Performance measurements are listed right now as a flat list, which is
fine when we measure big blocks. But when we start adding more and
more measurements, some of them could be just part of a bigger
measurement and a flat list gives a wrong impression that they are
executed at the same level instead of nested.
Add trace_performance_enter() and trace_performance_leave() to allow
indent these nested measurements. For now it does not help much
because the only nested thing is (lazy) name hash initialization
(e.g. called in diff-index from "git status"). This will help more
because I'm going to add some more tracing that's actually nested.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git add" files in the same directory, but spelling the directory
path in different cases on case insensitive filesystem, corrupted
the name hash data structure and led to unexpected results. This
has been corrected.
* bp/name-hash-dirname-fix:
name-hash: properly fold directory names in adjust_dirname_case()
Correct the pointer arithmetic in adjust_dirname_case() so that it calls
find_dir_entry() with the correct string length. Previously passing in
"dir1/foo" would pass a length of 6 instead of the correct 4. This resulted in
find_dir_entry() never finding the entry and so the subsequent memcpy that would
fold the name to the version with the correct case never executed.
Add a test to validate the corrected behavior with name folding of directories.
Signed-off-by: Ben Peart <benpeart@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
All the known heavy code blocks are measured (except object database
access). This should help identify if an optimization is effective or
not. An unoptimized git-status would give something like below:
0.001791141 s: read cache ...
0.004011363 s: preload index
0.000516161 s: refresh index
0.003139257 s: git command: ... 'status' '--porcelain=2'
0.006788129 s: diff-files
0.002090267 s: diff-index
0.001885735 s: initialize name hash
0.032013138 s: read directory
0.051781209 s: git command: './git' 'status'
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is to address concerns raised by ThreadSanitizer on the mailing list
about threaded unprotected R/W access to map.size with my previous "disallow
rehash" change (0607e10009).
See:
https://public-inbox.org/git/adb37b70139fd1e2bac18bfd22c8b96683ae18eb.1502780344.git.martin.agren@gmail.com/
Add API to hashmap to disable item counting and thus automatic rehashing.
Also include API to later re-enable them.
When item counting is disabled, the map.size field is invalid. So to
prevent accidents, the field has been renamed and an accessor function
hashmap_get_size() has been added. All direct references to this
field have been been updated. And the name of the field changed
to map.private_size to communicate this.
Here is the relevant output from ThreadSanitizer showing the problem:
WARNING: ThreadSanitizer: data race (pid=10554)
Read of size 4 at 0x00000082d488 by thread T2 (mutexes: write M16):
#0 hashmap_add hashmap.c:209
#1 hash_dir_entry_with_parent_and_prefix name-hash.c:302
#2 handle_range_dir name-hash.c:347
#3 handle_range_1 name-hash.c:415
#4 lazy_dir_thread_proc name-hash.c:471
#5 <null> <null>
Previous write of size 4 at 0x00000082d488 by thread T1 (mutexes: write M31):
#0 hashmap_add hashmap.c:209
#1 hash_dir_entry_with_parent_and_prefix name-hash.c:302
#2 handle_range_dir name-hash.c:347
#3 handle_range_1 name-hash.c:415
#4 handle_range_dir name-hash.c:380
#5 handle_range_1 name-hash.c:415
#6 lazy_dir_thread_proc name-hash.c:471
#7 <null> <null>
Martin gives instructions for running TSan on test t3008 in this post:
https://public-inbox.org/git/CAN0heSoJDL9pWELD6ciLTmWf-a=oyxe4EXXOmCKvsG5MSuzxsA@mail.gmail.com/
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When using the hashmap a common need is to have access to caller provided
data in the compare function. A couple of times we abuse the keydata field
to pass in the data needed. This happens for example in patch-ids.c.
This patch changes the function signature of the compare function
to have one more void pointer available. The pointer given for each
invocation of the compare function must be defined in the init function
of the hashmap and is just passed through.
Documentation of this new feature is deferred to a later patch.
This is a rather mechanical conversion, just adding the new pass-through
parameter. However while at it improve the naming of the fields of all
compare functions used by hashmaps by ensuring unused parameters are
prefixed with 'unused_' and naming the parameters what they are (instead
of 'unused' make it 'unused_keydata').
Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add check for the end of the entries for the thread partition.
Add test for lazy init name hash with specific directory structure
The lazy init hash name was causing a buffer overflow when the last
entry in the index was multiple folder deep with parent folders that
did not have any files in them.
This adds a test for the boundary condition of the thread partitions
with the folder structure that was triggering the buffer overflow.
The fix was to check if it is the last entry for the thread partition
in the handle_range_dir and not try to use the next entry in the cache.
Signed-off-by: Kevin Willford <kewillf@microsoft.com>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Improve performance of lazy_init_name_hash() when
ignore_case is set. Teach name-hash to build the
istate.name_hash and istate.dir_hash simultaneously
using a forward-diving technique on the pathname
of the index_entry, rather than adding name_hash
entries and then searching backwards in the pathname
for parent directories.
This borrows algorithm ideas from clear_ce_flags_{1,dir}.
Multiple threads are used with the new algorithm to
speed hashmap construction.
This new code path is only used when threads are present
(a compiler settings) and when the index is large enough
to warrant the pthread complexity.
The code in clear_ce_flags_dir() uses a linear search to
find the adjacent index entries with the same prefix; a
binary search is used here handle_range_dir() to further
speed things up.
The size of LAZY_THREAD_COST was determined from rough
analysis using:
t/helper/test-lazy-init-name-hash --analyze
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Specify an initial size for the istate.dir_hash HashMap matching
the size of the istate.name_hash.
Previously hashmap_init() was given 0, causing a 64 bucket
hashmap to be created. When working with very large
repositories, this would cause numerous rehash() calls to
realloc and rebalance the hashmap. This is especially true
when the worktree is deep, with many directories containing
a few files.
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Using FLEX_ARRAY macros reduces the amount of manual
computation size we have to do. It also ensures we don't
overflow size_t, and it makes sure we write the same number
of bytes that we allocated.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Stop reusing cache_entry in dir_entry; doing so causes a
use-after-free bug.
During merges, we free entries that we no longer need in the
destination index. But those entries might have also been stored in
the dir_entry cache, and when a later call to add_to_index found them,
they would be used after being freed.
To prevent this, change dir_entry to store a copy of the name instead
of a pointer to a cache_entry. This entails some refactoring of code
that expects the cache_entry.
Keith McGuigan <kmcguigan@twitter.com> diagnosed this bug and wrote
the initial patch, but this version does not use any of Keith's code.
Helped-by: Keith McGuigan <kmcguigan@twitter.com>
Helped-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: David Turner <dturner@twopensource.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Hashmap entries are typically looked up by just a key. The hashmap_get()
API expects an initialized entry structure instead, to support compound
keys. This flexibility is currently only needed by find_dir_entry() in
name-hash.c (and compat/win32/fscache.c in the msysgit fork). All other
(currently five) call sites of hashmap_get() have to set up a near emtpy
entry structure, resulting in duplicate code like this:
struct hashmap_entry keyentry;
hashmap_entry_init(&keyentry, hash(key));
return hashmap_get(map, &keyentry, key);
Add a hashmap_get_from_hash() API that allows hashmap lookups by just
specifying the key and its hash code, i.e.:
return hashmap_get_from_hash(map, hash(key), key);
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The same_name() private function wants a quick-and-exact check to
see if they two names are byte-for-byte identical first and then
fall back to the slow path. Use memcmp(3) for the former to make it
clear that we do not want any "name" specific comparison.
Signed-off-by: Jeremiah Mahler <jmmahler@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
db5360f3f4 (name-hash: refactor polymorphic index_name_exists();
2013-09-17) split index_name_exists() into index_file_exists() and
index_dir_exists() but retained index_name_exists() as a thin wrapper
to avoid disturbing possible in-flight topics. Since this change
landed in 'master' some time ago and there are no in-flight topics
referencing index_name_exists(), retire it.
Signed-off-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The new hashmap implementation supports remove, so really remove unused
cache entries from the name hashmap instead of just marking them.
The CE_UNHASHED flag and CE_STATE_MASK are no longer needed.
Keep the CE_HASHED flag to prevent adding entries twice.
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Note: the "ce->next = NULL;" in unpack-trees.c::do_add_entry can safely be
removed, as ce->next (now ce->ent.next) is always properly initialized in
name-hash.c::hash_index_entry.
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The new hashmap implementation supports remove, so remove and free
directory entries that are no longer referenced by active cache entries.
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When 5102c617 (Add case insensitivity support for directories when using
git status, 2010-10-03) added directories to the name-hash there was
only a single hash table in which both real cache entries and leading
directory prefixes were registered. To distinguish between the two types
of entries, directories were stored with a trailing '/'.
2092678c (name-hash.c: fix endless loop with core.ignorecase=true,
2013-02-28), however, moved directories to a separate hash table
(index_state.dir_hash) but retained the (now) redundant trailing '/',
thus callers continue to bear the burden of ensuring the slash's
presence before searching the index for a directory. Eliminate this
redundancy by storing paths in the dir-hash without the trailing '/'.
An important benefit of this change is that it eliminates undocumented
and dangerous behavior of dir.c:directory_exists_in_index_icase() in
which it assumes not only that it can validly access one character
beyond the end of its incoming directory argument, but also that that
character will unconditionally be a '/'. This perilous behavior was
"tolerated" because the string passed in by its lone caller always had a
'/' in that position, however, things broke [1] when 2eac2a4c (ls-files
-k: a directory only can be killed if the index has a non-directory,
2013-08-15) added a new caller which failed to respect the undocumented
assumption.
[1]: http://thread.gmane.org/gmane.comp.version-control.git/232727
Signed-off-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Depending upon the absence or presence of a trailing '/' on the incoming
pathname, index_name_exists() checks either if a file is present in the
index or if a directory is represented within the index. Each caller
explicitly chooses the mode of operation by adding or removing a
trailing '/' before invoking index_name_exists().
Since these two modes of operations are disjoint and have no code in
common (one searches index_state.name_hash; the other dir_hash), they
can be represented more naturally as distinct functions: one to search
for a file, and one for a directory.
Splitting index searching into two functions relieves callers of the
artificial burden of having to add or remove a slash to select the mode
of operation; instead they just call the desired function. A subsequent
patch will take advantage of this benefit in order to eliminate the
requirement that the incoming pathname for a directory search must have
a trailing slash.
(In order to avoid disturbing in-flight topics, index_name_exists() is
retained as a thin wrapper dispatching either to index_dir_exists() or
index_file_exists().)
Signed-off-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The code to keep track of what directory names are known to Git on
platforms with case insensitive filesystems can get confused upon
a hash collision between these pathnames and looped forever.
* kb/name-hash:
name-hash.c: fix endless loop with core.ignorecase=true
When we know approximately how many entries we will have in the
hash-table, it makes sense to size the hash table to that number
from the beginning to avoid unnecessary rehashing.
* nd/preallocate-hash:
Preallocate hash tables when the number of inserts are known in advance
This avoids unnecessary re-allocations and reinsertions. On webkit.git
(i.e. about 182k inserts to the name hash table), this reduces about
100ms out of 3s user time.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
With core.ignorecase=true, name-hash.c builds a case insensitive index of
all tracked directories. Currently, the existing cache entry structures are
added multiple times to the same hashtable (with different name lengths and
hash codes). However, there's only one dir_next pointer, which gets
completely messed up in case of hash collisions. In the worst case, this
causes an endless loop if ce == ce->dir_next (see t7062).
Use a separate hashtable and separate structures for the directory index
so that each directory entry has its own next pointer. Use reference
counting to track which directory entry contains files.
There are only slight changes to the name-hash.c API:
- new free_name_hash() used by read_cache.c::discard_index()
- remove_name_hash() takes an additional index_state parameter
- index_name_exists() for a directory (trailing '/') may return a cache
entry that has been removed (CE_UNHASHED). This is not a problem as the
return value is only used to check if the directory exists (dir.c) or to
normalize casing of directory names (read-cache.c).
Getting rid of cache_entry.dir_next reduces memory consumption, especially
with core.ignorecase=false (which doesn't use that member at all).
With core.ignorecase=true, building the directory index is slightly faster
as we add / check the parent directory first (instead of going through all
directory levels for each file in the index). E.g. with WebKit (~200k
files, ~7k dirs), time spent in lazy_init_name_hash is reduced from 176ms
to 130ms.
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Usually we do not pass an empty string to the function hash_name()
because we almost always ask for hash values for a path that is a
candidate to be added to the index. However, check-ignore (and most
likely check-attr, but I didn't check) apparently has a callchain
to ask the hash value for an empty path when it was given a "." from
the top-level directory to ask "Is the path . excluded by default?"
Make sure that hash_name() does not overrun the end of the given
pathname even when it is empty.
Remove a sweep-the-issue-under-the-rug conditional in check-ignore
that avoided to pass an empty string to the callchain while at it.
It is a valid question to ask for check-ignore if the top-level is
set to be ignored by default, even though the answer is most likely
no, if only because there is currently no way to specify such an
entry in the .gitignore file. But it is an unusual thing to ask and
it is not worth optimizing for it by special casing at the top level
of the call chain.
Signed-off-by: Adam Spiers <git@adamspiers.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Test t2021-checkout-overwrite.sh reveals a segfault in 'git add' on a
case-insensitive file system when git is compiled with XMALLOC_POISON
defined. The reason is that 2548183b (fix phantom untracked files when
core.ignorecase is set) added a new member dir_next to struct cache_entry,
but forgot to initialize it in all cases.
Signed-off-by: Johannes Sixt <j6t@kdbg.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When core.ignorecase is turned on and there are stale index
entries, "git commit" can sometimes report directories as
untracked, even though they contain tracked files.
You can see an example of this with:
# make a case-insensitive repo
git init repo && cd repo &&
git config core.ignorecase true &&
# with some tracked files in a subdir
mkdir subdir &&
> subdir/one &&
> subdir/two &&
git add . &&
git commit -m base &&
# now make the index entries stale
touch subdir/* &&
# and then ask commit to update those entries and show
# us the status template
git commit -a
which will report "subdir/" as untracked, even though it
clearly contains two tracked files. What is happening in the
commit program is this:
1. We load the index, and for each entry, insert it into the index's
name_hash. In addition, if ignorecase is turned on, we make an
entry in the name_hash for the directory (e.g., "contrib/"), which
uses the following code from 5102c61's hash_index_entry_directories:
hash = hash_name(ce->name, ptr - ce->name);
if (!lookup_hash(hash, &istate->name_hash)) {
pos = insert_hash(hash, &istate->name_hash);
if (pos) {
ce->next = *pos;
*pos = ce;
}
}
Note that we only add the directory entry if there is not already an
entry.
2. We run add_files_to_cache, which gets updated information for each
cache entry. It helpfully inserts this information into the cache,
which calls replace_index_entry. This in turn calls
remove_name_hash() on the old entry, and add_name_hash() on the new
one. But remove_name_hash doesn't actually remove from the hash, it
only marks it as "no longer interesting" (from cache.h):
/*
* We don't actually *remove* it, we can just mark it invalid so that
* we won't find it in lookups.
*
* Not only would we have to search the lists (simple enough), but
* we'd also have to rehash other hash buckets in case this makes the
* hash bucket empty (common). So it's much better to just mark
* it.
*/
static inline void remove_name_hash(struct cache_entry *ce)
{
ce->ce_flags |= CE_UNHASHED;
}
This is OK in the specific-file case, since the entries in the hash
form a linked list, and we can just skip the "not here anymore"
entries during lookup.
But for the directory hash entry, we will _not_ write a new entry,
because there is already one there: the old one that is actually no
longer interesting!
3. While traversing the directories, we end up in the
directory_exists_in_index_icase function to see if a directory is
interesting. This in turn checks index_name_exists, which will
look up the directory in the index's name_hash. We see the old,
deleted record, and assume there is nothing interesting. The
directory gets marked as untracked, even though there are index
entries in it.
The problem is in the code I showed above:
hash = hash_name(ce->name, ptr - ce->name);
if (!lookup_hash(hash, &istate->name_hash)) {
pos = insert_hash(hash, &istate->name_hash);
if (pos) {
ce->next = *pos;
*pos = ce;
}
}
Having a single cache entry that represents the directory is
not enough; that entry may go away if the index is changed.
It may be tempting to say that the problem is in our removal
method; if we removed the entry entirely instead of simply
marking it as "not here anymore", then we would know we need
to insert a new entry. But that only covers this particular
case of remove-replace. In the more general case, consider
something like this:
1. We add "foo/bar" and "foo/baz" to the index. Each gets
their own entry in name_hash, plus we make a "foo/"
entry that points to "foo/bar".
2. We remove the "foo/bar" entry from the index, and from
the name_hash.
3. We ask if "foo/" exists, and see no entry, even though
"foo/baz" exists.
So we need that directory entry to have the list of _all_
cache entries that indicate that the directory is tracked.
So that implies making a linked list as we do for other
entries, like:
hash = hash_name(ce->name, ptr - ce->name);
pos = insert_hash(hash, &istate->name_hash);
if (pos) {
ce->next = *pos;
*pos = ce;
}
But that's not right either. In fact, it shows a second bug
in the current code, which is that the "ce->next" pointer is
supposed to be linking entries for a specific filename
entry, but here we are overwriting it for the directory
entry. So the same cache entry ends up in two linked lists,
but they share the same "next" pointer.
As it turns out, this second bug can't be triggered in the
current code. The "if (pos)" conditional is totally dead
code; pos will only be non-NULL if there was an existing
hash entry, and we already checked that there wasn't one
through our call to lookup_hash.
But fixing the first bug means taking out that call to
lookup_hash, which is going to activate the buggy dead code,
and we'll end up splicing the two linked lists together.
So we need to have a separate next pointer for the list in
the directory bucket, and we need to traverse that list in
index_name_exists when we are looking up a directory.
This bloats "struct cache_entry" by a few bytes. Which is
annoying, because it's only necessary when core.ignorecase
is enabled. There's not an easy way around it, short of
separating out the "next" pointers from cache_entry entirely
(i.e., having a separate "cache_entry_list" struct that gets
stored in the name_hash). In practice, it probably doesn't
matter; we have thousands of cache entries, compared to the
millions of objects (where adding 4 bytes to the struct
actually does impact performance).
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When using a case preserving but case insensitive file system, directory
case can differ but still refer to the same physical directory. git
status reports the directory with the alternate case as an Untracked
file. (That is, when mydir/filea.txt is added to the repository and
then the directory on disk is renamed from mydir/ to MyDir/, git status
shows MyDir/ as being untracked.)
Support has been added in name-hash.c for hashing directories with a
terminating slash into the name hash. When index_name_exists() is called
with a directory (a name with a terminating slash), the name is not
found via the normal cache_name_compare() call, but it is found in the
slow_same_name() function.
Additionally, in dir.c, directory_exists_in_index_icase() allows newly
added directories deeper in the directory chain to be identified.
Ultimately, it would be better if the file list was read in case
insensitive alphabetical order from disk, but this change seems to
suffice for now.
The end result is the directory is looked up in a case insensitive
manner and does not show in the Untracked files list.
Signed-off-by: Joshua Jensen <jjensen@workspacewhiz.com>
Signed-off-by: Johannes Sixt <j6t@kdbg.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Right now nobody uses it, but "index_name_exists()" gets a flag so
you can enable it on a case-by-case basis.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This allows verify_absent() in unpack_trees() to use the hash chains
rather than looking it up using the binary search.
Perhaps more importantly, it's also going to be useful for the next phase,
where we actually start looking at the cache entry when we do
case-insensitive lookups and checking the result.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
It's really totally separate functionality, and if we want to start
doing case-insensitive hash lookups, I'd rather do it when it's
separated out.
It also renames "remove_index_entry()" to "remove_name_hash()", because
that really describes the thing better. It doesn't actually remove the
index entry, that's done by "remove_index_entry_at()", which is something
very different, despite the similarity in names.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
