c1 A/file1 &&
test_commit c2 A/B/file2 &&
test_commit c3 A/B/C/file3 &&
test_commit c4 A/file1 &&
test_commit c5 A/B/file2 &&
test_commit c6 A/B/C/file3 &&
test_commit c7 A/file1 &&
test_commit c8 A/B/file2 &&
test_commit c9 A/B/C/file3 &&
test_commit c10 file_to_be_deleted &&
git checkout -b side HEAD~4 &&
test_commit side-1 file4 &&
git checkout main &&
git merge side &&
test_commit c11 file5 &&
mv file5 file5_renamed &&
git add file5_renamed &&
git commit -m "rename" &&
rm file_to_be_deleted &&
git add . &&
git commit -m "file removed" &&
bloom: encode out-of-bounds filters as non-empty 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>
2020-09-18 02:59:44 +00:00
git commit --allow-empty -m "empty" &&
commit-graph: use the "hash version" byte 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>
2020-08-17 14:04:47 +00:00
git commit-graph write --reachable --changed-paths &&
test_oid_cache <<-EOF
oid_version sha1:1
oid_version sha256:2
EOF
'
bloom: encode out-of-bounds filters as non-empty 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>
2020-09-18 02:59:44 +00:00
graph_read_expect () {
commit-graph: implement generation data chunk 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>
2021-01-16 18:11:15 +00:00
NUM_CHUNKS=6
cat >expect <<- EOF
commit-graph: use the "hash version" byte 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>
2020-08-17 14:04:47 +00:00
header: 43475048 1 $(test_oid oid_version) $NUM_CHUNKS 0
num_commits: $1
commit-graph: implement generation data chunk 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>
2021-01-16 18:11:15 +00:00
chunks: oid_fanout oid_lookup commit_metadata generation_data bloom_indexes bloom_data
commit-graph: start parsing generation v2 (again) The 'read_generation_data' member of 'struct commit_graph' was introduced by 1fdc383c5 (commit-graph: use generation v2 only if entire chain does, 2021-01-16). The intention was to avoid using corrected commit dates if not all layers of a commit-graph had that data stored. The logic in validate_mixed_generation_chain() at that point incorrectly initialized read_generation_data to 1 if and only if the tip commit-graph contained the Corrected Commit Date chunk. This was "fixed" in 448a39e65 (commit-graph: validate layers for generation data, 2021-02-02) to validate that read_generation_data was either non-zero for all layers, or it would set read_generation_data to zero for all layers. The problem here is that read_generation_data is not initialized to be non-zero anywhere! This change initializes read_generation_data immediately after the chunk is parsed, so each layer will have its value present as soon as possible. The read_generation_data member is used in fill_commit_graph_info() to determine if we should use the corrected commit date or the topological levels stored in the Commit Data chunk. Due to this bug, all previous versions of Git were defaulting to topological levels in all cases! This can be measured with some performance tests. Using the Linux kernel as a testbed, I generated a complete commit-graph containing corrected commit dates and tested the 'new' version against the previous, 'old' version. First, rev-list with --topo-order demonstrates a 26% improvement using corrected commit dates: hyperfine \ -n "old" "$OLD_GIT rev-list --topo-order -1000 v3.6" \ -n "new" "$NEW_GIT rev-list --topo-order -1000 v3.6" \ --warmup=10 Benchmark 1: old Time (mean ± σ): 57.1 ms ± 3.1 ms Range (min … max): 52.9 ms … 62.0 ms 55 runs Benchmark 2: new Time (mean ± σ): 45.5 ms ± 3.3 ms Range (min … max): 39.9 ms … 51.7 ms 59 runs Summary 'new' ran 1.26 ± 0.11 times faster than 'old' These performance improvements are due to the algorithmic improvements given by walking fewer commits due to the higher cutoffs from corrected commit dates. However, this comes at a cost. The additional I/O cost of parsing the corrected commit dates is visible in case of merge-base commands that do not reduce the overall number of walked commits. hyperfine \ -n "old" "$OLD_GIT merge-base v4.8 v4.9" \ -n "new" "$NEW_GIT merge-base v4.8 v4.9" \ --warmup=10 Benchmark 1: old Time (mean ± σ): 110.4 ms ± 6.4 ms Range (min … max): 96.0 ms … 118.3 ms 25 runs Benchmark 2: new Time (mean ± σ): 150.7 ms ± 1.1 ms Range (min … max): 149.3 ms … 153.4 ms 19 runs Summary 'old' ran 1.36 ± 0.08 times faster than 'new' Performance issues like this are what motivated 702110aac (commit-graph: use config to specify generation type, 2021-02-25). In the future, we could fix this performance problem by inserting the corrected commit date offsets into the Commit Date chunk instead of having that data in an extra chunk. Signed-off-by: Derrick Stolee <derrickstolee@github.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-03-01 19:48:31 +00:00
options: bloom(1,10,7) read_generation_data
EOF
test-tool read-graph >actual &&
test_cmp expect actual
}
test_expect_success 'commit-graph write wrote out the bloom chunks' '
bloom: encode out-of-bounds filters as non-empty 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>
2020-09-18 02:59:44 +00:00
graph_read_expect 16
'
# Turn off any inherited trace2 settings for this test.
sane_unset GIT_TRACE2 GIT_TRACE2_PERF GIT_TRACE2_EVENT
sane_unset GIT_TRACE2_PERF_BRIEF
sane_unset GIT_TRACE2_CONFIG_PARAMS
setup () {
rm -f "$TRASH_DIRECTORY/trace.perf" &&
git -c core.commitGraph=false log --pretty="format:%s" $1 >log_wo_bloom &&
GIT_TRACE2_PERF="$TRASH_DIRECTORY/trace.perf" git -c core.commitGraph=true log --pretty="format:%s" $1 >log_w_bloom
}
test_bloom_filters_used () {
log_args=$1
commit-graph: introduce 'get_bloom_filter_settings()' Many places in the code often need a pointer to the commit-graph's 'struct bloom_filter_settings', in which case they often take the value from the top-most commit-graph. In the non-split case, this works as expected. In the split case, however, things get a little tricky. Not all layers in a chain of incremental commit-graphs are required to themselves have Bloom data, and so whether or not some part of the code uses Bloom filters depends entirely on whether or not the top-most level of the commit-graph chain has Bloom filters. This has been the behavior since Bloom filters were introduced, and has been codified into the tests since a759bfa9ee (t4216: add end to end tests for git log with Bloom filters, 2020-04-06). In fact, t4216.130 requires that Bloom filters are not used in exactly the case described earlier. There is no reason that this needs to be the case, since it is perfectly valid for commits in an earlier layer to have Bloom filters when commits in a newer layer do not. Since Bloom settings are guaranteed in practice to be the same for any layer in a chain that has Bloom data, it is sufficient to traverse the '->base_graph' pointer until either (1) a non-null 'struct bloom_filter_settings *' is found, or (2) until we are at the root of the commit-graph chain. Introduce a 'get_bloom_filter_settings()' function that does just this, and use it instead of purely dereferencing the top-most graph's '->bloom_filter_settings' pointer. While we're at it, add an additional test in t5324 to guard against code in the commit-graph writing machinery that doesn't correctly handle a NULL 'struct bloom_filter *'. Co-authored-by: Derrick Stolee <dstolee@microsoft.com> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-09-09 15:22:44 +00:00
bloom_trace_prefix="statistics:{\"filter_not_present\":${2:-0},\"maybe\""
setup "$log_args" &&
grep -q "$bloom_trace_prefix" "$TRASH_DIRECTORY/trace.perf" &&
test_cmp log_wo_bloom log_w_bloom &&
test_path_is_file "$TRASH_DIRECTORY/trace.perf"
}
test_bloom_filters_not_used () {
log_args=$1
setup "$log_args" &&
! grep -q "statistics:{\"filter_not_present\":" "$TRASH_DIRECTORY/trace.perf" &&
test_cmp log_wo_bloom log_w_bloom
}
for path in A A/B A/B/C A/file1 A/B/file2 A/B/C/file3 file4 file5 file5_renamed file_to_be_deleted
do
for option in "" \
"--all" \
"--full-history" \
"--full-history --simplify-merges" \
"--simplify-merges" \
"--simplify-by-decoration" \
"--follow" \
"--first-parent" \
"--topo-order" \
"--date-order" \
"--author-date-order" \
"--ancestry-path side..main"
do
test_expect_success "git log option: $option for path: $path" '
test_bloom_filters_used "$option -- $path" &&
test_config commitgraph.readChangedPaths false &&
test_bloom_filters_not_used "$option -- $path"
'
done
done
test_expect_success 'git log -- folder works with and without the trailing slash' '
test_bloom_filters_used "-- A" &&
test_bloom_filters_used "-- A/"
'
test_expect_success 'git log for path that does not exist. ' '
test_bloom_filters_used "-- path_does_not_exist"
'
test_expect_success 'git log with --walk-reflogs does not use Bloom filters' '
test_bloom_filters_not_used "--walk-reflogs -- A"
'
test_expect_success 'git log -- multiple path specs does not use Bloom filters' '
test_bloom_filters_not_used "-- file4 A/file1"
'
test_expect_success 'git log -- "." pathspec at root does not use Bloom filters' '
test_bloom_filters_not_used "-- ."
'
test_expect_success 'git log with wildcard that resolves to a single path uses Bloom filters' '
test_bloom_filters_used "-- *4" &&
test_bloom_filters_used "-- *renamed"
'
test_expect_success 'git log with wildcard that resolves to a multiple paths does not uses Bloom filters' '
test_bloom_filters_not_used "-- *" &&
test_bloom_filters_not_used "-- file*"
'
test_expect_success 'setup - add commit-graph to the chain without Bloom filters' '
test_commit c14 A/anotherFile2 &&
test_commit c15 A/B/anotherFile2 &&
test_commit c16 A/B/C/anotherFile2 &&
git commit-graph write --reachable --split --no-changed-paths &&
test_line_count = 2 .git/objects/info/commit-graphs/commit-graph-chain
'
commit-graph: introduce 'get_bloom_filter_settings()' 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>
2020-09-09 15:22:44 +00:00
test_expect_success 'use Bloom filters even if the latest graph does not have Bloom filters' '
# Ensure that the number of empty filters is equal to the number of
# filters in the latest graph layer to prove that they are loaded (and
# ignored).
test_bloom_filters_used "-- A/B" 3
'
test_expect_success 'setup - add commit-graph to the chain with Bloom filters' '
test_commit c17 A/anotherFile3 &&
git commit-graph write --reachable --changed-paths --split &&
test_line_count = 3 .git/objects/info/commit-graphs/commit-graph-chain
'
test_bloom_filters_used_when_some_filters_are_missing () {
log_args=$1
bloom: encode out-of-bounds filters as non-empty 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>
2020-09-18 02:59:44 +00:00
bloom_trace_prefix="statistics:{\"filter_not_present\":3,\"maybe\":6,\"definitely_not\":9"
setup "$log_args" &&
grep -q "$bloom_trace_prefix" "$TRASH_DIRECTORY/trace.perf" &&
test_cmp log_wo_bloom log_w_bloom
}
test_expect_success 'Use Bloom filters if they exist in the latest but not all commit graphs in the chain.' '
test_bloom_filters_used_when_some_filters_are_missing "-- A/B"
'
test_expect_success 'persist filter settings' '
test_when_finished rm -rf .git/objects/info/commit-graph* &&
rm -rf .git/objects/info/commit-graph* &&
GIT_TRACE2_EVENT="$(pwd)/trace2.txt" \
GIT_TEST_BLOOM_SETTINGS_NUM_HASHES=9 \
GIT_TEST_BLOOM_SETTINGS_BITS_PER_ENTRY=15 \
git commit-graph write --reachable --changed-paths &&
grep "{\"hash_version\":1,\"num_hashes\":9,\"bits_per_entry\":15,\"max_changed_paths\":512" trace2.txt &&
GIT_TRACE2_EVENT="$(pwd)/trace2-auto.txt" \
git commit-graph write --reachable --changed-paths &&
grep "{\"hash_version\":1,\"num_hashes\":9,\"bits_per_entry\":15,\"max_changed_paths\":512" trace2-auto.txt
'
bloom/diff: properly short-circuit on max_changes Commit e3696980 (diff: halt tree-diff early after max_changes, 2020-03-30) intended to create a mechanism to short-circuit a diff calculation after a certain number of paths were modified. By incrementing a "num_changes" counter throughout the recursive ll_diff_tree_paths(), this was supposed to match the number of changes that would be written into the changed-path Bloom filters. Unfortunately, this was not implemented correctly and instead misses simple cases like file modifications. This then does not stop very large changed-path filters from being written (unless they add or remove many files). To start, change the implementation in ll_diff_tree_paths() to instead use the global diff_queue_diff struct's 'nr' member as the count. This is a way to simplify the logic instead of making more mistakes in the complicated diff code. This has a drawback: the diff_queue_diff struct only lists the paths corresponding to blob changes, not their leading directories. Thus, get_or_compute_bloom_filter() needs an additional check to see if the hashmap with the leading directories becomes too large. One reason why this was not caught by test cases was that the test in t4216-log-bloom.sh that was supposed to check this "too many changes" condition only checked this on the initial commit of a repository. The old logic counted these values correctly. Update this test in a few ways: 1. Use GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS to reduce the limit, allowing smaller commits to engage with this logic. 2. Create several interesting cases of edits, adds, removes, and mode changes (in the second commit). By testing both sides of the inequality with the *_MAX_CHANGED_PATHS variable, we can see that the count is exactly correct, so none of these changes are missed or over-counted. 3. Use the trace2 data value filter_found_large to verify that these commits are on the correct side of the limit. Another way to verify the behavior is correct is through performance tests. By testing on my local copies of the Git repository and the Linux kernel repository, I could measure the effect of these short-circuits when computing a fresh commit-graph file with changed-path Bloom filters using the command GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS=N time \ git commit-graph write --reachable --changed-paths and reporting the wall time and resulting commit-graph size. For Git, the results are | | N=1 | N=10 | N=512 | |--------|----------------|----------------|----------------| | HEAD~1 | 10.90s 9.18MB | 11.11s 9.34MB | 11.31s 9.35MB | | HEAD | 9.21s 8.62MB | 11.11s 9.29MB | 11.29s 9.34MB | For Linux, the results are | | N=1 | N=20 | N=512 | |--------|----------------|---------------|---------------| | HEAD~1 | 61.28s 64.3MB | 76.9s 72.6MB | 77.6s 72.6MB | | HEAD | 49.44s 56.3MB | 68.7s 65.9MB | 69.2s 65.9MB | Naturally, the improvement becomes much less as the limit grows, as fewer commits satisfy the short-circuit. Reported-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Derrick Stolee <dstolee@microsoft.com> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-09-16 18:07:52 +00:00
test_max_changed_paths () {
grep "\"max_changed_paths\":$1" $2
}
bloom: encode out-of-bounds filters as non-empty 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>
2020-09-18 02:59:44 +00:00
test_filter_not_computed () {
grep "\"key\":\"filter-not-computed\",\"value\":\"$1\"" $2
}
bloom/diff: properly short-circuit on max_changes Commit e3696980 (diff: halt tree-diff early after max_changes, 2020-03-30) intended to create a mechanism to short-circuit a diff calculation after a certain number of paths were modified. By incrementing a "num_changes" counter throughout the recursive ll_diff_tree_paths(), this was supposed to match the number of changes that would be written into the changed-path Bloom filters. Unfortunately, this was not implemented correctly and instead misses simple cases like file modifications. This then does not stop very large changed-path filters from being written (unless they add or remove many files). To start, change the implementation in ll_diff_tree_paths() to instead use the global diff_queue_diff struct's 'nr' member as the count. This is a way to simplify the logic instead of making more mistakes in the complicated diff code. This has a drawback: the diff_queue_diff struct only lists the paths corresponding to blob changes, not their leading directories. Thus, get_or_compute_bloom_filter() needs an additional check to see if the hashmap with the leading directories becomes too large. One reason why this was not caught by test cases was that the test in t4216-log-bloom.sh that was supposed to check this "too many changes" condition only checked this on the initial commit of a repository. The old logic counted these values correctly. Update this test in a few ways: 1. Use GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS to reduce the limit, allowing smaller commits to engage with this logic. 2. Create several interesting cases of edits, adds, removes, and mode changes (in the second commit). By testing both sides of the inequality with the *_MAX_CHANGED_PATHS variable, we can see that the count is exactly correct, so none of these changes are missed or over-counted. 3. Use the trace2 data value filter_found_large to verify that these commits are on the correct side of the limit. Another way to verify the behavior is correct is through performance tests. By testing on my local copies of the Git repository and the Linux kernel repository, I could measure the effect of these short-circuits when computing a fresh commit-graph file with changed-path Bloom filters using the command GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS=N time \ git commit-graph write --reachable --changed-paths and reporting the wall time and resulting commit-graph size. For Git, the results are | | N=1 | N=10 | N=512 | |--------|----------------|----------------|----------------| | HEAD~1 | 10.90s 9.18MB | 11.11s 9.34MB | 11.31s 9.35MB | | HEAD | 9.21s 8.62MB | 11.11s 9.29MB | 11.29s 9.34MB | For Linux, the results are | | N=1 | N=20 | N=512 | |--------|----------------|---------------|---------------| | HEAD~1 | 61.28s 64.3MB | 76.9s 72.6MB | 77.6s 72.6MB | | HEAD | 49.44s 56.3MB | 68.7s 65.9MB | 69.2s 65.9MB | Naturally, the improvement becomes much less as the limit grows, as fewer commits satisfy the short-circuit. Reported-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Derrick Stolee <dstolee@microsoft.com> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-09-16 18:07:52 +00:00
test_filter_computed () {
grep "\"key\":\"filter-computed\",\"value\":\"$1\"" $2
}
bloom: encode out-of-bounds filters as non-empty 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>
2020-09-18 02:59:44 +00:00
test_filter_trunc_empty () {
grep "\"key\":\"filter-trunc-empty\",\"value\":\"$1\"" $2
}
bloom/diff: properly short-circuit on max_changes Commit e3696980 (diff: halt tree-diff early after max_changes, 2020-03-30) intended to create a mechanism to short-circuit a diff calculation after a certain number of paths were modified. By incrementing a "num_changes" counter throughout the recursive ll_diff_tree_paths(), this was supposed to match the number of changes that would be written into the changed-path Bloom filters. Unfortunately, this was not implemented correctly and instead misses simple cases like file modifications. This then does not stop very large changed-path filters from being written (unless they add or remove many files). To start, change the implementation in ll_diff_tree_paths() to instead use the global diff_queue_diff struct's 'nr' member as the count. This is a way to simplify the logic instead of making more mistakes in the complicated diff code. This has a drawback: the diff_queue_diff struct only lists the paths corresponding to blob changes, not their leading directories. Thus, get_or_compute_bloom_filter() needs an additional check to see if the hashmap with the leading directories becomes too large. One reason why this was not caught by test cases was that the test in t4216-log-bloom.sh that was supposed to check this "too many changes" condition only checked this on the initial commit of a repository. The old logic counted these values correctly. Update this test in a few ways: 1. Use GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS to reduce the limit, allowing smaller commits to engage with this logic. 2. Create several interesting cases of edits, adds, removes, and mode changes (in the second commit). By testing both sides of the inequality with the *_MAX_CHANGED_PATHS variable, we can see that the count is exactly correct, so none of these changes are missed or over-counted. 3. Use the trace2 data value filter_found_large to verify that these commits are on the correct side of the limit. Another way to verify the behavior is correct is through performance tests. By testing on my local copies of the Git repository and the Linux kernel repository, I could measure the effect of these short-circuits when computing a fresh commit-graph file with changed-path Bloom filters using the command GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS=N time \ git commit-graph write --reachable --changed-paths and reporting the wall time and resulting commit-graph size. For Git, the results are | | N=1 | N=10 | N=512 | |--------|----------------|----------------|----------------| | HEAD~1 | 10.90s 9.18MB | 11.11s 9.34MB | 11.31s 9.35MB | | HEAD | 9.21s 8.62MB | 11.11s 9.29MB | 11.29s 9.34MB | For Linux, the results are | | N=1 | N=20 | N=512 | |--------|----------------|---------------|---------------| | HEAD~1 | 61.28s 64.3MB | 76.9s 72.6MB | 77.6s 72.6MB | | HEAD | 49.44s 56.3MB | 68.7s 65.9MB | 69.2s 65.9MB | Naturally, the improvement becomes much less as the limit grows, as fewer commits satisfy the short-circuit. Reported-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Derrick Stolee <dstolee@microsoft.com> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-09-16 18:07:52 +00:00
test_filter_trunc_large () {
grep "\"key\":\"filter-trunc-large\",\"value\":\"$1\"" $2
}
test_expect_success 'correctly report changes over limit' '
bloom/diff: properly short-circuit on max_changes Commit e3696980 (diff: halt tree-diff early after max_changes, 2020-03-30) intended to create a mechanism to short-circuit a diff calculation after a certain number of paths were modified. By incrementing a "num_changes" counter throughout the recursive ll_diff_tree_paths(), this was supposed to match the number of changes that would be written into the changed-path Bloom filters. Unfortunately, this was not implemented correctly and instead misses simple cases like file modifications. This then does not stop very large changed-path filters from being written (unless they add or remove many files). To start, change the implementation in ll_diff_tree_paths() to instead use the global diff_queue_diff struct's 'nr' member as the count. This is a way to simplify the logic instead of making more mistakes in the complicated diff code. This has a drawback: the diff_queue_diff struct only lists the paths corresponding to blob changes, not their leading directories. Thus, get_or_compute_bloom_filter() needs an additional check to see if the hashmap with the leading directories becomes too large. One reason why this was not caught by test cases was that the test in t4216-log-bloom.sh that was supposed to check this "too many changes" condition only checked this on the initial commit of a repository. The old logic counted these values correctly. Update this test in a few ways: 1. Use GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS to reduce the limit, allowing smaller commits to engage with this logic. 2. Create several interesting cases of edits, adds, removes, and mode changes (in the second commit). By testing both sides of the inequality with the *_MAX_CHANGED_PATHS variable, we can see that the count is exactly correct, so none of these changes are missed or over-counted. 3. Use the trace2 data value filter_found_large to verify that these commits are on the correct side of the limit. Another way to verify the behavior is correct is through performance tests. By testing on my local copies of the Git repository and the Linux kernel repository, I could measure the effect of these short-circuits when computing a fresh commit-graph file with changed-path Bloom filters using the command GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS=N time \ git commit-graph write --reachable --changed-paths and reporting the wall time and resulting commit-graph size. For Git, the results are | | N=1 | N=10 | N=512 | |--------|----------------|----------------|----------------| | HEAD~1 | 10.90s 9.18MB | 11.11s 9.34MB | 11.31s 9.35MB | | HEAD | 9.21s 8.62MB | 11.11s 9.29MB | 11.29s 9.34MB | For Linux, the results are | | N=1 | N=20 | N=512 | |--------|----------------|---------------|---------------| | HEAD~1 | 61.28s 64.3MB | 76.9s 72.6MB | 77.6s 72.6MB | | HEAD | 49.44s 56.3MB | 68.7s 65.9MB | 69.2s 65.9MB | Naturally, the improvement becomes much less as the limit grows, as fewer commits satisfy the short-circuit. Reported-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Derrick Stolee <dstolee@microsoft.com> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-09-16 18:07:52 +00:00
git init limits &&
(
bloom/diff: properly short-circuit on max_changes Commit e3696980 (diff: halt tree-diff early after max_changes, 2020-03-30) intended to create a mechanism to short-circuit a diff calculation after a certain number of paths were modified. By incrementing a "num_changes" counter throughout the recursive ll_diff_tree_paths(), this was supposed to match the number of changes that would be written into the changed-path Bloom filters. Unfortunately, this was not implemented correctly and instead misses simple cases like file modifications. This then does not stop very large changed-path filters from being written (unless they add or remove many files). To start, change the implementation in ll_diff_tree_paths() to instead use the global diff_queue_diff struct's 'nr' member as the count. This is a way to simplify the logic instead of making more mistakes in the complicated diff code. This has a drawback: the diff_queue_diff struct only lists the paths corresponding to blob changes, not their leading directories. Thus, get_or_compute_bloom_filter() needs an additional check to see if the hashmap with the leading directories becomes too large. One reason why this was not caught by test cases was that the test in t4216-log-bloom.sh that was supposed to check this "too many changes" condition only checked this on the initial commit of a repository. The old logic counted these values correctly. Update this test in a few ways: 1. Use GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS to reduce the limit, allowing smaller commits to engage with this logic. 2. Create several interesting cases of edits, adds, removes, and mode changes (in the second commit). By testing both sides of the inequality with the *_MAX_CHANGED_PATHS variable, we can see that the count is exactly correct, so none of these changes are missed or over-counted. 3. Use the trace2 data value filter_found_large to verify that these commits are on the correct side of the limit. Another way to verify the behavior is correct is through performance tests. By testing on my local copies of the Git repository and the Linux kernel repository, I could measure the effect of these short-circuits when computing a fresh commit-graph file with changed-path Bloom filters using the command GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS=N time \ git commit-graph write --reachable --changed-paths and reporting the wall time and resulting commit-graph size. For Git, the results are | | N=1 | N=10 | N=512 | |--------|----------------|----------------|----------------| | HEAD~1 | 10.90s 9.18MB | 11.11s 9.34MB | 11.31s 9.35MB | | HEAD | 9.21s 8.62MB | 11.11s 9.29MB | 11.29s 9.34MB | For Linux, the results are | | N=1 | N=20 | N=512 | |--------|----------------|---------------|---------------| | HEAD~1 | 61.28s 64.3MB | 76.9s 72.6MB | 77.6s 72.6MB | | HEAD | 49.44s 56.3MB | 68.7s 65.9MB | 69.2s 65.9MB | Naturally, the improvement becomes much less as the limit grows, as fewer commits satisfy the short-circuit. Reported-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Derrick Stolee <dstolee@microsoft.com> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-09-16 18:07:52 +00:00
cd limits &&
mkdir d &&
mkdir d/e &&
for i in $(test_seq 1 2)
do
bloom/diff: properly short-circuit on max_changes Commit e3696980 (diff: halt tree-diff early after max_changes, 2020-03-30) intended to create a mechanism to short-circuit a diff calculation after a certain number of paths were modified. By incrementing a "num_changes" counter throughout the recursive ll_diff_tree_paths(), this was supposed to match the number of changes that would be written into the changed-path Bloom filters. Unfortunately, this was not implemented correctly and instead misses simple cases like file modifications. This then does not stop very large changed-path filters from being written (unless they add or remove many files). To start, change the implementation in ll_diff_tree_paths() to instead use the global diff_queue_diff struct's 'nr' member as the count. This is a way to simplify the logic instead of making more mistakes in the complicated diff code. This has a drawback: the diff_queue_diff struct only lists the paths corresponding to blob changes, not their leading directories. Thus, get_or_compute_bloom_filter() needs an additional check to see if the hashmap with the leading directories becomes too large. One reason why this was not caught by test cases was that the test in t4216-log-bloom.sh that was supposed to check this "too many changes" condition only checked this on the initial commit of a repository. The old logic counted these values correctly. Update this test in a few ways: 1. Use GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS to reduce the limit, allowing smaller commits to engage with this logic. 2. Create several interesting cases of edits, adds, removes, and mode changes (in the second commit). By testing both sides of the inequality with the *_MAX_CHANGED_PATHS variable, we can see that the count is exactly correct, so none of these changes are missed or over-counted. 3. Use the trace2 data value filter_found_large to verify that these commits are on the correct side of the limit. Another way to verify the behavior is correct is through performance tests. By testing on my local copies of the Git repository and the Linux kernel repository, I could measure the effect of these short-circuits when computing a fresh commit-graph file with changed-path Bloom filters using the command GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS=N time \ git commit-graph write --reachable --changed-paths and reporting the wall time and resulting commit-graph size. For Git, the results are | | N=1 | N=10 | N=512 | |--------|----------------|----------------|----------------| | HEAD~1 | 10.90s 9.18MB | 11.11s 9.34MB | 11.31s 9.35MB | | HEAD | 9.21s 8.62MB | 11.11s 9.29MB | 11.29s 9.34MB | For Linux, the results are | | N=1 | N=20 | N=512 | |--------|----------------|---------------|---------------| | HEAD~1 | 61.28s 64.3MB | 76.9s 72.6MB | 77.6s 72.6MB | | HEAD | 49.44s 56.3MB | 68.7s 65.9MB | 69.2s 65.9MB | Naturally, the improvement becomes much less as the limit grows, as fewer commits satisfy the short-circuit. Reported-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Derrick Stolee <dstolee@microsoft.com> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-09-16 18:07:52 +00:00
printf $i >d/file$i.txt &&
printf $i >d/e/file$i.txt || return 1
done &&
bloom/diff: properly short-circuit on max_changes Commit e3696980 (diff: halt tree-diff early after max_changes, 2020-03-30) intended to create a mechanism to short-circuit a diff calculation after a certain number of paths were modified. By incrementing a "num_changes" counter throughout the recursive ll_diff_tree_paths(), this was supposed to match the number of changes that would be written into the changed-path Bloom filters. Unfortunately, this was not implemented correctly and instead misses simple cases like file modifications. This then does not stop very large changed-path filters from being written (unless they add or remove many files). To start, change the implementation in ll_diff_tree_paths() to instead use the global diff_queue_diff struct's 'nr' member as the count. This is a way to simplify the logic instead of making more mistakes in the complicated diff code. This has a drawback: the diff_queue_diff struct only lists the paths corresponding to blob changes, not their leading directories. Thus, get_or_compute_bloom_filter() needs an additional check to see if the hashmap with the leading directories becomes too large. One reason why this was not caught by test cases was that the test in t4216-log-bloom.sh that was supposed to check this "too many changes" condition only checked this on the initial commit of a repository. The old logic counted these values correctly. Update this test in a few ways: 1. Use GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS to reduce the limit, allowing smaller commits to engage with this logic. 2. Create several interesting cases of edits, adds, removes, and mode changes (in the second commit). By testing both sides of the inequality with the *_MAX_CHANGED_PATHS variable, we can see that the count is exactly correct, so none of these changes are missed or over-counted. 3. Use the trace2 data value filter_found_large to verify that these commits are on the correct side of the limit. Another way to verify the behavior is correct is through performance tests. By testing on my local copies of the Git repository and the Linux kernel repository, I could measure the effect of these short-circuits when computing a fresh commit-graph file with changed-path Bloom filters using the command GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS=N time \ git commit-graph write --reachable --changed-paths and reporting the wall time and resulting commit-graph size. For Git, the results are | | N=1 | N=10 | N=512 | |--------|----------------|----------------|----------------| | HEAD~1 | 10.90s 9.18MB | 11.11s 9.34MB | 11.31s 9.35MB | | HEAD | 9.21s 8.62MB | 11.11s 9.29MB | 11.29s 9.34MB | For Linux, the results are | | N=1 | N=20 | N=512 | |--------|----------------|---------------|---------------| | HEAD~1 | 61.28s 64.3MB | 76.9s 72.6MB | 77.6s 72.6MB | | HEAD | 49.44s 56.3MB | 68.7s 65.9MB | 69.2s 65.9MB | Naturally, the improvement becomes much less as the limit grows, as fewer commits satisfy the short-circuit. Reported-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Derrick Stolee <dstolee@microsoft.com> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-09-16 18:07:52 +00:00
mkdir mode &&
printf bash >mode/script.sh &&
mkdir foo &&
touch foo/bar &&
touch foo.txt &&
git add d foo foo.txt mode &&
git commit -m "files" &&
bloom/diff: properly short-circuit on max_changes Commit e3696980 (diff: halt tree-diff early after max_changes, 2020-03-30) intended to create a mechanism to short-circuit a diff calculation after a certain number of paths were modified. By incrementing a "num_changes" counter throughout the recursive ll_diff_tree_paths(), this was supposed to match the number of changes that would be written into the changed-path Bloom filters. Unfortunately, this was not implemented correctly and instead misses simple cases like file modifications. This then does not stop very large changed-path filters from being written (unless they add or remove many files). To start, change the implementation in ll_diff_tree_paths() to instead use the global diff_queue_diff struct's 'nr' member as the count. This is a way to simplify the logic instead of making more mistakes in the complicated diff code. This has a drawback: the diff_queue_diff struct only lists the paths corresponding to blob changes, not their leading directories. Thus, get_or_compute_bloom_filter() needs an additional check to see if the hashmap with the leading directories becomes too large. One reason why this was not caught by test cases was that the test in t4216-log-bloom.sh that was supposed to check this "too many changes" condition only checked this on the initial commit of a repository. The old logic counted these values correctly. Update this test in a few ways: 1. Use GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS to reduce the limit, allowing smaller commits to engage with this logic. 2. Create several interesting cases of edits, adds, removes, and mode changes (in the second commit). By testing both sides of the inequality with the *_MAX_CHANGED_PATHS variable, we can see that the count is exactly correct, so none of these changes are missed or over-counted. 3. Use the trace2 data value filter_found_large to verify that these commits are on the correct side of the limit. Another way to verify the behavior is correct is through performance tests. By testing on my local copies of the Git repository and the Linux kernel repository, I could measure the effect of these short-circuits when computing a fresh commit-graph file with changed-path Bloom filters using the command GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS=N time \ git commit-graph write --reachable --changed-paths and reporting the wall time and resulting commit-graph size. For Git, the results are | | N=1 | N=10 | N=512 | |--------|----------------|----------------|----------------| | HEAD~1 | 10.90s 9.18MB | 11.11s 9.34MB | 11.31s 9.35MB | | HEAD | 9.21s 8.62MB | 11.11s 9.29MB | 11.29s 9.34MB | For Linux, the results are | | N=1 | N=20 | N=512 | |--------|----------------|---------------|---------------| | HEAD~1 | 61.28s 64.3MB | 76.9s 72.6MB | 77.6s 72.6MB | | HEAD | 49.44s 56.3MB | 68.7s 65.9MB | 69.2s 65.9MB | Naturally, the improvement becomes much less as the limit grows, as fewer commits satisfy the short-circuit. Reported-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Derrick Stolee <dstolee@microsoft.com> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-09-16 18:07:52 +00:00
# Commit has 7 file and 4 directory adds
GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS=10 \
GIT_TRACE2_EVENT="$(pwd)/trace" \
git commit-graph write --reachable --changed-paths &&
test_max_changed_paths 10 trace &&
test_filter_computed 1 trace &&
test_filter_trunc_large 1 trace &&
for path in $(git ls-tree -r --name-only HEAD)
do
git -c commitGraph.readChangedPaths=false log \
-- $path >expect &&
git log -- $path >actual &&
test_cmp expect actual || return 1
done &&
# Make a variety of path changes
printf new1 >d/e/file1.txt &&
printf new2 >d/file2.txt &&
rm d/e/file2.txt &&
rm -r foo &&
printf text >foo &&
mkdir f &&
printf new1 >f/file1.txt &&
# including a mode-only change (counts as modified)
git update-index --chmod=+x mode/script.sh &&
git add foo d f &&
git commit -m "complicated" &&
# start from scratch and rebuild
rm -f .git/objects/info/commit-graph &&
GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS=10 \
GIT_TRACE2_EVENT="$(pwd)/trace-edit" \
git commit-graph write --reachable --changed-paths &&
test_max_changed_paths 10 trace-edit &&
test_filter_computed 2 trace-edit &&
test_filter_trunc_large 2 trace-edit &&
for path in $(git ls-tree -r --name-only HEAD)
do
git -c commitGraph.readChangedPaths=false log \
-- $path >expect &&
git log -- $path >actual &&
test_cmp expect actual || return 1
done &&
# start from scratch and rebuild
rm -f .git/objects/info/commit-graph &&
GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS=11 \
GIT_TRACE2_EVENT="$(pwd)/trace-update" \
git commit-graph write --reachable --changed-paths &&
test_max_changed_paths 11 trace-update &&
test_filter_computed 2 trace-update &&
test_filter_trunc_large 0 trace-update &&
for path in $(git ls-tree -r --name-only HEAD)
do
bloom/diff: properly short-circuit on max_changes Commit e3696980 (diff: halt tree-diff early after max_changes, 2020-03-30) intended to create a mechanism to short-circuit a diff calculation after a certain number of paths were modified. By incrementing a "num_changes" counter throughout the recursive ll_diff_tree_paths(), this was supposed to match the number of changes that would be written into the changed-path Bloom filters. Unfortunately, this was not implemented correctly and instead misses simple cases like file modifications. This then does not stop very large changed-path filters from being written (unless they add or remove many files). To start, change the implementation in ll_diff_tree_paths() to instead use the global diff_queue_diff struct's 'nr' member as the count. This is a way to simplify the logic instead of making more mistakes in the complicated diff code. This has a drawback: the diff_queue_diff struct only lists the paths corresponding to blob changes, not their leading directories. Thus, get_or_compute_bloom_filter() needs an additional check to see if the hashmap with the leading directories becomes too large. One reason why this was not caught by test cases was that the test in t4216-log-bloom.sh that was supposed to check this "too many changes" condition only checked this on the initial commit of a repository. The old logic counted these values correctly. Update this test in a few ways: 1. Use GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS to reduce the limit, allowing smaller commits to engage with this logic. 2. Create several interesting cases of edits, adds, removes, and mode changes (in the second commit). By testing both sides of the inequality with the *_MAX_CHANGED_PATHS variable, we can see that the count is exactly correct, so none of these changes are missed or over-counted. 3. Use the trace2 data value filter_found_large to verify that these commits are on the correct side of the limit. Another way to verify the behavior is correct is through performance tests. By testing on my local copies of the Git repository and the Linux kernel repository, I could measure the effect of these short-circuits when computing a fresh commit-graph file with changed-path Bloom filters using the command GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS=N time \ git commit-graph write --reachable --changed-paths and reporting the wall time and resulting commit-graph size. For Git, the results are | | N=1 | N=10 | N=512 | |--------|----------------|----------------|----------------| | HEAD~1 | 10.90s 9.18MB | 11.11s 9.34MB | 11.31s 9.35MB | | HEAD | 9.21s 8.62MB | 11.11s 9.29MB | 11.29s 9.34MB | For Linux, the results are | | N=1 | N=20 | N=512 | |--------|----------------|---------------|---------------| | HEAD~1 | 61.28s 64.3MB | 76.9s 72.6MB | 77.6s 72.6MB | | HEAD | 49.44s 56.3MB | 68.7s 65.9MB | 69.2s 65.9MB | Naturally, the improvement becomes much less as the limit grows, as fewer commits satisfy the short-circuit. Reported-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Derrick Stolee <dstolee@microsoft.com> Signed-off-by: Taylor Blau <me@ttaylorr.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-09-16 18:07:52 +00:00
git -c commitGraph.readChangedPaths=false log \
-- $path >expect &&
git log -- $path >actual &&
test_cmp expect actual || return 1
done
)
'
bloom: encode out-of-bounds filters as non-empty 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>
2020-09-18 02:59:44 +00:00
test_expect_success 'correctly report commits with no changed paths' '
git init empty &&
test_when_finished "rm -fr empty" &&
(
cd empty &&
git commit --allow-empty -m "initial commit" &&
GIT_TRACE2_EVENT="$(pwd)/trace.event" \
git commit-graph write --reachable --changed-paths &&
test_filter_computed 1 trace.event &&
test_filter_not_computed 0 trace.event &&
test_filter_trunc_empty 1 trace.event &&
test_filter_trunc_large 0 trace.event
)
'
2020-09-18 13:27:27 +00:00
test_expect_success 'Bloom generation is limited by --max-new-filters' '
(
cd limits &&
test_commit c2 filter &&
test_commit c3 filter &&
test_commit c4 no-filter &&
rm -f trace.event &&
GIT_TRACE2_EVENT="$(pwd)/trace.event" \
git commit-graph write --reachable --split=replace \
--changed-paths --max-new-filters=2 &&
test_filter_computed 2 trace.event &&
test_filter_not_computed 3 trace.event &&
test_filter_trunc_empty 0 trace.event &&
test_filter_trunc_large 0 trace.event
)
'
test_expect_success 'Bloom generation backfills previously-skipped filters' '
# Check specifying commitGraph.maxNewFilters over "git config" works.
test_config -C limits commitGraph.maxNewFilters 1 &&
2020-09-18 13:27:27 +00:00
(
cd limits &&
rm -f trace.event &&
GIT_TRACE2_EVENT="$(pwd)/trace.event" \
git commit-graph write --reachable --changed-paths \
--split=replace &&
2020-09-18 13:27:27 +00:00
test_filter_computed 1 trace.event &&
test_filter_not_computed 4 trace.event &&
test_filter_trunc_empty 0 trace.event &&
test_filter_trunc_large 0 trace.event
)
'
test_expect_success '--max-new-filters overrides configuration' '
git init override &&
test_when_finished "rm -fr override" &&
test_config -C override commitGraph.maxNewFilters 2 &&
(
cd override &&
test_commit one &&
test_commit two &&
rm -f trace.event &&
GIT_TRACE2_EVENT="$(pwd)/trace.event" \
git commit-graph write --reachable --changed-paths \
--max-new-filters=1 &&
test_filter_computed 1 trace.event &&
test_filter_not_computed 1 trace.event &&
test_filter_trunc_empty 0 trace.event &&
test_filter_trunc_large 0 trace.event
)
'
2020-09-18 13:27:27 +00:00
test_expect_success 'Bloom generation backfills empty commits' '
git init empty &&
test_when_finished "rm -fr empty" &&
(
cd empty &&
for i in $(test_seq 1 6)
do
git commit --allow-empty -m "$i" || return 1
2020-09-18 13:27:27 +00:00
done &&
# Generate Bloom filters for empty commits 1-6, two at a time.
for i in $(test_seq 1 3)
do
rm -f trace.event &&
GIT_TRACE2_EVENT="$(pwd)/trace.event" \
git commit-graph write --reachable \
--changed-paths --max-new-filters=2 &&
test_filter_computed 2 trace.event &&
test_filter_not_computed 4 trace.event &&
test_filter_trunc_empty 2 trace.event &&
test_filter_trunc_large 0 trace.event || return 1
2020-09-18 13:27:27 +00:00
done &&
# Finally, make sure that once all commits have filters, that
# none are subsequently recomputed.
rm -f trace.event &&
GIT_TRACE2_EVENT="$(pwd)/trace.event" \
git commit-graph write --reachable \
--changed-paths --max-new-filters=2 &&
test_filter_computed 0 trace.event &&
test_filter_not_computed 6 trace.event &&
test_filter_trunc_empty 0 trace.event &&
test_filter_trunc_large 0 trace.event
)
'
commit-graph: check bounds when accessing BDAT chunk When loading Bloom filters from a commit-graph file, we use the offset values in the BIDX chunk to index into the memory mapped for the BDAT chunk. But since we don't record how big the BDAT chunk is, we just trust that the BIDX offsets won't cause us to read outside of the chunk memory. A corrupted or malicious commit-graph file will cause us to segfault (in practice this isn't a very interesting attack, since commit-graph files are local-only, and the worst case is an out-of-bounds read). We can't fix this by checking the chunk size during parsing, since the data in the BDAT chunk doesn't have a fixed size (that's why we need the BIDX in the first place). So we'll fix it in two parts: 1. Record the BDAT chunk size during parsing, and then later check that the BIDX offsets we look up are within bounds. 2. Because the offsets are relative to the end of the BDAT header, we must also make sure that the BDAT chunk is at least as large as the expected header size. Otherwise, we overflow when trying to move past the header, even for an offset of "0". We can check this early, during the parsing stage. The error messages are rather verbose, but since this is not something you'd expect to see outside of severe bugs or corruption, it makes sense to err on the side of too many details. Sadly we can't mention the filename during the chunk-parsing stage, as we haven't set g->filename at this point, nor passed it down through the stack. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-10-09 21:05:50 +00:00
corrupt_graph () {
graph=.git/objects/info/commit-graph &&
test_when_finished "rm -rf $graph" &&
git commit-graph write --reachable --changed-paths &&
corrupt_chunk_file $graph "$@"
}
check_corrupt_graph () {
corrupt_graph "$@" &&
git -c core.commitGraph=false log -- A/B/file2 >expect.out &&
git -c core.commitGraph=true log -- A/B/file2 >out 2>err &&
test_cmp expect.out out
}
test_expect_success 'Bloom reader notices too-small data chunk' '
check_corrupt_graph BDAT clear 00000000 &&
echo "warning: ignoring too-small changed-path chunk" \
"(4 < 12) in commit-graph file" >expect.err &&
test_cmp expect.err err
'
test_expect_success 'Bloom reader notices out-of-bounds filter offsets' '
check_corrupt_graph BIDX 12 FFFFFFFF &&
# use grep to avoid depending on exact chunk size
grep "warning: ignoring out-of-range offset (4294967295) for changed-path filter at pos 3 of .git/objects/info/commit-graph" err
'
test_expect_success 'Bloom reader notices too-small index chunk' '
# replace the index with a single entry, making most
# lookups out-of-bounds
check_corrupt_graph BIDX clear 00000000 &&
echo "warning: commit-graph changed-path index chunk" \
"is too small" >expect.err &&
test_cmp expect.err err
'
test_expect_success 'Bloom reader notices out-of-order index offsets' '
# we do not know any real offsets, but we can pick
# something plausible; we should not get to the point of
# actually reading from the bogus offsets anyway.
corrupt_graph BIDX 4 0000000c00000005 &&
echo "warning: ignoring decreasing changed-path index offsets" \
"(12 > 5) for positions 1 and 2 of .git/objects/info/commit-graph" >expect.err &&
git -c core.commitGraph=false log -- A/B/file2 >expect.out &&
git -c core.commitGraph=true log -- A/B/file2 >out 2>err &&
test_cmp expect.out out &&
test_cmp expect.err err
'
test_done