Apply "index-compatibility.pending.cocci" rule to "builtin/*", but
exclude those where we conflict with in-flight changes.
As a result some of them end up using only "the_index", so let's have
them use the more narrow "USE_THE_INDEX_VARIABLE" rather than
"USE_THE_INDEX_COMPATIBILITY_MACROS".
Manual changes not made by coccinelle, that were squashed in:
* Whitespace-wrap argument lists for repo_hold_locked_index(),
repo_read_index_preload() and repo_refresh_and_write_index(), in cases
where the line became too long after the transformation.
* Change "refresh_cache()" to "refresh_index()" in a comment in
"builtin/update-index.c".
* For those whose call was followed by perror("<macro-name>"), change
it to perror("<function-name>"), referring to the new function.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In a preceding commit we fully applied the
"index-compatibility.pending.cocci" rule to "t/helper/*".
Let's now stop defining "USE_THE_INDEX_COMPATIBILITY_MACROS" in
test-tool.h itself, and instead instead define
"USE_THE_INDEX_VARIABLE" in the individual test helpers that need
it. This mirrors how we do the same thing in the "builtin/" directory.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Split up the "USE_THE_INDEX_COMPATIBILITY_MACROS" into that setting
and a more narrow "USE_THE_INDEX_VARIABLE". In the case of these
built-ins we only need "the_index" variable, but not the compatibility
wrapper for functions we're not using.
Let's then have some users of "USE_THE_INDEX_COMPATIBILITY_MACROS" use
this more narrow and descriptive define.
For context: The USE_THE_INDEX_COMPATIBILITY_MACROS macro was added to
test-tool.h in f8adbec9fe (cache.h: flip
NO_THE_INDEX_COMPATIBILITY_MACROS switch, 2019-01-24).
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Apply the "index-compatibility.pending.cocci" rule to the "t/helper/*"
directory, a subsequent commit will extend cache.h to further narrow
down the use of "USE_THE_INDEX_COMPATIBILITY_MACROS" in this area.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Mostly apply the part of "index-compatibility.pending.cocci" that
renames the global variables like "active_nr", which are a shorthand
to referencing (in that case) a struct member as "the_index.cache_nr".
In doing so move more of "index-compatibility.pending.cocci" to
"index-compatibility.cocci".
In the case of "active_nr" we'd have a textual conflict with
"ab/various-leak-fixes" in "next"[1]. Let's exclude that specific case
while moving the rule over from "pending".
1. 407b94280f8 (commit: discard partial cache before (re-)reading it,
2022-11-08)
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Apply a selection of rules in "index-compatibility.pending.cocci"
tree-wide, and in doing so migrate them to
"index-compatibility.cocci".
As in preceding commits the only manual changes here are the macro
removals in "cache.h", and the update to the '*.cocci" rules. The rest
of the C code changes are the result of applying those updated rules.
Move rules for some rarely used cache compatibility macros from
"index-compatibility.pending.cocci" to "index-compatibility.cocci" and
apply them.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add a coccinelle rule which covers the rest of the macros guarded by
"USE_THE_INDEX_COMPATIBILITY_MACROS" cache.h. If the result of this
were applied it can be reduced down to just:
#ifdef USE_THE_INDEX_COMPATIBILITY_MACROS
extern struct index_state the_index;
#endif
But that patch is just under 2000 lines, so let's first add this as a
"pending", and then incrementally pick changes from it in subsequent
commits. In doing that we'll migrate rules from this
"index-compatibility.pending.cocci" to the "index-compatibility.cocci"
created in a preceding commit.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The discard_index() function has not returned non-zero since
7a51ed66f6 (Make on-disk index representation separate from in-core
one, 2008-01-14), but we've had various code in-tree still acting as
though that might be the case.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since 4aab5b46f4 (Make read-cache.c "the_index" free., 2007-04-01)
we've been undergoing a slow migration away from these macros, but
haven't made much progress since f8adbec9fe (cache.h: flip
NO_THE_INDEX_COMPATIBILITY_MACROS switch, 2019-01-24).
Let's move forward a bit by changing the users of those macros that
are rare enough that we can convert them in one go, and then remove
the compatibility shim.
The only manual change to the C code here is to "cache.h", the rest is
all the result of applying the new "index-compatibility.cocci".
Even though it's a one-off, let's keep the coccinelle rules for
now. We'll extend them in subsequent commits, and this will help
anything that's in-flight or out-of-tree to migrate.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Adding "USE_THE_INDEX_COMPATIBILITY_MACROS" to these two appears to
have been unnecessary from the start, as going back and compiling
f8adbec9fe (cache.h: flip NO_THE_INDEX_COMPATIBILITY_MACROS switch,
2019-01-24) without that addition works.
Let's not have these ask for the compatibility macros from cache.h
that they don't need.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The "active_alloc" macro added in 228e94f935 (Move index-related
variables into a structure., 2007-04-01) has not been used since
4aab5b46f4 (Make read-cache.c "the_index" free., 2007-04-01). Let's
remove it.
The rest of these are likewise unused, so let's not keep them
around. E.g. 12cd0bf9b0 (dir: stop using the index compatibility
macros, 2017-05-05) is the last use of "cache_dir_exists".
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Avoid calling 'cache_tree_update()' when doing so would be redundant.
* vd/skip-cache-tree-update:
rebase: use 'skip_cache_tree_update' option
read-tree: use 'skip_cache_tree_update' option
reset: use 'skip_cache_tree_update' option
unpack-trees: add 'skip_cache_tree_update' option
cache-tree: add perf test comparing update and prime
Update the credential-cache documentation to provide a more realistic
example.
* mh/increase-credential-cache-timeout:
Documentation: increase example cache timeout to 1 hour
`git rebase --update-refs` would delete references when all `update-ref`
commands in the sequencer were removed, which has been corrected.
* vd/update-refs-delete:
rebase --update-refs: avoid unintended ref deletion
"git repack" learns to send cruft objects out of the way into
packfiles outside the repository.
* tb/repack-expire-to:
builtin/repack.c: implement `--expire-to` for storing pruned objects
builtin/repack.c: write cruft packs to arbitrary locations
builtin/repack.c: pass "cruft_expiration" to `write_cruft_pack`
builtin/repack.c: pass "out" to `prepare_pack_objects`
Makefile comments updates and reordering to clarify knobs used to
choose SHA implementations.
* ab/sha-makefile-doc:
Makefile: discuss SHAttered in *_SHA{1,256} discussion
Makefile: document default SHA-1 backend on OSX
Makefile & test-tool: replace "DC_SHA1" variable with a "define"
Makefile: document SHA-1 and SHA-256 default and selection order
Makefile: document default SHA-256 backend
Makefile: rephrase the discussion of *_SHA1 knobs
Makefile: create and use sections for "define" flag listing
Makefile: correct DC_SHA1 documentation
INSTALL: remove discussion of SHA-1 backends
Makefile: always (re)set DC_SHA1 on fallback
Various test updates.
* ab/misc-hook-submodule-run-command:
run-command tests: test stdout of run_command_parallel()
submodule tests: reset "trace.out" between "grep" invocations
hook tests: fix redirection logic error in 96e7225b31
On my use case involving 771 islands of Linux on kernel.org,
this reduces memory usage by around 25MB. The bulk of that
comes from free_remote_islands, since free_config_regexes only
saves around 40k.
This memory is saved early in the memory-intensive pack process,
making it available for the remainder of the long process.
Signed-off-by: Eric Wong <e@80x24.org>
Co-authored-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
In parse_object(), we try to handle blobs by streaming rather than
loading them entirely into memory. The most common case here will be
that we haven't seen the object yet and check oid_object_info(), which
tells us we have a blob.
But we trigger this code on one other case: when we have an in-memory
object struct with type OBJ_BLOB (and without its "parsed" flag set,
since otherwise we'd return early from the function). This indicates
that some other part of the code suspected we have a blob (e.g., it was
mentioned by a tree or tag) but we haven't yet looked at the on-disk
copy.
In this case before hitting the streaming path, we check if we have the
object on-disk at all. This is mostly pointless extra work, as the
streaming path would complain if it couldn't open the object (albeit
with the message "hash mismatch", which is a little misleading).
But it's also insufficient to catch all problems. The streaming code
will only tell us "yes, the on-disk object matches the oid". But it
doesn't actually confirm that what we found was indeed a blob, and
neither does repo_has_object_file().
One way to improve this would be to teach stream_object_signature() to
check the type (either by returning it to us to check, or taking an
"expected" type). But there's an even simpler fix here: if we suspect
the object is a blob, just call oid_object_info() to confirm that we
have it on-disk, and that it really is a blob.
This is slightly less efficient than teaching stream_object_signature()
to do it (since it has to open the object already). But this case very
rarely comes up. In practice, we usually don't have any clue what the
type is, in which case we already call oid_object_info(). This
"suspected" case happens only when some other code created an object
struct but didn't actually parse the blob, which is actually tricky to
trigger at all (see the discussion of the test below).
I reworked the conditional a bit so that instead of:
if ((suspected_blob && oid_object_info() == OBJ_BLOB)
(no_clue && oid_object_info() == OBJ_BLOB)
we have the simpler:
if ((suspected_blob || no_clue) && oid_object_info() == OBJ_BLOB)
This is shorter, but also reflects what we really want say, which is
"have we ruled out this being a blob; if not, check it on-disk".
In either case, if oid_object_info() fails to tell us it's a blob, we'll
skip the streaming code path and call repo_read_object_file(), just as
before. And if we really do have a mismatch with the existing object
struct, we'll eventually call lookup_commit(), etc, via
parse_object_buffer(), which will complain that it doesn't match our
existing obj->type.
So this fixes one of the lingering expect_failure cases from 0616617c7e
(t: introduce tests for unexpected object types, 2019-04-09). That test
works by peeling a tag that claims to point to a blob (triggering us to
create the struct), but really points to something else, which we later
discover when we call parse_object() as part of the actual traversal).
Prior to this commit, we'd quietly check the sha1 and mark the blob as
"parsed". Now we correctly complain about the mismatch.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
When parsing an object of unknown type, we check to see if it's a blob,
so we can use our streaming code path. This uses oid_object_info() to
check the type, but before doing so we call repo_has_object_file(). This
latter is pointless, as oid_object_info() will already fail if the
object is missing. Checking it ahead of time just complicates the code
and is a waste of resources (albeit small).
Let's drop the redundant check.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
When serving a push, git-receive-pack(1) needs to verify that the
packfile sent by the client contains all objects that are required by
the updated references. This connectivity check works by marking all
preexisting references as uninteresting and using the new reference tips
as starting point for a graph walk.
Marking all preexisting references as uninteresting can be a problem
when it comes to performance. Git forges tend to do internal bookkeeping
to keep alive sets of objects for internal use or make them easy to find
via certain references. These references are typically hidden away from
the user so that they are neither advertised nor writeable. At GitLab,
we have one particular repository that contains a total of 7 million
references, of which 6.8 million are indeed internal references. With
the current connectivity check we are forced to load all these
references in order to mark them as uninteresting, and this alone takes
around 15 seconds to compute.
We can optimize this by only taking into account the set of visible refs
when marking objects as uninteresting. This means that we may now walk
more objects until we hit any object that is marked as uninteresting.
But it is rather unlikely that clients send objects that make large
parts of objects reachable that have previously only ever been hidden,
whereas the common case is to push incremental changes that build on top
of the visible object graph.
This provides a huge boost to performance in the mentioned repository,
where the vast majority of its refs hidden. Pushing a new commit into
this repo with `transfer.hideRefs` set up to hide 6.8 million of 7 refs
as it is configured in Gitaly leads to a 4.5-fold speedup:
Benchmark 1: main
Time (mean ± σ): 30.977 s ± 0.157 s [User: 30.226 s, System: 1.083 s]
Range (min … max): 30.796 s … 31.071 s 3 runs
Benchmark 2: pks-connectivity-check-hide-refs
Time (mean ± σ): 6.799 s ± 0.063 s [User: 6.803 s, System: 0.354 s]
Range (min … max): 6.729 s … 6.850 s 3 runs
Summary
'pks-connectivity-check-hide-refs' ran
4.56 ± 0.05 times faster than 'main'
As we mostly go through the same codepaths even in the case where there
are no hidden refs at all compared to the code before there is no change
in performance when no refs are hidden:
Benchmark 1: main
Time (mean ± σ): 48.188 s ± 0.432 s [User: 49.326 s, System: 5.009 s]
Range (min … max): 47.706 s … 48.539 s 3 runs
Benchmark 2: pks-connectivity-check-hide-refs
Time (mean ± σ): 48.027 s ± 0.500 s [User: 48.934 s, System: 5.025 s]
Range (min … max): 47.504 s … 48.500 s 3 runs
Summary
'pks-connectivity-check-hide-refs' ran
1.00 ± 0.01 times faster than 'main'
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Add a new `--exclude-hidden=` option that is similar to the one we just
added to git-rev-list(1). Given a section name `uploadpack` or `receive`
as argument, it causes us to exclude all references that would be hidden
by the respective `$section.hideRefs` configuration.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Users can optionally hide refs from remote users in git-upload-pack(1),
git-receive-pack(1) and others via the `transfer.hideRefs`, but there is
not an easy way to obtain the list of all visible or hidden refs right
now. We'll require just that though for a performance improvement in our
connectivity check.
Add a new option `--exclude-hidden=` that excludes any hidden refs from
the next pseudo-ref like `--all` or `--branches`.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
The functions that handle exclusion of refs work on a single string
list. We're about to add a second mechanism for excluding refs though,
and it makes sense to reuse much of the same architecture for both kinds
of exclusion.
Introduce a new `struct ref_exclusions` that encapsulates all the logic
related to excluding refs and move the `struct string_list` that holds
all wildmatch patterns of excluded refs into it. Rename functions that
operate on this struct to match its name.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Move together the definitions of functions that handle exclusions of
refs so that related functionality sits in a single place, only.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
We're about to add a new argument to git-rev-list(1) that allows it to
add all references that are visible when taking `transfer.hideRefs` et
al into account. This will require us to potentially parse multiple sets
of hidden refs, which is not easily possible right now as there is only
a single, global instance of the list of parsed hidden refs.
Refactor `parse_hide_refs_config()` and `ref_is_hidden()` so that both
take the list of hidden references as input and adjust callers to keep a
local list, instead. This allows us to easily use multiple hidden-ref
lists. Furthermore, it allows us to properly free this list before we
exit.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
When parsing the hideRefs configuration, we first duplicate the config
value so that we can modify it. We then subsequently append it to the
`hide_refs` string list, which is initialized with `strdup_strings`
enabled. As a consequence we again reallocate the string, but never
free the first duplicate and thus have a memory leak.
While we never clean up the static `hide_refs` variable anyway, this is
no excuse to make the leak worse by leaking every value twice. We are
also about to change the way this variable will be handled so that we do
indeed start to clean it up. So let's fix the memory leak by using the
`string_list_append_nodup()` so that we pass ownership of the allocated
string to `hide_refs`.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
When `git notes` prepares the template it adds an empty newline between
the comment header and the content:
>
> #
> # Write/edit the notes for the following object:
>
> # commit 0f3c55d4c2b7864bffb2d92278eff08d0b2e083f
> # etc
This is wrong structurally because that newline is part of the comment,
too, and thus should be commented. Also, it throws off some positioning
strategies of editors and plugins, and it differs from how we do commit
templates.
Change this to follow the standard set by `git commit`:
>
> #
> # Write/edit the notes for the following object:
> #
> # commit 0f3c55d4c2b7864bffb2d92278eff08d0b2e083f
>
Tests pass unchanged after this code change.
Signed-off-by: Michael J Gruber <git@grubix.eu>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
On MinGW the "/dev/null" is translated to "nul" on command-lines, even
though as in this case it'll never end up referring to an actual file.
So on Windows the fix for the previous "example.com" timeout issue in
8354cf752e (t7610: fix flaky timeout issue, don't clone from
example.com, 2022-11-05) would yield:
fatal: repo URL: 'nul' must be absolute or begin with ./|../
Let's evade this yet again by prefixing this with "file://", which
makes this pass in the Windows CI.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
While trying to fix a move based on an uninitialized value (along with a
declaration after the first statement), be0fd57228
(maintenance --unregister: fix uninit'd data use &
-Wdeclaration-after-statement, 2022-11-15) unintentionally introduced a
use-after-free.
The problem arises when `maintenance_unregister()` sees a non-NULL
`config_file` string and thus tries to call
git_configset_get_value_multi() to lookup the corresponding values.
We store the result off, and then call git_configset_clear(), which
frees the pointer that we just stored. We then try to read that
now-freed pointer a few lines below, and there we have our
use-after-free:
$ ./t7900-maintenance.sh -vxi --run=23 --valgrind
[...]
+ git maintenance unregister --config-file ./other
==3048727== Invalid read of size 8
==3048727== at 0x1869CA: maintenance_unregister (gc.c:1590)
==3048727== by 0x188F42: cmd_maintenance (gc.c:2651)
==3048727== by 0x128C62: run_builtin (git.c:466)
==3048727== by 0x12907E: handle_builtin (git.c:721)
==3048727== by 0x1292EC: run_argv (git.c:788)
==3048727== by 0x12988E: cmd_main (git.c:926)
==3048727== by 0x21ED39: main (common-main.c:57)
==3048727== Address 0x4b38bc8 is 24 bytes inside a block of size 64 free'd
==3048727== at 0x484617B: free (vg_replace_malloc.c:872)
==3048727== by 0x2D207E: free_individual_entries (hashmap.c:188)
==3048727== by 0x2D2153: hashmap_clear_ (hashmap.c:207)
==3048727== by 0x270B5C: git_configset_clear (config.c:2375)
==3048727== by 0x1869AC: maintenance_unregister (gc.c:1585)
==3048727== by 0x188F42: cmd_maintenance (gc.c:2651)
==3048727== by 0x128C62: run_builtin (git.c:466)
==3048727== by 0x12907E: handle_builtin (git.c:721)
==3048727== by 0x1292EC: run_argv (git.c:788)
==3048727== by 0x12988E: cmd_main (git.c:926)
==3048727== by 0x21ED39: main (common-main.c:57)
[...]
Resolve this via a partial-revert of be0fd57228. The config_set struct
now gets a zero initialization, which makes free()-ing it a noop even
without calling git_configset_init(). When we do initialize it to a
non-zero value, it is only free()'d after our last read of `list`.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Since (maintenance: add option to register in a specific config,
2022-11-09) we've been unable to build with "DEVELOPER=1" without
"DEVOPTS=no-error", as the added code triggers a
"-Wdeclaration-after-statement" warning.
And worse than that, the data handed to git_configset_clear() is
uninitialized, as can be spotted with e.g.:
./t7900-maintenance.sh -vixd --run=23 --valgrind
[...]
+ git maintenance unregister --force
Conditional jump or move depends on uninitialised value(s)
at 0x6B5F1E: git_configset_clear (config.c:2367)
by 0x4BA64E: maintenance_unregister (gc.c:1619)
by 0x4BD278: cmd_maintenance (gc.c:2650)
by 0x409905: run_builtin (git.c:466)
by 0x40A21C: handle_builtin (git.c:721)
by 0x40A58E: run_argv (git.c:788)
by 0x40AF68: cmd_main (git.c:926)
by 0x5D39FE: main (common-main.c:57)
Uninitialised value was created by a stack allocation
at 0x4BA22C: maintenance_unregister (gc.c:1557)
Let's fix both of these issues, and also move the scope of the
variable to the "if" statement it's used in, to make it obvious where
it's used.
Helped-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
maintenance register currently records the maintenance repo exclusively
within the user's global configuration, but other configuration files
may be relevant when running maintenance if they are included from the
global config. This option allows the user to choose where maintenance
repos are recorded.
Signed-off-by: Ronan Pigott <ronan@rjp.ie>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
This is a quality of life change for git-maintenance, so repos can be
recorded with the tilde syntax. The register subcommand will not record
repos in this format by default.
Signed-off-by: Ronan Pigott <ronan@rjp.ie>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Add trace2 counters to the region to clear skip worktree bits in a
sparse checkout.
* al/trace2-clearing-skip-worktree:
index: raise a bug if the index is materialised more than once
index: add trace2 region for clear skip worktree
Previously the docs only described storage helpers.
A concrete example: Git Credential Manager can generate credentials
for GitHub and GitLab via OAuth.
https://github.com/GitCredentialManager/git-credential-manager
Signed-off-by: M Hickford <mirth.hickford@gmail.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
With GIT_TRACE_CURL=1 or GIT_CURL_VERBOSE=1, sensitive headers like
"Authorization" and "Cookie" get redacted. However, since [1], curl's
h2h3 module (invoked when using HTTP/2) also prints headers in its
"info", which don't get redacted. For example,
echo 'github.com TRUE / FALSE 1698960413304 o foo=bar' >cookiefile &&
GIT_TRACE_CURL=1 GIT_TRACE_CURL_NO_DATA=1 git \
-c 'http.cookiefile=cookiefile' \
-c 'http.version=' \
ls-remote https://github.com/git/git refs/heads/main 2>output &&
grep 'cookie' output
produces output like:
23:04:16.920495 http.c:678 == Info: h2h3 [cookie: o=foo=bar]
23:04:16.920562 http.c:637 => Send header: cookie: o=<redacted>
Teach http.c to check for h2h3 headers in info and redact them using the
existing header redaction logic. This fixes the broken redaction logic
that we noted in the previous commit, so mark the redaction tests as
passing under HTTP2.
[1] f8c3724aa9
Helped-by: Jeff King <peff@peff.net>
Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
We have occasionally seen bugs that affect Git running only against an
HTTP/2 web server, not an HTTP one. For instance, b66c77a64e (http:
match headers case-insensitively when redacting, 2021-09-22). But since
we have no test coverage using HTTP/2, we only uncover these bugs in the
wild.
That commit gives a recipe for converting our Apache setup to support
HTTP/2, but:
- it's not necessarily portable
- we don't want to just test HTTP/2; we really want to do a variety of
basic tests for _both_ protocols
This patch handles both problems by running a duplicate of t5551
(labeled as t5559 here) with an alternate-universe setup that enables
HTTP/2. So we'll continue to run t5551 as before, but run the same
battery of tests again with HTTP/2. If HTTP/2 isn't supported on a given
platform, then t5559 should bail during the webserver setup, and
gracefully skip all tests (unless GIT_TEST_HTTPD has been changed from
"auto" to "yes", where the point is to complain when webserver setup
fails).
In theory other http-related test scripts could benefit from the same
duplication, but doing t5551 should give us a reasonable check of basic
functionality, and would have caught both bugs we've seen in the wild
with HTTP/2.
A few notes on the implementation:
- a script enables the server side config by calling enable_http2
before starting the webserver. This avoids even trying to load any
HTTP/2 config for t5551 (which is what lets it keep working with
regular HTTP even on systems that don't support it). This also sets
a prereq which can be used by individual tests.
- As discussed in b66c77a64e, the http2 module isn't compatible with
the "prefork" mpm, so we need to pick something else. I chose
"event" here, which works on my Debian system, but it's possible
there are platforms which would prefer something else. We can adjust
that later if somebody finds such a platform.
- The test "large fetch-pack requests can be sent using chunked
encoding" makes sure we use a chunked transfer-encoding by looking
for that header in the trace. But since HTTP/2 has its own streaming
mechanisms, we won't find such a header. We could skip the test
entirely by marking it with !HTTP2. But there's some value in making
sure that the fetch itself succeeded. So instead, we'll confirm that
either we're using HTTP2 _or_ we saw the expected chunked header.
- the redaction tests fail under HTTP/2 with recent versions of curl.
This is a bug! I've marked them with !HTTP2 here to skip them under
t5559 for the moment. Using test_expect_failure would be more
appropriate, but would require a bunch of boilerplate. Since we'll
be fixing them momentarily, let's just skip them for now to keep the
test suite bisectable, and we can re-enable them in the commit that
fixes the bug.
- one alternative layout would be to push most of t5551 into a
lib-t5551.sh script, then source it from both t5551 and t5559.
Keeping t5551 intact seemed a little simpler, as its one less level
of indirection for people fixing bugs/regressions in the non-HTTP/2
tests.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Git learned pushing submodules without pushing the superproject by
the user specifying --recurse-submodules=only through 6c656c3fe4
("submodules: add RECURSE_SUBMODULES_ONLY value", 2016-12-20) and
225e8bf778 ("push: add option to push only submodules", 2016-12-20).
For users who use this feature regularly, it is desirable to have an
equivalent configuration.
It turns out that such a configuration (push.recurseSubmodules=only) is
already supported, even though it is neither documented nor mentioned
in the commit messages, due to the way the --recurse-submodules=only
feature was implemented (a function used to parse --recurse-submodules
was updated to support "only", but that same function is used to parse
push.recurseSubmodules too). What is left is to document it and test it,
which is what this commit does.
There is a possible point of confusion when recursing into a submodule
that itself has the push.recurseSubmodules=only configuration, because
if a repository has only its submodules pushed and not itself, its
superproject can never be pushed. Therefore, treat such configurations
as being "on-demand", and print a warning message.
Signed-off-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
