When running 'read-tree' with a single tree and no prefix,
'prime_cache_tree()' is called after the tree is unpacked. In that
situation, skip a redundant call to 'cache_tree_update()' in
'unpack_trees()' by enabling the 'skip_cache_tree_update' unpack option.
Removing the redundant cache tree update provides a substantial performance
improvement to 'git read-tree <tree-ish>', as shown by a test added to
'p0006-read-tree-checkout.sh':
Test before after
----------------------------------------------------------------------
read-tree br_ballast_plus_1 3.94(1.80+1.57) 3.00(1.14+1.28) -23.9%
Note that the 'read-tree' in 't1022-read-tree-partial-clone.sh' is updated
to read two trees, rather than one. The test was first introduced in
d3da223f22 (cache-tree: prefetch in partial clone read-tree, 2021-07-23) to
exercise the 'cache_tree_update()' code path, as used in 'git merge'. Since
this patch drops the call to 'cache_tree_update()' in single-tree 'git
read-tree', change the test to use the two-tree variant so that
'cache_tree_update()' is called as intended.
Signed-off-by: Victoria Dye <vdye@github.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Enable the 'skip_cache_tree_update' option in the variants that call
'prime_cache_tree()' after 'unpack_trees()' (specifically, 'git reset
--mixed' and 'git reset --hard'). This avoids redundantly rebuilding the
cache tree in both 'cache_tree_update()' at the end of 'unpack_trees()' and
in 'prime_cache_tree()', resulting in a small (but consistent) performance
improvement. From the newly-added 'p7102-reset.sh' test:
Test before after
--------------------------------------------------------------------
7102.1: reset --hard (...) 2.11(0.40+1.54) 1.97(0.38+1.47) -6.6%
Signed-off-by: Victoria Dye <vdye@github.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Add (disabled by default) option to skip the 'cache_tree_update()' at the
end of 'unpack_trees()'. In many cases, this cache tree update is redundant
because the caller of 'unpack_trees()' immediately follows it with
'prime_cache_tree()', rebuilding the entire cache tree from scratch. While
these operations aren't the most expensive part of operations like 'git
reset', the duplicate calls still create a minor unnecessary slowdown.
Introduce an option for callers to skip the 'cache_tree_update()' in
'unpack_trees()' if it is redundant (that is, if 'prime_cache_tree()' is
called afterwards). At the moment, no 'unpack_trees()' callers use the new
option; they will be updated in subsequent patches.
Signed-off-by: Victoria Dye <vdye@github.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Add a performance test comparing the execution times of 'prime_cache_tree()'
and 'cache_tree_update(_, WRITE_TREE_SILENT | WRITE_TREE_REPAIR)'. The goal
of comparing these two is to identify which is the faster method for
rebuilding an invalid cache tree, ultimately to remove one when both are
(reundantly) called in immediate succession.
Both methods are fast, so the new tests in 'p0090-cache-tree.sh' must call
each tested function multiple times to ensure the reported times (to 0.01s
resolution) convey the differences between them.
The tests compare the timing of a 'test-tool cache-tree' run as a no-op (to
capture a baseline for the overhead associated with running the tool),
'cache_tree_update()', and 'prime_cache_tree()' on four scenarios:
- A completely valid cache tree
- A cache tree with 2 invalid paths
- A cache tree with 50 invalid paths
- A completely empty cache tree
Example results:
Test this tree
-----------------------------------------------------------
0090.2: no-op, clean 1.27(0.48+0.52)
0090.3: prime_cache_tree, clean 2.02(0.83+0.85)
0090.4: cache_tree_update, clean 1.30(0.49+0.54)
0090.5: no-op, invalidate 2 1.29(0.48+0.54)
0090.6: prime_cache_tree, invalidate 2 1.98(0.81+0.83)
0090.7: cache_tree_update, invalidate 2 2.12(0.94+0.86)
0090.8: no-op, invalidate 50 1.32(0.50+0.55)
0090.9: prime_cache_tree, invalidate 50 2.10(0.86+0.89)
0090.10: cache_tree_update, invalidate 50 2.35(1.14+0.90)
0090.11: no-op, empty 1.33(0.50+0.54)
0090.12: prime_cache_tree, empty 2.04(0.84+0.87)
0090.13: cache_tree_update, empty 2.51(1.27+0.92)
These timings show that, while 'cache_tree_update()' is faster when the
cache tree is completely valid, it is equal to or slower than
'prime_cache_tree()' when there are any invalid paths. Since the redundant
calls are mostly in scenarios where the cache tree will be at least
partially invalid (e.g., 'git reset --hard'), 'prime_cache_tree()' will
likely perform better than 'cache_tree_update()' in typical cases.
Helped-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Victoria Dye <vdye@github.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Simplify the run-command API.
* rs/no-more-run-command-v:
replace and remove run_command_v_opt()
replace and remove run_command_v_opt_cd_env_tr2()
replace and remove run_command_v_opt_tr2()
replace and remove run_command_v_opt_cd_env()
use child_process members "args" and "env" directly
use child_process member "args" instead of string array variable
sequencer: simplify building argument list in do_exec()
bisect--helper: factor out do_bisect_run()
bisect: simplify building "checkout" argument list
am: simplify building "show" argument list
run-command: fix return value comment
merge: remove always-the-same "verbose" arguments
"git archive" mistakenly complained twice about a missing executable,
which has been corrected.
* rs/archive-filter-error-once:
archive-tar: report filter start error only once
A redundant diagnostic message is dropped from test_path_is_missing().
* ma/drop-redundant-diagnostic:
test-lib-functions: drop redundant diagnostic print
Various tests exercising the transfer.credentialsInUrl configuration
are taught to avoid making requests which require resolving localhost
to reduce CI-flakiness.
* jk/ref-filter-parsing-bugs:
ref-filter: fix parsing of signatures with CRLF and no body
ref-filter: fix parsing of signatures without blank lines
The glossary entries for "commit-graph file" and "reachability
bitmap" have been added.
* po/glossary-around-traversal:
glossary: add reachability bitmap description
glossary: add "commit graph" description
doc: use 'object database' not ODB or abbreviation
doc: use "commit-graph" hyphenation consistently
The adjust_shared_perm() helper function learned to refrain from
setting the "g+s" bit on directories when it is not necessary.
* jc/set-gid-bit-less-aggressively:
adjust_shared_perm(): leave g+s alone when the group does not matter
Enable gc.cruftpacks by default for those who opt into
feature.experimental setting.
* es/mark-gc-cruft-as-experimental:
config: let feature.experimental imply gc.cruftPacks=true
gc: add tests for --cruft and friends
Various tests exercising the transfer.credentialsInUrl configuration
are taught to avoid making requests which require resolving localhost
to reduce CI-flakiness.
* jk/avoid-localhost:
t5516/t5601: be less strict about the number of credential warnings
t5516: move plaintext-password tests from t5601 and t5516
This commit fixes a bug when parsing tags that have CRLF line endings, a
signature, and no body, like this (the "^M" are marking the CRs):
this is the subject^M
-----BEGIN PGP SIGNATURE-----^M
^M
...some stuff...^M
-----END PGP SIGNATURE-----^M
When trying to find the start of the body, we look for a blank line
separating the subject and body. In this case, there isn't one. But we
search for it using strstr(), which will find the blank line in the
signature.
In the non-CRLF code path, we check whether the line we found is past
the start of the signature, and if so, put the body pointer at the start
of the signature (effectively making the body empty). But the CRLF code
path doesn't catch the same case, and we end up with the body pointer in
the middle of the signature field. This has two visible problems:
- printing %(contents:subject) will show part of the signature, too,
since the subject length is computed as (body - subject)
- the length of the body is (sig - body), which makes it negative.
Asking for %(contents:body) causes us to cast this to a very large
size_t when we feed it to xmemdupz(), which then complains about
trying to allocate too much memory.
These are essentially the same bugs fixed in the previous commit, except
that they happen when there is a CRLF blank line in the signature,
rather than no blank line at all. Both are caused by the refactoring in
9f75ce3d8f (ref-filter: handle CRLF at end-of-line more gracefully,
2020-10-29).
We can fix this by doing the same "sigstart" check that we do in the
non-CRLF case. And rather than repeat ourselves, we can just use
short-circuiting OR to collapse both cases into a single conditional.
I.e., rather than:
if (strstr("\n\n"))
...found blank, check if it's in signature...
else if (strstr("\r\n\r\n"))
...found blank, check if it's in signature...
else
...no blank line found...
we can collapse this to:
if (strstr("\n\n")) ||
strstr("\r\n\r\n")))
...found blank, check if it's in signature...
else
...no blank line found...
The tests show the problem and the fix. Though it wasn't broken, I
included contents:signature here to make sure it still behaves as
expected, but note the shell hackery needed to make it work. A
less-clever option would be to skip using test_atom and just "append_cr
>expected" ourselves.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
When ref-filter is asked to show %(content:subject), etc, we end up in
find_subpos() to parse out the three major parts: the subject, the body,
and the signature (if any).
When searching for the blank line between the subject and body, if we
don't find anything, we try to treat the whole message as the subject,
with no body. But our idea of "the whole message" needs to take into
account the signature, too. Since 9f75ce3d8f (ref-filter: handle CRLF at
end-of-line more gracefully, 2020-10-29), the code instead goes all the
way to the end of the buffer, which produces confusing output.
Here's an example. If we have a tag message like this:
this is the subject
-----BEGIN SSH SIGNATURE-----
...some stuff...
-----END SSH SIGNATURE-----
then the current parser will put the start of the body at the end of the
whole buffer. This produces two buggy outcomes:
- since the subject length is computed as (body - subject), showing
%(contents:subject) will print both the subject and the signature,
rather than just the single line
- since the body length is computed as (sig - body), and the body now
starts _after_ the signature, we end up with a negative length!
Fortunately we never access out-of-bounds memory, because the
negative length is fed to xmemdupz(), which casts it to a size_t,
and xmalloc() bails trying to allocate an absurdly large value.
In theory it would be possible for somebody making a malicious tag
to wrap it around to a more reasonable value, but it would require a
tag on the order of 2^63 bytes. And even if they did, all they get
is an out of bounds string read. So the security implications are
probably not interesting.
We can fix both by correctly putting the start of the body at the same
index as the start of the signature (effectively making the body empty).
Note that this is a real issue with signatures generated with gpg.format
set to "ssh", which would look like the example above. In the new tests
here I use a hard-coded tag message, for a few reasons:
- regardless of what the ssh-signing code produces now or in the
future, we should be testing this particular case
- skipping the actual signature makes the tests simpler to write (and
allows them to run on more systems)
- t6300 has helpers for working with gpg signatures; for the purposes
of this bug, "BEGIN PGP" is just as good a demonstration, and this
simplifies the tests
Curiously, the same issue doesn't happen with real gpg signatures (and
there are even existing tests in t6300 with cover this). Those have a
blank line between the header and the content, like:
this is the subject
-----BEGIN PGP SIGNATURE-----
...some stuff...
-----END PGP SIGNATURE-----
Because we search for the subject/body separator line with a strstr(),
we find the blank line in the signature, even though it's outside of
what we'd consider the body. But that puts us unto a separate code path,
which realizes that we're now in the signature and adjusts the line back
to "sigstart". So this patch is basically just making the "no line found
at all" case match that. And note that "sigstart" is always defined (if
there is no signature, it points to the end of the buffer as you'd
expect).
Reported-by: Martin Englund <martin@englund.nu>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
It is unclear as to _why_, but under certain circumstances the warning
about credentials being passed as part of the URL seems to be swallowed
by the `git remote-https` helper in the Windows jobs of Git's CI builds.
Since it is not actually important how many times Git prints the
warning/error message, as long as it prints it at least once, let's just
make the test a bit more lenient and test for the latter instead of the
former, which works around these CI issues.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Commit 6dcbdc0d66 (remote: create fetch.credentialsInUrl config,
2022-06-06) added tests for our handling of passwords in URLs. Since the
obvious URL to be affected is git-over-http, the tests use http. However
they don't set up a test server; they just try to access
https://localhost, assuming it will fail (because the nothing is
listening there).
This causes some possible problems:
- There might be a web server running on localhost, and we do not
actually want to connect to that.
- The DNS resolver, or the local firewall, might take a substantial
amount of time (or forever, whichever comes first) to fail to
connect, slowing down the tests cases unnecessarily.
- Since there's no server, our tests for "allow" and "warn" still
expect the clone/fetch/push operations to fail, even though in the
real world we'd expect these to succeed. We scrape stderr to see
what happened, but it's not as robust as a more realistic test.
Let's instead move these to t5551, which is all about testing http and
where we have a real server. That eliminates any issues with contacting
a strange URL, and lets the "allow" and "warn" tests confirm that the
operation actually succeeds.
It's not quite a verbatim move for a few reasons:
- we can drop the LIBCURL dependency; it's already part of
lib-httpd.sh
- we'll use HTTPD_URL_USER_PASS, etc, instead of our fake URL. To
avoid repetition, we'll add a few extra variables.
- the "https://username:@localhost" test uses a funny URL that
lib-httpd.sh doesn't provide. We'll similarly construct it in a
variable. Note that we're hard-coding the lib-httpd username here,
but t5551 already does that everywhere.
- for the "domain:port" test, the URL provided by lib-httpd is fine,
since our test server will always be on an exotic port. But we'll
confirm in the test that this is so.
- since our message-matching is done via grep, I simplified it to use
a regex, rather than trying to massage lib-httpd's variables.
Arguably this makes it more readable, too, while retaining the bits
we care about: the fatal/warning distinction, the "uses plaintext"
message, and the fact that the password was redacted.
- we'll use the /auth/ path for the repo, which shows that we are
indeed making use of the auth information when needed.
- we'll also use /smart/; most of these tests could be done via /dumb/
in t5550, but setting up pushes there requires extra effort and
dependencies. The smart protocol is what most everyone is using
these days anyway.
This patch is my own, but I stole the analysis and a few bits of the
commit message from a patch by Johannes Schindelin.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
`test_path_is_missing` was introduced back in 2caf20c52b ("test-lib:
user-friendly alternatives to test [-d|-f|-e]", 2010-08-10). It took the
path that was supposed to be missing, as well as an optional "diagnosis"
that would be echoed if the path was found to be alive.
Commit 45a2686441 ("test-lib-functions: remove bug-inducing
"diagnostics" helper param", 2021-02-12) dropped this diagnostic
functionality from several `test_path_is_foo` helpers, but note how it
tweaked the README entry on `test_path_is_missing` without actually
adjusting its implementation.
Commit e7884b353b ("test-lib-functions: assert correct parameter count",
2021-02-12) then followed up by asserting that we get just a single
argument.
This history leaves us in a state where we assert that we have exactly
one argument, then go on to anyway check for arguments, echoing them
all. It's clear that we can simplify this code. We should also note that
we run `ls -ld "$1"`, so printing the filename a second time doesn't
really buy us anything. Thus, we can drop the whole `if` block as
redundant.
Signed-off-by: Martin Ågren <martin.agren@gmail.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Rewrite a deep recursion in the skipping negotiator to use a loop
with on-heap prio queue to avoid stack wastage.
* jt/skipping-negotiator-wo-recursion:
negotiator/skipping: avoid stack overflow
"git merge-tree --stdin" is a new way to request a series of merges
and report the merge results.
* en/merge-tree-sequence:
merge-tree: support multiple batched merges with --stdin
merge-tree: update documentation for differences in -z output
Define the logical elements of a "bundle list", data structure to
store them in-core, format to transfer them, and code to parse
them.
* ds/bundle-uri-3:
bundle-uri: suppress stderr from remote-https
bundle-uri: quiet failed unbundlings
bundle: add flags to verify_bundle()
bundle-uri: fetch a list of bundles
bundle: properly clear all revision flags
bundle-uri: limit recursion depth for bundle lists
bundle-uri: parse bundle list in config format
bundle-uri: unit test "key=value" parsing
bundle-uri: create "key=value" line parsing
bundle-uri: create base key-value pair parsing
bundle-uri: create bundle_list struct and helpers
bundle-uri: use plain string in find_temp_filename()
"git branch --edit-description" can exit with status -1 which is
not a good practice; it learned to use 1 as everybody else instead.
* rj/branch-do-not-exit-with-minus-one-status:
branch: error code with --edit-description
The role the security mailing list plays in an embargoed release
has been documented.
* jr/embargoed-releases-doc:
embargoed releases: also describe the git-security list and the process
Merging a branch with directory renames into a branch that changes
the directory to a symlink was mishandled by the ort merge
strategy, which has been corrected.
* en/ort-dir-rename-and-symlink-fix:
merge-ort: fix bug with dir rename vs change dir to symlink
A bugfix to "git subtree" in its split and merge features.
* pb/subtree-split-and-merge-after-squashing-tag-fix:
subtree: fix split after annotated tag was squashed merged
subtree: fix squash merging after annotated tag was squashed merged
subtree: process 'git-subtree-split' trailer in separate function
subtree: use named variables instead of "$@" in cmd_pull
subtree: define a variable before its first use in 'find_latest_squash'
subtree: prefix die messages with 'fatal'
subtree: add 'die_incompatible_opt' function to reduce duplication
subtree: use 'git rev-parse --verify [--quiet]' for better error messages
test-lib-functions: mark 'test_commit' variables as 'local'
Fix some bugs in the reflog messages when rebasing and changes the
reflog messages of "rebase --apply" to match "rebase --merge" with
the aim of making the reflog easier to parse.
* pw/rebase-reflog-fixes:
rebase: cleanup action handling
rebase --abort: improve reflog message
rebase --apply: make reflog messages match rebase --merge
rebase --apply: respect GIT_REFLOG_ACTION
rebase --merge: fix reflog message after skipping
rebase --merge: fix reflog when continuing
t3406: rework rebase reflog tests
rebase --apply: remove duplicated code
"git rebase --keep-base" used to discard the commits that are
already cherry-picked to the upstream, even when "keep-base" meant
that the base, on top of which the history is being rebuilt, does
not yet include these cherry-picked commits. The --keep-base
option now implies --reapply-cherry-picks and --no-fork-point
options.
* pw/rebase-keep-base-fixes:
rebase --keep-base: imply --no-fork-point
rebase --keep-base: imply --reapply-cherry-picks
rebase: factor out branch_base calculation
rebase: rename merge_base to branch_base
rebase: store orig_head as a commit
rebase: be stricter when reading state files containing oids
t3416: set $EDITOR in subshell
t3416: tighten two tests
Two new facilities, "timer" and "counter", are introduced to the
trace2 API.
* jh/trace2-timers-and-counters:
trace2: add global counter mechanism
trace2: add stopwatch timers
trace2: convert ctx.thread_name from strbuf to pointer
trace2: improve thread-name documentation in the thread-context
trace2: rename the thread_name argument to trace2_thread_start
api-trace2.txt: elminate section describing the public trace2 API
tr2tls: clarify TLS terminology
trace2: use size_t alloc,nr_open_regions in tr2tls_thread_ctx
"git shortlog" learned to group by the "format" string.
* tb/shortlog-group:
shortlog: implement `--group=committer` in terms of `--group=<format>`
shortlog: implement `--group=author` in terms of `--group=<format>`
shortlog: extract `shortlog_finish_setup()`
shortlog: support arbitrary commit format `--group`s
shortlog: extract `--group` fragment for translation
shortlog: make trailer insertion a noop when appropriate
shortlog: accept `--date`-related options
Code simplification by using strvec_pushf() instead of building an
argument in a separate strbuf.
* rs/absorb-git-dir-simplify:
submodule: use strvec_pushf() for --super-prefix
The way "git repack" creared temporary files when it received a
signal was prone to deadlocking, which has been corrected.
* jk/repack-tempfile-cleanup:
t7700: annotate cruft-pack failure with ok=sigpipe
repack: drop remove_temporary_files()
repack: use tempfiles for signal cleanup
repack: expand error message for missing pack files
repack: populate extension bits incrementally
repack: convert "names" util bitfield to array
Make sure generated dependency file is stably sorted to help
developers debugging their build issues.
* sg/stable-docdep:
Documentation/build-docdep.perl: generate sorted output
A new "--include-whitespace" option is added to "git patch-id", and
existing bugs in the internal patch-id logic that did not match
what "git patch-id" produces have been corrected.
* jz/patch-id:
builtin: patch-id: remove unused diff-tree prefix
builtin: patch-id: add --verbatim as a command mode
patch-id: fix patch-id for mode changes
builtin: patch-id: fix patch-id with binary diffs
patch-id: use stable patch-id for rebases
patch-id: fix stable patch id for binary / header-only
Git has an additional "commit graph" capability that supplements the
normal commit object's directed acyclic graph (DAG). The supplemental
commit graph file is designed for speed of access.
Describe the commit graph both from the normative DAG view point and
from the commit graph file perspective.
Also, clarify the link between the branch ref and branch tip
by linking to the `ref` glossary entry, matching this commit graph
entry.
The commit-graph file is also distinguished by its hyphenation.
Subsequent commit catches the few cases where the hyphenation of
commit-graph was missing.
Signed-off-by: Philip Oakley <philipoakley@iee.email>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
The abbreviation 'ODB' is used in the technical documentation
sections for commit-graph and parallel-checkout, along with an
'odb' option in `git-pack-redundant`, without expansion.
Use 'object database' in full, in those entries. The text has not
been reflowed to keep the changes minimal.
While in the glossary for `object` terms, add the common`oid`
abbreviation to its entry.
Signed-off-by: Philip Oakley <philipoakley@iee.email>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Note, historical release notes have not been updated.
Signed-off-by: Philip Oakley <philipoakley@iee.email>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
A missing tar filter is reported by start_command() using error(), but
also by its caller, write_tar_filter_archive(), using die():
$ git -c tar.invalid.command=foo archive --format=invalid HEAD
error: cannot run foo: No such file or directory
fatal: unable to start 'foo' filter: No such file or directory
The second message contains all relevant information and even says that
the failed command was intended to be used as a filter. Silence the
first one because it's redundant.
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Replace the remaining calls of run_command_v_opt() with run_command()
calls and explict struct child_process variables. This is more verbose,
but not by much overall. The code becomes more flexible, e.g. it's easy
to extend to conditionally add a new argument.
Then remove the now unused function and its own flag names, simplifying
the run-command API.
Suggested-by: Jeff King <peff@peff.net>
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
The convenience function run_command_v_opt_cd_env_tr2() has no external
callers left. Inline it and remove it from the API.
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
The convenience function run_command_v_opt_tr2() is only used by a
single caller. Use struct child_process and run_command() directly
instead and remove the underused function.
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
