Regression fix for a change made during this cycle.
* cs/http-use-basic-after-failed-negotiate:
Revert "remote-curl: fall back to basic auth if Negotiate fails"
t5551: test http interaction with credential helpers
When there are many renames between the old base of a series of commits
and the new base, the way sequencer.c, merge-recursive.c, and
diffcore-rename.c have traditionally split the work resulted in
redetecting the same renames with each and every commit being
transplanted. To address this, the last several commits have been
creating a cache of rename detection results, determining when it was
safe to use such a cache in subsequent merge operations, adding helper
functions, and so on. See the previous half dozen commit messages for
additional discussion of this optimization, particularly the message a
few commits ago entitled "add code to check for whether cached renames
can be reused". This commit finally ties all of that work together,
modifying the merge algorithm to make use of these cached renames.
For the testcases mentioned in commit 557ac0350d ("merge-ort: begin
performance work; instrument with trace2_region_* calls", 2020-10-28),
this change improves the performance as follows:
Before After
no-renames: 5.665 s ± 0.129 s 5.622 s ± 0.059 s
mega-renames: 11.435 s ± 0.158 s 10.127 s ± 0.073 s
just-one-mega: 494.2 ms ± 6.1 ms 500.3 ms ± 3.8 ms
That's a fairly small improvement, but mostly because the previous
optimizations were so effective for these particular testcases; this
optimization only kicks in when the others don't. If we undid the
basename-guided rename detection and skip-irrelevant-renames
optimizations, then we'd see that this series by itself improved
performance as follows:
Before Basename Series After Just This Series
no-renames: 13.815 s ± 0.062 s 5.697 s ± 0.080 s
mega-renames: 1799.937 s ± 0.493 s 205.709 s ± 0.457 s
Since this optimization kicks in to help accelerate cases where the
previous optimizations do not apply, this last comparison shows that
this cached-renames optimization has the potential to help signficantly
in cases that don't meet the requirements for the other optimizations to
be effective.
The changes made in this optimization also lay some important groundwork
for a future optimization around having collect_merge_info() avoid
recursing into subtrees in more cases.
However, for this optimization to be effective, merge_switch_to_result()
should only be called when the rebase or cherry-pick operation has
either completed or hit a case where the user needs to resolve a
conflict or edit the result. If it is called after every commit, as
sequencer.c does, then the working tree and index are needlessly updated
with every commit and the cached metadata is tossed, defeating this
optimization. Refactoring sequencer.c to only call
merge_switch_to_result() at the end of the operation is a bigger
undertaking, and the practical benefits of this optimization will not be
realized until that work is performed. Since `test-tool fast-rebase`
only updates at the end of the operation, it was used to obtain the
timings above.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
As documented in Documentation/technical/remembering-renames.txt, and as
tested for in the two testcases in t6429 with "rename same file
identically" in their description, there is one case where we need to
have renames in one commit NOT be cached for the next commit in our
rebase sequence -- namely, rename/rename(1to1) cases. Rather than
specifically trying to uncache those and fix up dir_rename_counts() to
match (which would also be valid but more work), we simply disable the
optimization when this really rare type of rename occurs.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If we have a usable rename cache, then we can remove from
relevant_sources all the paths that were cached;
diffcore_rename_extended() can then consider an even smaller set of
relevant_sources in its rename detection.
However, when diffcore_rename_extended() is done, we will need to take
the renames it detected and then add back in all the ones we had cached
from before.
Add helper functions for doing these two operations; the next commit
will make use of them.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Previous commits created an in-memory cache of the results of rename
detection, and added logic to detect when that cache could appropriately
be used in a subsequent merge operation -- but we were still
unconditionally clearing the cache with each new merge operation anyway.
If it is valid to reuse the cache from one of the two sides of history,
preserve that side.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Previously, callers of the merge-ort API could have passed an
uninitialized value for struct merge_result *result. However, we want
to check result to see if it has cached renames from a previous merge
that we can reuse; such values would be found behind result->priv.
However, if result->priv is uninitialized, attempting to access behind
it will give a segfault. So, we need result->priv to be NULL (which
will be the case if the caller does a memset(&result, 0)), or be written
by a previous call to the merge-ort machinery. Documenting this
requirement may help, but despite being the person who introduced this
requirement, I still missed it once and it did not fail in a very clear
way and led to a long debugging session.
Add a _properly_initialized field to merge_result; that value will be
0 if the caller zero'ed the merge_result, it will be set to a very
specific value by a previous run by the merge-ort machinery, and if it's
uninitialized it will most likely either be 0 or some value that does
not match the specific one we'd expect allowing us to throw a much more
meaningful error.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We need to know when renames detected in a previous merge operation can
be reused in a later merge operation. Consider the following setup
(from the git-rebase manpage):
A---B---C topic
/
D---E---F---G master
After rebasing, this will appear as:
A'--B'--C' topic
/
D---E---F---G master
Further, let's say that 'oldfile' was renamed to 'newfile' between E
and G. The rebase or cherry-pick of A onto G will involve a three-way
merge between E (as the merge base) and G and A. After detecting the
rename between E:oldfile and G:newfile, there will be a three-way
content merge of the following:
E:oldfile
G:newfile
A:oldfile
and produce a new result:
A':newfile
Now, when we want to pick B onto A', we will need to do a three-way
merge between A (as the merge-base) and A' and B. This will involve
a three-way content merge of
A:oldfile
A':newfile
B:oldfile
but only if we can detect that A:oldfile is similar enough to A':newfile
to be used together in a three-way content merge, i.e. only if we can
detect that A:oldfile and A':newfile are a rename. But we already know
that A:oldfile and A':newfile are similar enough to be used in a
three-way content merge, because that is precisely where A':newfile came
from in the previous merge.
Note that A & A' both appear in both merges. That gives us the
condition under which we can reuse renames.
There are a couple important points about this optimization:
- If the rebase or cherry-pick halts for user conflicts, these caches
are NOT saved anywhere. Thus, resuming a halted rebase or
cherry-pick will result in no reused renames for the next commit.
This is intentional, as user resolution can change files
significantly and in ways that violate the similarity assumptions
here.
- Technically, in a *very* narrow case this might give slightly
different results for rename detection. Using the example above,
if:
* E:oldfile had 20 lines
* G:newfile added 10 new lines at the beginning of the file
* A:oldfile deleted all but the first three lines of the file
then
=> A':newfile would have 13 lines, 3 of which matches those
in A:oldfile.
Consider the two cases:
* Without this optimization:
- the next step of the rebase operation (moving B to B')
would not detect the rename betwen A:oldfile and A':newfile
- we'd thus get a modify/delete conflict with the rebase
operation halting for the user to resolve, and have both
A':newfile and B:oldfile sitting in the working tree.
* With this optimization:
- the rename between A:oldfile and A':newfile would be detected
via the cache of renames
- a three-way merge between A:oldfile, A':newfile, and B:oldfile
would commence and be written to A':newfile
Now, is the difference in behavior a bug...or a bugfix? I can't
tell. Given that A:oldfile and A':newfile are not very similar,
when we three-way merge with B:oldfile it seems likely we'll hit a
conflict for the user to resolve. And it shouldn't be too hard for
users to see why we did that three-way merge; oldfile and newfile
*were* renames somewhere in the sequence. So, most of these corner
cases will still behave similarly -- namely, a conflict given to the
user to resolve. Also, consider the interesting case when commit B
is a clean revert of commit A. Without this optimization, a rebase
could not both apply a weird patch like A and then immediately
revert it; users would be forced to resolve merge conflicts. With
this optimization, it would successfully apply the clean revert.
So, there is certainly at least one case that behaves better. Even
if it's considered a "difference in behavior", I think both behaviors
are reasonable, and the time savings provided by this optimization
justify using the slightly altered rename heuristics.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Fill in cache_pairs, cached_target_names, and cached_irrelevant based on
rename detection results. Future commits will make use of these values.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When there are many renames between the old base of a series of commits
and the new base for a series of commits, the sequence of merges
employed to transplant those commits (from a cherry-pick or rebase
operation) will repeatedly detect the exact same renames. This is
wasted effort.
Add data structures which will be used to cache rename detection
results, along with the initialization and deallocation of these data
structures. Future commits will populate these caches, detect the
appropriate circumstances when they can be used, and employ them to
avoid re-detecting the same renames repeatedly.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We will soon be adding an optimization that caches (in memory only,
never written to disk) upstream renames during a sequence of merges such
as occurs during a cherry-pick or rebase operation. Add several tests
meant to stress such an implementation to ensure it does the right
thing, and include a test whose outcome we will later change due to this
optimization as well.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Previously, when fast-rebase hit a conflict, it simply aborted and left
HEAD, the index, and the working tree where they were before the
operation started. While fast-rebase does not support restarting from a
conflicted state, write the conflicted state out anyway as it gives us a
way to see what the conflicts are and write tests that check for them.
This will be important in the upcoming commits, because sequencer.c is
only superficially integrated with merge-ort.c; in particular, it calls
merge_switch_to_result() after EACH merge instead of only calling it at
the end of all the sequence of merges (or when a conflict is hit). This
not only causes needless updates to the working copy and index, but also
causes all intermediate data to be freed and tossed, preventing caching
information from one merge to the next. However, integrating
sequencer.c more deeply with merge-ort.c is a big task, and making this
small extension to fast-rebase.c provides us with a simple way to test
the edge and corner cases that we want to make sure continue working.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
assert() can succinctly document expectations for the code, and do so in
a way that may be useful to future folks trying to refactor the code and
change basic assumptions; it allows them to more quickly find some
places where their violations of previous assumptions trips things up.
Unfortunately, assert() can surround a function call with important
side-effects, which is a huge mistake since some users will compile with
assertions disabled. I've had to debug such mistakes before in other
codebases, so I should know better. Luckily, this was only in test
code, but it's still very embarrassing. Change an assert() to an if
(...) BUG (...).
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Remembering renames on the upstream side of history in an early merge of
a rebase or cherry-pick for re-use in a latter merge of the same
operation makes pretty good intuitive sense. However, trying to show
that it doesn't cause some subtle behavioral difference or some funny
edge or corner case is much more involved. And, in fact, it does
introduce a subtle behavioral change.
Document all the assumptions, special cases, and logic involved in such
an optimization, and describe why this optimization is safe under the
current optimizations/features/etc. -- even when the subtle behavioral
change is triggered.
Part of the point of adding this document that goes over the
optimization in such laborious detail, is that it is possible that
significant future changes (optimizations or feature changes) could
interact with this optimization in interesting ways; this document is
here to help folks making big changes sanity check that the assumptions
and arguments underlying this optimization are still valid. (As a side
note, creating this document forced me to review things in sufficient
detail that I found I was not properly caching directory-rename-induced
renames, resulting in the code not being aware of those renames and
causing unnecessary diffcore_rename_extended() calls in subsequent
merges.)
A subsequent commit will add several testcases based on this document
meant to stress-test the implementation and also document the case with
the subtle behavioral change.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The current documentation for color.pager is technically correct, but
slightly misleading and doesn't really clarify the purpose of the
variable. As explained in the original thread which added it:
https://lore.kernel.org/git/E1G6zPH-00062L-Je@moooo.ath.cx/
the point is to deal with pagers that don't understand colors. And hence it
being set to "true" is necessary for colorizing output to the pager, but
not sufficient by itself (you must also have enabled one of the other
color options, though note that these are set to "auto" by default these
days).
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Find typos in "po/TEAMS" file using the "git-po-helper" program. These
typos were introduced from commit v2.24.0-1-g9917eca794 (l10n: zh_TW:
add translation for v2.24.0, 2019-11-20 19:14:22 +0800).
Signed-off-by: Jiang Xin <worldhello.net@gmail.com>
The "chainlint" feature in the test framework is a handy way to
catch common mistakes in writing new tests, but tends to get
expensive. An knob to selectively disable it has been introduced
to help running tests that the developer has not modified.
* jk/test-chainlint-softer:
t: avoid sed-based chain-linting in some expensive cases
The handling of "%(push)" formatting element of "for-each-ref" and
friends was broken when the same codepath started handling
"%(push:<what>)", which has been corrected.
* zh/ref-filter-push-remote-fix:
ref-filter: fix read invalid union member bug
"git clone" from SHA256 repository by Git built with SHA-1 as the
default hash algorithm over the dumb HTTP protocol did not
correctly set up the resulting repository, which has been corrected.
* ew/sha256-clone-remote-curl-fix:
remote-curl: fix clone on sha256 repos
"git clean" and "git ls-files -i" had confusion around working on
or showing ignored paths inside an ignored directory, which has
been corrected.
* en/dir-traversal:
dir: introduce readdir_skip_dot_and_dotdot() helper
dir: update stale description of treat_directory()
dir: traverse into untracked directories if they may have ignored subfiles
dir: avoid unnecessary traversal into ignored directory
t3001, t7300: add testcase showcasing missed directory traversal
t7300: add testcase showing unnecessary traversal into ignored directory
ls-files: error out on -i unless -o or -c are specified
dir: report number of visited directories and paths with trace2
dir: convert trace calls to trace2 equivalents
When fetching with the v0 protocol over ssh (or a local upload-pack with
pipes), the server closes the connection as soon as it is finished
sending the pack. So even though the client may still be operating on
the data via index-pack (e.g., resolving deltas, checking connectivity,
etc), the server has released all resources.
With the v2 protocol, however, the server considers the ssh session only
as a transport, with individual requests coming over it. After sending
the pack, it goes back to its main loop, waiting for another request to
come from the client. As a result, the ssh session hangs around until
the client process ends, which may be much later (because resolving
deltas, etc, may consume a lot of CPU).
This is bad for two reasons:
- it's consuming resources on the server to leave open a connection
that won't see any more use
- if something bad happens to the ssh connection in the meantime (say,
it gets killed by the network because it's idle, as happened in a
real-world report), then ssh will exit non-zero, and we'll propagate
the error up the stack.
The server is correct here not to hang up after serving the pack. The v2
protocol's design is meant to allow multiple requests like this, and
hanging up would be the wrong thing for a hypothetical client which was
planning to make more requests (though in practice, the git.git client
never would, and I doubt any other implementations would either).
The right thing is instead for the client to signal to the server that
it's not interested in making more requests. We can do that by closing
the pipe descriptor we use to write to ssh. This will propagate to the
server upload-pack as an EOF when it tries to read the next request (and
then it will close its half, and the whole connection will go away).
It's important to do this "half duplex" shutdown, because we have to do
it _before_ we actually receive the pack. This is an artifact of the way
fetch-pack and index-pack (or unpack-objects) interact. We hand the
connection off to index-pack (really, a sideband demuxer which feeds
it), and then wait until it returns. And it doesn't do that until it has
resolved all of the deltas in the pack, even though it was done reading
from the server long before.
So just closing the connection fully after index-pack returns would be
too late; we'd have held it open much longer than was necessary. And
teaching index-pack to close the connection is awkward. It's not even
seeing the whole conversation (the sideband demuxer is, but it doesn't
actually know what's in the packets, or when the end comes).
Note that this close() is happening deep within the transport code. It's
possible that a caller would want to perform other operations over the
same ssh transport after receiving the pack. But as of the current code,
none of the callers do, and there haven't been discussions of any plans
to change this. If we need to support that later, we can probably do so
by passing down a flag for "you're the last request on the transport;
it's OK to close" instead of the code just assuming that's true.
The description above all discusses v2 ssh, so it's worth thinking about
how this interacts with other protocols:
- in v0 protocols, we could do the same half-duplex shutdown (it just
goes into the v0 do_fetch_pack() instead). This does work, but since
it doesn't have the same persistence problem in the first place,
there's little reason to change it at this point.
- local fetches against git-upload-pack on the same machine will
behave the same as ssh (they are talking over two pipes, and see EOF
on their input pipe)
- fetches against git-daemon will run this same code, and close one of
the descriptors. In practice, this won't do anything, since there
our two descriptors are dups of each other, and not part of a
half-duplex pair. The right thing would probably be to call
shutdown(SHUT_WR) on it. I didn't bother with that here. It doesn't
face the same error-code problem (since it's just a TCP connection),
so it's really only an optimization problem. And git:// is not that
widely used these days, and has less impact on server resources than
an ssh termination.
- v2 http doesn't suffer from this problem in the first place, as our
pipes terminate at a local git-remote-https, which is passing data
along as individual requests via curl. Probably curl is keeping the
TCP/TLS connection open for more requests, and we might be able to
tell it manually "hey, we are done making requests now". But I think
that's much less important. It again doesn't suffer from the
error-code problem, and HTTP keepalive is pretty well understood
(importantly, the timeouts can be set low, because clients like curl
know how to reconnect for subsequent requests if necessary). So it's
probably not worth figuring out how to tell curl that we're done
(though if we do, this patch is probably the first step anyway;
fetch-pack closes the pipe back to remote-https, which would be the
signal that it should tell curl we're done).
The code is pretty straightforward. We close the pipe at the right
moment, and set it to -1 to mark it as invalid. I modified the later
cleanup code to avoid calling close(-1). That's not strictly necessary,
since close(-1) is a noop, but hopefully makes things a bit more obvious
to a reader.
I suspect that trying to call more transport functions after the close()
(e.g., calling transport_fetch_refs() again) would fail, as it's not
smart enough to realize we need to re-open the ssh connection. But
that's already true when v0 is in use. And no current callers want to do
that (and again, the solution is probably a flag in the transport code
to keep things open, which can be added later).
There's no test here, as the situation it covers is inherently racy (the
question is when upload-pack exits, compared to when index-pack finishes
resolving deltas and exits). The rather gross shell snippet below does
recreate the problematic situation; when run on a sufficiently-large
repository (git.git works fine), it kills an "idle" upload-pack while
the client is resolving deltas, leading to a failed clone.
(
git clone --no-local --progress . foo.git 2>&1
echo >&2 "clone exit code=$?"
) |
tr '\r' '\n' |
while read line
do
case "$done,$line" in
,Resolving*)
echo "hit resolving deltas; killing upload-pack"
killall -9 git-upload-pack
done=t
;;
esac
done
Reported-by: Greg Pflaum <greg.pflaum@pnp-hcl.com>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
git-clone started respecting errors from the transport subsystem in
aab179d937 (builtin/clone.c: don't ignore transport_fetch_refs() errors,
2020-12-03). However, that commit didn't handle the cleanup of the
filesystem quite right.
The cleanup of the directory that cmd_clone() creates is done by an
atexit() handler, which we control with a flag. It starts as
JUNK_LEAVE_NONE ("clean up everything"), then progresses to
JUNK_LEAVE_REPO when we know we have a valid repo but not working tree,
and then finally JUNK_LEAVE_ALL when we have a successful checkout.
Most errors cause us to die(), which then triggers the handler to do the
right thing based on how far into cmd_clone() we got. But the checks
added by aab179d937 instead set the "err" variable and then jump to a
new "cleanup" label, which then returns our non-zero status. However,
the code after the cleanup label includes setting the flag to
JUNK_LEAVE_ALL, and so we accidentally leave the repository and working
tree in place.
One obvious option to fix this is to reorder the end of the function to
set the flag first, before cleanup code, and put the label between them.
But we can observe another small bug: the error return from
transport_fetch_refs() is generally "-1", and we propagate that to the
return value of cmd_clone(), which ultimately becomes the exit code of
the process. And we try to avoid transmitting negative values via exit
codes (only the low 8 bits are passed along as an unsigned value, though
in practice for "-1" this at least retains the property that it's
non-zero).
Instead, let's just die(). That makes us consistent with rest of the
code in the function. It does add a new "fatal:" line to the output, but
I'd argue that's a good thing:
- in the rare case that the transport code didn't say anything, now
the user gets _some_ error message
- even if the transport code said something like "error: ssh died of
signal 9", it's nice to also say "fatal" to indicate that we
considered that to be a show-stopper.
Triggering this in the test suite turns out to be surprisingly
difficult. Almost every error we'd encounter, including ones deep inside
the transport code, cause us to just die() right there! However, one way
is to put a fake wrapper around git-upload-pack that sends the complete
packfile but exits with a failure code.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The text was somewhat confusing between the revision itself and the author.
Signed-off-by: Reuven Yagel <robi@post.jce.ac.il>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
These methods were marked as MAYBE_UNUSED in the previous change to
avoid a complicated diff. Delete them entirely, since we now use the
hashfile API instead of this custom hashing code.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The do_write_index() method in read-cache.c has its own hashing logic
and buffering mechanism. Specifically, the ce_write() method was
introduced by 4990aadc (Speed up index file writing by chunking it
nicely, 2005-04-20) and similar mechanisms were introduced a few months
later in c38138cd (git-pack-objects: write the pack files with a SHA1
csum, 2005-06-26). Based on the timing, in the early days of the Git
codebase, I figured that these roughly equivalent code paths were never
unified only because it got lost in the shuffle. The hashfile API has
since been used extensively in other file formats, such as pack-indexes,
multi-pack-indexes, and commit-graphs. Therefore, it seems prudent to
unify the index writing code to use the same mechanism.
I discovered this disparity while trying to create a new index format
that uses the chunk-format API. That API uses a hashfile as its base, so
it is incompatible with the custom code in read-cache.c.
This rewrite is rather straightforward. It replaces all writes to the
temporary file with writes to the hashfile struct. This takes care of
many of the direct interactions with the_hash_algo.
There are still some git_hash_ctx uses remaining: the extension headers
are hashed for use in the End of Index Entries (EOIE) extension. This
use of the git_hash_ctx is left as-is. There are multiple reasons to not
use a hashfile here, including the fact that the data is not actually
writing to a file, just a hash computation. These hashes do not block
our adoption of the chunk-format API in a future change to the index, so
leave it as-is.
The internals of the algorithms are mostly identical. Previously, the
hashfile API used a smaller 8KB buffer instead of the 128KB buffer from
read-cache.c. The previous change already unified these sizes.
There is one subtle point: we do not pass the CSUM_FSYNC to the
finalize_hashfile() method, which differs from most consumers of the
hashfile API. The extra fsync() call indicated by this flag causes a
significant peformance degradation that is noticeable for quick
commands that write the index, such as "git add". Other consumers can
absorb this cost with their more complicated data structure
organization, and further writing structures such as pack-files and
commit-graphs is rarely in the critical path for common user
interactions.
Some static methods become orphaned in this diff, so I marked them as
MAYBE_UNUSED. The diff is much harder to read if they are deleted during
this change. Instead, they will be deleted in the following change.
In addition to the test suite passing, I computed indexes using the
previous binaries and the binaries compiled after this change, and found
the index data to be exactly equal. Finally, I did extensive performance
testing of "git update-index --force-write" on repos of various sizes,
including one with over 2 million paths at HEAD. These tests
demonstrated less than 1% difference in behavior. As expected, the
performance should be considered unchanged. The previous changes to
increase the hashfile buffer size from 8K to 128K ensured this change
would not create a peformance regression.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The hashfile API uses a hard-coded buffer size of 8KB and has ever since
it was introduced in c38138c (git-pack-objects: write the pack files
with a SHA1 csum, 2005-06-26). It performs a similar function to the
hashing buffers in read-cache.c, but that code was updated from 8KB to
128KB in f279894 (read-cache: make the index write buffer size 128K,
2021-02-18). The justification there was that do_write_index() improves
from 1.02s to 0.72s. Since our end goal is to have the index writing
code use the hashfile API, we need to unify this buffer size to avoid a
performance regression.
There is a buffer, 'check_buffer', that is used to verify the check_fd
file descriptor. When this buffer increases to 128K to fit the data
being flushed, it causes the stack to overflow the limits placed in the
test suite. To avoid issues with stack size, move both 'buffer' and
'check_buffer' to be heap pointers within 'struct hashfile'. The
'check_buffer' member is left as NULL unless check_fd is set in
hashfd_check(). Both buffers are cleared as part of finalize_hashfile()
which also frees the full structure.
Since these buffers are now on the heap, we can adjust their size based
on the needs of the consumer. In particular, callers to
hashfd_throughput() are expecting to report progress indicators as the
buffer flushes. These callers would prefer the smaller 8k buffer to
avoid large delays between updates, especially for users with slower
networks. When the progress indicator is not used, the larger buffer is
preferrable.
By adding a new trace2 region in the chunk-format API, we can see that
the writing portion of 'git multi-pack-index write' lowers from ~1.49s
to ~1.47s on a Linux machine. These effects may be more pronounced or
diminished on other filesystems. The end-to-end timing is too noisy to
have a definitive change either way.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The xsize_t helper aims to safely convert an off_t to a size_t,
erroring out when a file offset is too large to fit into a memory
address. It does this by using two casts:
size_t size = (size_t) len;
if (len != (off_t) size)
... error out ...
On a platform with sizeof(size_t) < sizeof(off_t), this check is safe
and correct. The first cast truncates to a size_t by finding the
remainder modulo SIZE_MAX+1 (see C99 section 6.3.1.3 Signed and
unsigned integers) and the second promotes to an off_t, meaning the
result is true if and only if len is representable as a size_t.
On other platforms, this two-casts strategy still works well (always
succeeds) for len >= 0. But for len < 0, when the first cast succeeds
and produces SIZE_MAX + 1 + len, the resulting value is too large to
be represented as an off_t, so the second cast produces implementation
defined behavior. In practice, it is likely to produce a result of
true despite len not being representable as size_t.
Simplify by replacing with a more straightforward check: compare len
to the relevant bounds and then cast it. (To avoid a -Wsign-compare
warning, after checking that len >= 0, we explicitly convert to a
sufficiently-large unsigned type before comparing to SIZE_MAX.)
In practice, this is not likely to come up since typical callers use
nonnegative len. Still, it's helpful to handle this case to make the
behavior easy to reason about.
Historical note: the original bounds-checking in 46be82dfd0 (xsize_t:
check whether we lose bits, 2010-07-28) did not produce this
implementation-defined behavior, though it still did not handle
negative offsets. It was not until 73560c793a (git-compat-util.h:
xsize_t() - avoid -Wsign-compare warnings, 2017-09-21) introduced the
double cast that the implementation-defined behavior was triggered.
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This reverts commit 1b0d9545bb85912a16b367229d414f55d140d3be.
That commit does fix the situation it intended to (avoiding Negotiate
even when the credentials were provided in the URL), but it creates a
more serious regression: we now never hit the conditional for "we had a
username and password, tried them, but the server still gave us a 401".
That has two bad effects:
1. we never call credential_reject(), and thus a bogus credential
stored by a helper will live on forever
2. we never return HTTP_NOAUTH, so the error message the user gets is
"The requested URL returned error: 401", instead of "Authentication
failed".
Doing this correctly seems non-trivial, as we don't know whether the
Negotiate auth was a problem. Since this is a regression in the upcoming
v2.23.0 release (for which we're in -rc0), let's revert for now and work
on a fix separately.
(Note that this isn't a pure revert; the previous commit added a test
showing the regression, so we can now flip it to expect_success).
Reported-by: Ben Humphreys <behumphreys@atlassian.com>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We test authentication with http, and we independently test that
credential helpers work, but we don't have any tests that cover the
two features working together. Let's add two:
1. Make sure that a successful request asks the helper to save the
credential. This works as expected.
2. Make sure that a failed request asks the helper to forget the
credential. This is marked as expect_failure, as it was recently
regressed by 1b0d9545bb (remote-curl: fall back to basic auth if
Negotiate fails, 2021-03-22). The symptom here is that the second
request should prompt the user, but doesn't.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Sometimes new people are confused by how a revision "range" works,
in that it is not a random collection of commits but a set of
commits that are all connected to each other, and most Git commands
work on a single such "range".
Give an example to clarify it.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The flush() logic in csum-file.c was introduced originally by c38138c
(git-pack-objects: write the pack files with a SHA1 csum, 2005-06-26)
and a portion of the logic performs similar utility to write_in_full()
in wrapper.c. The history of write_in_full() is full of moves and
renames, but was originally introduced by 7230e6d (Add write_or_die(), a
helper function, 2006-08-21).
The point of these sections of code are to flush a write buffer using
xwrite() and report errors in the case of disk space issues or other
generic input/output errors. The logic in flush() can interpret the
output of write_in_full() to provide the correct error messages to
users.
The logic in the hashfile API has an additional set of logic to augment
the progress indicator between calls to xwrite(). This was introduced by
2a128d6 (add throughput display to git-push, 2007-10-30). It seems that
since the hashfile's buffer is only 8KB, these additional progress
indicators might not be incredibly necessary. Instead, update the
progress only when write_in_full() complete.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
While testing the sparse-index, I verified a test with --valgrind and it
complained about an uninitialized value being used in a jump in the
path_matches_pattern_list() method. The line was this one:
if (*dtype == DT_UNKNOWN)
In the call stack, the culprit was the initialization of the dtype
variable in convert_to_sparse_rec().
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
An object_id storing a SHA-1 name has some unused bytes at the end of
the hash array. Since these bytes are not used, they are usually not
initialized to any value either. However, at
parallel_checkout.c:send_one_item() the object_id of a cache entry is
copied into a buffer which is later sent to a checkout worker through a
pipe write(). This makes Valgrind complain about passing uninitialized
bytes to a syscall. The worker won't use these uninitialized bytes
either, but the warning could confuse someone trying to debug this code;
So instead of using oidcpy(), send_one_item() uses hashcpy() to only
copy the used/initialized bytes of the object_id, and leave the
remaining part with zeros.
However, since cf0983213c ("hash: add an algo member to struct
object_id", 2021-04-26), using hashcpy() is no longer sufficient here as
it won't copy the new algo field from the object_id. Let's add and use a
new function which meets both our requirements of copying all the
important object_id data while still avoiding the uninitialized bytes,
by padding the end of the hash array in the destination object_id. With
this change, we also no longer need the destination buffer from
send_one_item() to be initialized with zeros, so let's switch from
xcalloc() to xmalloc() to make this clear.
Signed-off-by: Matheus Tavares <matheus.bernardino@usp.br>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The C_LOCALE_OUTPUT prerequisite was removed in b1e079807b (tests:
remove last uses of C_LOCALE_OUTPUT, 2021-02-11), where Ævar noted:
I'm not leaving the prerequisite itself in place for in-flight changes
as there currently are none that introduce new tests that rely on it,
and because C_LOCALE_OUTPUT is currently a noop on the master branch
we likely won't have any new submissions that use it.
One more use of C_LOCALE_OUTPUT did creep in with 3d1bda6b5b (t7500: add
tests for --fixup=[amend|reword] options, 2021-03-15). This causes a
number of the tests to be skipped by default:
ok 35 # SKIP --fixup=reword: incompatible with --all (missing C_LOCALE_OUTPUT)
ok 36 # SKIP --fixup=reword: incompatible with --include (missing C_LOCALE_OUTPUT)
ok 37 # SKIP --fixup=reword: incompatible with --only (missing C_LOCALE_OUTPUT)
ok 38 # SKIP --fixup=reword: incompatible with --interactive (missing C_LOCALE_OUTPUT)
ok 39 # SKIP --fixup=reword: incompatible with --patch (missing C_LOCALE_OUTPUT)
Remove the C_LOCALE_OUTPUT prerequisite from these tests so they are
not skipped.
Signed-off-by: Todd Zullinger <tmz@pobox.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
These error messages are intended for the user. Let's touch them up
since we're here from the previous commit.
Signed-off-by: Wolfgang Müller <wolf@oriole.systems>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Calling "git rev-parse --path-format" without an argument segfaults
instead of giving an error message. Commit fac60b8925 (rev-parse: add
option for absolute or relative path formatting, 2020-12-13) added the
argument parsing code but forgot to handle NULL.
Returning an error makes sense here because there is no default value we
could use. Add a test case to verify.
Signed-off-by: Wolfgang Müller <wolf@oriole.systems>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* Eliminated 'Negation of emptiness' of 'nenhum' (not one/none)
* Eliminated 'Negation of emptiness' of 'nada' (nothing)
* Transformed 'Não' (No) into affirmative
* Some other translations
* Transforming 'não' (no) into affirmative
* From junção-de-3 to tri-junção
Signed-off-by: Daniel Santos <hello@brighterdan.com>
Clarify the default length used for the abbreviated form used for
commits in git describe.
The behavior was modified in Git 2.11.0, but the documentation was not
updated to clarify the new behavior.
Signed-off-by: Anders Höckersten <anders@hockersten.se>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
I sometimes receive patches from people with short mononyms, and in my
cultural environment these are not uncommon. To my dismay, git-am
currently discards their names, and replaces them with their email
addresses.
Link: https://www.kalzumeus.com/2010/06/17/falsehoods-programmers-believe-about-names/
Signed-off-by: edef <edef@edef.eu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
