git/read-cache.c

1397 строки
35 KiB
C
Исходник Обычный вид История

/*
* GIT - The information manager from hell
*
* Copyright (C) Linus Torvalds, 2005
*/
#define NO_THE_INDEX_COMPATIBILITY_MACROS
#include "cache.h"
#include "cache-tree.h"
#include "refs.h"
#include "dir.h"
/* Index extensions.
*
* The first letter should be 'A'..'Z' for extensions that are not
* necessary for a correct operation (i.e. optimization data).
* When new extensions are added that _needs_ to be understood in
* order to correctly interpret the index file, pick character that
* is outside the range, to cause the reader to abort.
*/
#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
#define CACHE_EXT_TREE 0x54524545 /* "TREE" */
struct index_state the_index;
static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
{
istate->cache[nr] = ce;
add_name_hash(istate, ce);
}
Create pathname-based hash-table lookup into index This creates a hash index of every single file added to the index. Right now that hash index isn't actually used for much: I implemented a "cache_name_exists()" function that uses it to efficiently look up a filename in the index without having to do the O(logn) binary search, but quite frankly, that's not why this patch is interesting. No, the whole and only reason to create the hash of the filenames in the index is that by modifying the hash function, you can fairly easily do things like making it always hash equivalent names into the same bucket. That, in turn, means that suddenly questions like "does this name exist in the index under an _equivalent_ name?" becomes much much cheaper. Guiding principles behind this patch: - it shouldn't be too costly. In fact, my primary goal here was to actually speed up "git commit" with a fully populated kernel tree, by being faster at checking whether a file already existed in the index. I did succeed, but only barely: Best before: [torvalds@woody linux]$ time git commit > /dev/null real 0m0.255s user 0m0.168s sys 0m0.088s Best after: [torvalds@woody linux]$ time ~/git/git commit > /dev/null real 0m0.233s user 0m0.144s sys 0m0.088s so some things are actually faster (~8%). Caveat: that's really the best case. Other things are invariably going to be slightly slower, since we populate that index cache, and quite frankly, few things really use it to look things up. That said, the cost is really quite small. The worst case is probably doing a "git ls-files", which will do very little except puopulate the index, and never actually looks anything up in it, just lists it. Before: [torvalds@woody linux]$ time git ls-files > /dev/null real 0m0.016s user 0m0.016s sys 0m0.000s After: [torvalds@woody linux]$ time ~/git/git ls-files > /dev/null real 0m0.021s user 0m0.012s sys 0m0.008s and while the thing has really gotten relatively much slower, we're still talking about something almost unmeasurable (eg 5ms). And that really should be pretty much the worst case. So we lose 5ms on one "benchmark", but win 22ms on another. Pick your poison - this patch has the advantage that it will _likely_ speed up the cases that are complex and expensive more than it slows down the cases that are already so fast that nobody cares. But if you look at relative speedups/slowdowns, it doesn't look so good. - It should be simple and clean The code may be a bit subtle (the reasons I do hash removal the way I do etc), but it re-uses the existing hash.c files, so it really is fairly small and straightforward apart from a few odd details. Now, this patch on its own doesn't really do much, but I think it's worth looking at, if only because if done correctly, the name hashing really can make an improvement to the whole issue of "do we have a filename that looks like this in the index already". And at least it gets real testing by being used even by default (ie there is a real use-case for it even without any insane filesystems). NOTE NOTE NOTE! The current hash is a joke. I'm ashamed of it, I'm just not ashamed of it enough to really care. I took all the numbers out of my nether regions - I'm sure it's good enough that it works in practice, but the whole point was that you can make a really much fancier hash that hashes characters not directly, but by their upper-case value or something like that, and thus you get a case-insensitive hash, while still keeping the name and the index itself totally case sensitive. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-01-23 05:41:14 +03:00
static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
{
struct cache_entry *old = istate->cache[nr];
remove_name_hash(old);
Fix name re-hashing semantics We handled the case of removing and re-inserting cache entries badly, which is something that merging commonly needs to do (removing the different stages, and then re-inserting one of them as the merged state). We even had a rather ugly special case for this failure case, where replace_index_entry() basically turned itself into a no-op if the new and the old entries were the same, exactly because the hash routines didn't handle it on their own. So what this patch does is to not just have the UNHASHED bit, but a HASHED bit too, and when you insert an entry into the name hash, that involves: - clear the UNHASHED bit, because now it's valid again for lookup (which is really all that UNHASHED meant) - if we're being lazy, we're done here (but we still want to clear the UNHASHED bit regardless of lazy mode, since we can become unlazy later, and so we need the UNHASHED bit to always be set correctly, even if we never actually insert the entry into the hash list) - if it was already hashed, we just leave it on the list - otherwise mark it HASHED and insert it into the list this all means that unhashing and rehashing a name all just works automatically. Obviously, you cannot change the name of an entry (that would be a serious bug), but nothing can validly do that anyway (you'd have to allocate a new struct cache_entry anyway since the name length could change), so that's not a new limitation. The code actually gets simpler in many ways, although the lazy hashing does mean that there are a few odd cases (ie something can be marked unhashed even though it was never on the hash in the first place, and isn't actually marked hashed!). Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-02-23 07:37:40 +03:00
set_index_entry(istate, nr, ce);
Create pathname-based hash-table lookup into index This creates a hash index of every single file added to the index. Right now that hash index isn't actually used for much: I implemented a "cache_name_exists()" function that uses it to efficiently look up a filename in the index without having to do the O(logn) binary search, but quite frankly, that's not why this patch is interesting. No, the whole and only reason to create the hash of the filenames in the index is that by modifying the hash function, you can fairly easily do things like making it always hash equivalent names into the same bucket. That, in turn, means that suddenly questions like "does this name exist in the index under an _equivalent_ name?" becomes much much cheaper. Guiding principles behind this patch: - it shouldn't be too costly. In fact, my primary goal here was to actually speed up "git commit" with a fully populated kernel tree, by being faster at checking whether a file already existed in the index. I did succeed, but only barely: Best before: [torvalds@woody linux]$ time git commit > /dev/null real 0m0.255s user 0m0.168s sys 0m0.088s Best after: [torvalds@woody linux]$ time ~/git/git commit > /dev/null real 0m0.233s user 0m0.144s sys 0m0.088s so some things are actually faster (~8%). Caveat: that's really the best case. Other things are invariably going to be slightly slower, since we populate that index cache, and quite frankly, few things really use it to look things up. That said, the cost is really quite small. The worst case is probably doing a "git ls-files", which will do very little except puopulate the index, and never actually looks anything up in it, just lists it. Before: [torvalds@woody linux]$ time git ls-files > /dev/null real 0m0.016s user 0m0.016s sys 0m0.000s After: [torvalds@woody linux]$ time ~/git/git ls-files > /dev/null real 0m0.021s user 0m0.012s sys 0m0.008s and while the thing has really gotten relatively much slower, we're still talking about something almost unmeasurable (eg 5ms). And that really should be pretty much the worst case. So we lose 5ms on one "benchmark", but win 22ms on another. Pick your poison - this patch has the advantage that it will _likely_ speed up the cases that are complex and expensive more than it slows down the cases that are already so fast that nobody cares. But if you look at relative speedups/slowdowns, it doesn't look so good. - It should be simple and clean The code may be a bit subtle (the reasons I do hash removal the way I do etc), but it re-uses the existing hash.c files, so it really is fairly small and straightforward apart from a few odd details. Now, this patch on its own doesn't really do much, but I think it's worth looking at, if only because if done correctly, the name hashing really can make an improvement to the whole issue of "do we have a filename that looks like this in the index already". And at least it gets real testing by being used even by default (ie there is a real use-case for it even without any insane filesystems). NOTE NOTE NOTE! The current hash is a joke. I'm ashamed of it, I'm just not ashamed of it enough to really care. I took all the numbers out of my nether regions - I'm sure it's good enough that it works in practice, but the whole point was that you can make a really much fancier hash that hashes characters not directly, but by their upper-case value or something like that, and thus you get a case-insensitive hash, while still keeping the name and the index itself totally case sensitive. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-01-23 05:41:14 +03:00
istate->cache_changed = 1;
}
/*
* This only updates the "non-critical" parts of the directory
* cache, ie the parts that aren't tracked by GIT, and only used
* to validate the cache.
*/
void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
{
ce->ce_ctime = st->st_ctime;
ce->ce_mtime = st->st_mtime;
ce->ce_dev = st->st_dev;
ce->ce_ino = st->st_ino;
ce->ce_uid = st->st_uid;
ce->ce_gid = st->st_gid;
ce->ce_size = st->st_size;
"Assume unchanged" git This adds "assume unchanged" logic, started by this message in the list discussion recently: <Pine.LNX.4.64.0601311807470.7301@g5.osdl.org> This is a workaround for filesystems that do not have lstat() that is quick enough for the index mechanism to take advantage of. On the paths marked as "assumed to be unchanged", the user needs to explicitly use update-index to register the object name to be in the next commit. You can use two new options to update-index to set and reset the CE_VALID bit: git-update-index --assume-unchanged path... git-update-index --no-assume-unchanged path... These forms manipulate only the CE_VALID bit; it does not change the object name recorded in the index file. Nor they add a new entry to the index. When the configuration variable "core.ignorestat = true" is set, the index entries are marked with CE_VALID bit automatically after: - update-index to explicitly register the current object name to the index file. - when update-index --refresh finds the path to be up-to-date. - when tools like read-tree -u and apply --index update the working tree file and register the current object name to the index file. The flag is dropped upon read-tree that does not check out the index entry. This happens regardless of the core.ignorestat settings. Index entries marked with CE_VALID bit are assumed to be unchanged most of the time. However, there are cases that CE_VALID bit is ignored for the sake of safety and usability: - while "git-read-tree -m" or git-apply need to make sure that the paths involved in the merge do not have local modifications. This sacrifices performance for safety. - when git-checkout-index -f -q -u -a tries to see if it needs to checkout the paths. Otherwise you can never check anything out ;-). - when git-update-index --really-refresh (a new flag) tries to see if the index entry is up to date. You can start with everything marked as CE_VALID and run this once to drop CE_VALID bit for paths that are modified. Most notably, "update-index --refresh" honours CE_VALID and does not actively stat, so after you modified a file in the working tree, update-index --refresh would not notice until you tell the index about it with "git-update-index path" or "git-update-index --no-assume-unchanged path". This version is not expected to be perfect. I think diff between index and/or tree and working files may need some adjustment, and there probably needs other cases we should automatically unmark paths that are marked to be CE_VALID. But the basics seem to work, and ready to be tested by people who asked for this feature. Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-02-09 08:15:24 +03:00
if (assume_unchanged)
ce->ce_flags |= CE_VALID;
if (S_ISREG(st->st_mode))
ce_mark_uptodate(ce);
}
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 11:02:15 +03:00
static int ce_compare_data(struct cache_entry *ce, struct stat *st)
{
int match = -1;
int fd = open(ce->name, O_RDONLY);
if (fd >= 0) {
unsigned char sha1[20];
if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
match = hashcmp(sha1, ce->sha1);
/* index_fd() closed the file descriptor already */
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 11:02:15 +03:00
}
return match;
}
static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 11:02:15 +03:00
{
int match = -1;
char *target;
void *buffer;
unsigned long size;
enum object_type type;
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 11:02:15 +03:00
int len;
target = xmalloc(expected_size);
len = readlink(ce->name, target, expected_size);
if (len != expected_size) {
free(target);
return -1;
}
buffer = read_sha1_file(ce->sha1, &type, &size);
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 11:02:15 +03:00
if (!buffer) {
free(target);
return -1;
}
if (size == expected_size)
match = memcmp(buffer, target, size);
free(buffer);
free(target);
return match;
}
static int ce_compare_gitlink(struct cache_entry *ce)
{
unsigned char sha1[20];
/*
* We don't actually require that the .git directory
* under GITLINK directory be a valid git directory. It
* might even be missing (in case nobody populated that
* sub-project).
*
* If so, we consider it always to match.
*/
if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
return 0;
return hashcmp(sha1, ce->sha1);
}
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 11:02:15 +03:00
static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
{
switch (st->st_mode & S_IFMT) {
case S_IFREG:
if (ce_compare_data(ce, st))
return DATA_CHANGED;
break;
case S_IFLNK:
if (ce_compare_link(ce, xsize_t(st->st_size)))
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 11:02:15 +03:00
return DATA_CHANGED;
break;
case S_IFDIR:
if (S_ISGITLINK(ce->ce_mode))
return 0;
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 11:02:15 +03:00
default:
return TYPE_CHANGED;
}
return 0;
}
static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
{
unsigned int changed = 0;
if (ce->ce_flags & CE_REMOVE)
return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
switch (ce->ce_mode & S_IFMT) {
case S_IFREG:
changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
/* We consider only the owner x bit to be relevant for
* "mode changes"
*/
if (trust_executable_bit &&
(0100 & (ce->ce_mode ^ st->st_mode)))
changed |= MODE_CHANGED;
break;
case S_IFLNK:
if (!S_ISLNK(st->st_mode) &&
(has_symlinks || !S_ISREG(st->st_mode)))
changed |= TYPE_CHANGED;
break;
case S_IFGITLINK:
if (!S_ISDIR(st->st_mode))
changed |= TYPE_CHANGED;
else if (ce_compare_gitlink(ce))
changed |= DATA_CHANGED;
return changed;
default:
die("internal error: ce_mode is %o", ce->ce_mode);
}
if (ce->ce_mtime != (unsigned int) st->st_mtime)
changed |= MTIME_CHANGED;
if (ce->ce_ctime != (unsigned int) st->st_ctime)
changed |= CTIME_CHANGED;
if (ce->ce_uid != (unsigned int) st->st_uid ||
ce->ce_gid != (unsigned int) st->st_gid)
changed |= OWNER_CHANGED;
if (ce->ce_ino != (unsigned int) st->st_ino)
changed |= INODE_CHANGED;
#ifdef USE_STDEV
/*
* st_dev breaks on network filesystems where different
* clients will have different views of what "device"
* the filesystem is on
*/
if (ce->ce_dev != (unsigned int) st->st_dev)
changed |= INODE_CHANGED;
#endif
if (ce->ce_size != (unsigned int) st->st_size)
changed |= DATA_CHANGED;
return changed;
}
static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
{
return (!S_ISGITLINK(ce->ce_mode) &&
istate->timestamp &&
((unsigned int)istate->timestamp) <= ce->ce_mtime);
}
int ie_match_stat(const struct index_state *istate,
struct cache_entry *ce, struct stat *st,
unsigned int options)
{
"Assume unchanged" git This adds "assume unchanged" logic, started by this message in the list discussion recently: <Pine.LNX.4.64.0601311807470.7301@g5.osdl.org> This is a workaround for filesystems that do not have lstat() that is quick enough for the index mechanism to take advantage of. On the paths marked as "assumed to be unchanged", the user needs to explicitly use update-index to register the object name to be in the next commit. You can use two new options to update-index to set and reset the CE_VALID bit: git-update-index --assume-unchanged path... git-update-index --no-assume-unchanged path... These forms manipulate only the CE_VALID bit; it does not change the object name recorded in the index file. Nor they add a new entry to the index. When the configuration variable "core.ignorestat = true" is set, the index entries are marked with CE_VALID bit automatically after: - update-index to explicitly register the current object name to the index file. - when update-index --refresh finds the path to be up-to-date. - when tools like read-tree -u and apply --index update the working tree file and register the current object name to the index file. The flag is dropped upon read-tree that does not check out the index entry. This happens regardless of the core.ignorestat settings. Index entries marked with CE_VALID bit are assumed to be unchanged most of the time. However, there are cases that CE_VALID bit is ignored for the sake of safety and usability: - while "git-read-tree -m" or git-apply need to make sure that the paths involved in the merge do not have local modifications. This sacrifices performance for safety. - when git-checkout-index -f -q -u -a tries to see if it needs to checkout the paths. Otherwise you can never check anything out ;-). - when git-update-index --really-refresh (a new flag) tries to see if the index entry is up to date. You can start with everything marked as CE_VALID and run this once to drop CE_VALID bit for paths that are modified. Most notably, "update-index --refresh" honours CE_VALID and does not actively stat, so after you modified a file in the working tree, update-index --refresh would not notice until you tell the index about it with "git-update-index path" or "git-update-index --no-assume-unchanged path". This version is not expected to be perfect. I think diff between index and/or tree and working files may need some adjustment, and there probably needs other cases we should automatically unmark paths that are marked to be CE_VALID. But the basics seem to work, and ready to be tested by people who asked for this feature. Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-02-09 08:15:24 +03:00
unsigned int changed;
int ignore_valid = options & CE_MATCH_IGNORE_VALID;
int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
"Assume unchanged" git This adds "assume unchanged" logic, started by this message in the list discussion recently: <Pine.LNX.4.64.0601311807470.7301@g5.osdl.org> This is a workaround for filesystems that do not have lstat() that is quick enough for the index mechanism to take advantage of. On the paths marked as "assumed to be unchanged", the user needs to explicitly use update-index to register the object name to be in the next commit. You can use two new options to update-index to set and reset the CE_VALID bit: git-update-index --assume-unchanged path... git-update-index --no-assume-unchanged path... These forms manipulate only the CE_VALID bit; it does not change the object name recorded in the index file. Nor they add a new entry to the index. When the configuration variable "core.ignorestat = true" is set, the index entries are marked with CE_VALID bit automatically after: - update-index to explicitly register the current object name to the index file. - when update-index --refresh finds the path to be up-to-date. - when tools like read-tree -u and apply --index update the working tree file and register the current object name to the index file. The flag is dropped upon read-tree that does not check out the index entry. This happens regardless of the core.ignorestat settings. Index entries marked with CE_VALID bit are assumed to be unchanged most of the time. However, there are cases that CE_VALID bit is ignored for the sake of safety and usability: - while "git-read-tree -m" or git-apply need to make sure that the paths involved in the merge do not have local modifications. This sacrifices performance for safety. - when git-checkout-index -f -q -u -a tries to see if it needs to checkout the paths. Otherwise you can never check anything out ;-). - when git-update-index --really-refresh (a new flag) tries to see if the index entry is up to date. You can start with everything marked as CE_VALID and run this once to drop CE_VALID bit for paths that are modified. Most notably, "update-index --refresh" honours CE_VALID and does not actively stat, so after you modified a file in the working tree, update-index --refresh would not notice until you tell the index about it with "git-update-index path" or "git-update-index --no-assume-unchanged path". This version is not expected to be perfect. I think diff between index and/or tree and working files may need some adjustment, and there probably needs other cases we should automatically unmark paths that are marked to be CE_VALID. But the basics seem to work, and ready to be tested by people who asked for this feature. Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-02-09 08:15:24 +03:00
/*
* If it's marked as always valid in the index, it's
* valid whatever the checked-out copy says.
*/
if (!ignore_valid && (ce->ce_flags & CE_VALID))
"Assume unchanged" git This adds "assume unchanged" logic, started by this message in the list discussion recently: <Pine.LNX.4.64.0601311807470.7301@g5.osdl.org> This is a workaround for filesystems that do not have lstat() that is quick enough for the index mechanism to take advantage of. On the paths marked as "assumed to be unchanged", the user needs to explicitly use update-index to register the object name to be in the next commit. You can use two new options to update-index to set and reset the CE_VALID bit: git-update-index --assume-unchanged path... git-update-index --no-assume-unchanged path... These forms manipulate only the CE_VALID bit; it does not change the object name recorded in the index file. Nor they add a new entry to the index. When the configuration variable "core.ignorestat = true" is set, the index entries are marked with CE_VALID bit automatically after: - update-index to explicitly register the current object name to the index file. - when update-index --refresh finds the path to be up-to-date. - when tools like read-tree -u and apply --index update the working tree file and register the current object name to the index file. The flag is dropped upon read-tree that does not check out the index entry. This happens regardless of the core.ignorestat settings. Index entries marked with CE_VALID bit are assumed to be unchanged most of the time. However, there are cases that CE_VALID bit is ignored for the sake of safety and usability: - while "git-read-tree -m" or git-apply need to make sure that the paths involved in the merge do not have local modifications. This sacrifices performance for safety. - when git-checkout-index -f -q -u -a tries to see if it needs to checkout the paths. Otherwise you can never check anything out ;-). - when git-update-index --really-refresh (a new flag) tries to see if the index entry is up to date. You can start with everything marked as CE_VALID and run this once to drop CE_VALID bit for paths that are modified. Most notably, "update-index --refresh" honours CE_VALID and does not actively stat, so after you modified a file in the working tree, update-index --refresh would not notice until you tell the index about it with "git-update-index path" or "git-update-index --no-assume-unchanged path". This version is not expected to be perfect. I think diff between index and/or tree and working files may need some adjustment, and there probably needs other cases we should automatically unmark paths that are marked to be CE_VALID. But the basics seem to work, and ready to be tested by people who asked for this feature. Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-02-09 08:15:24 +03:00
return 0;
changed = ce_match_stat_basic(ce, st);
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 11:02:15 +03:00
/*
* Within 1 second of this sequence:
* echo xyzzy >file && git-update-index --add file
* running this command:
* echo frotz >file
* would give a falsely clean cache entry. The mtime and
* length match the cache, and other stat fields do not change.
*
* We could detect this at update-index time (the cache entry
* being registered/updated records the same time as "now")
* and delay the return from git-update-index, but that would
* effectively mean we can make at most one commit per second,
* which is not acceptable. Instead, we check cache entries
* whose mtime are the same as the index file timestamp more
"Assume unchanged" git This adds "assume unchanged" logic, started by this message in the list discussion recently: <Pine.LNX.4.64.0601311807470.7301@g5.osdl.org> This is a workaround for filesystems that do not have lstat() that is quick enough for the index mechanism to take advantage of. On the paths marked as "assumed to be unchanged", the user needs to explicitly use update-index to register the object name to be in the next commit. You can use two new options to update-index to set and reset the CE_VALID bit: git-update-index --assume-unchanged path... git-update-index --no-assume-unchanged path... These forms manipulate only the CE_VALID bit; it does not change the object name recorded in the index file. Nor they add a new entry to the index. When the configuration variable "core.ignorestat = true" is set, the index entries are marked with CE_VALID bit automatically after: - update-index to explicitly register the current object name to the index file. - when update-index --refresh finds the path to be up-to-date. - when tools like read-tree -u and apply --index update the working tree file and register the current object name to the index file. The flag is dropped upon read-tree that does not check out the index entry. This happens regardless of the core.ignorestat settings. Index entries marked with CE_VALID bit are assumed to be unchanged most of the time. However, there are cases that CE_VALID bit is ignored for the sake of safety and usability: - while "git-read-tree -m" or git-apply need to make sure that the paths involved in the merge do not have local modifications. This sacrifices performance for safety. - when git-checkout-index -f -q -u -a tries to see if it needs to checkout the paths. Otherwise you can never check anything out ;-). - when git-update-index --really-refresh (a new flag) tries to see if the index entry is up to date. You can start with everything marked as CE_VALID and run this once to drop CE_VALID bit for paths that are modified. Most notably, "update-index --refresh" honours CE_VALID and does not actively stat, so after you modified a file in the working tree, update-index --refresh would not notice until you tell the index about it with "git-update-index path" or "git-update-index --no-assume-unchanged path". This version is not expected to be perfect. I think diff between index and/or tree and working files may need some adjustment, and there probably needs other cases we should automatically unmark paths that are marked to be CE_VALID. But the basics seem to work, and ready to be tested by people who asked for this feature. Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-02-09 08:15:24 +03:00
* carefully than others.
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 11:02:15 +03:00
*/
if (!changed && is_racy_timestamp(istate, ce)) {
if (assume_racy_is_modified)
changed |= DATA_CHANGED;
else
changed |= ce_modified_check_fs(ce, st);
}
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 11:02:15 +03:00
return changed;
}
int ie_modified(const struct index_state *istate,
struct cache_entry *ce, struct stat *st, unsigned int options)
{
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 11:02:15 +03:00
int changed, changed_fs;
changed = ie_match_stat(istate, ce, st, options);
if (!changed)
return 0;
/*
* If the mode or type has changed, there's no point in trying
* to refresh the entry - it's not going to match
*/
if (changed & (MODE_CHANGED | TYPE_CHANGED))
return changed;
/* Immediately after read-tree or update-index --cacheinfo,
* the length field is zero. For other cases the ce_size
* should match the SHA1 recorded in the index entry.
*/
if ((changed & DATA_CHANGED) && ce->ce_size != 0)
return changed;
Racy GIT This fixes the longstanding "Racy GIT" problem, which was pretty much there from the beginning of time, but was first demonstrated by Pasky in this message on October 24, 2005: http://marc.theaimsgroup.com/?l=git&m=113014629716878 If you run the following sequence of commands: echo frotz >infocom git update-index --add infocom echo xyzzy >infocom so that the second update to file "infocom" does not change st_mtime, what is recorded as the stat information for the cache entry "infocom" exactly matches what is on the filesystem (owner, group, inum, mtime, ctime, mode, length). After this sequence, we incorrectly think "infocom" file still has string "frotz" in it, and get really confused. E.g. git-diff-files would say there is no change, git-update-index --refresh would not even look at the filesystem to correct the situation. Some ways of working around this issue were already suggested by Linus in the same thread on the same day, including waiting until the next second before returning from update-index if a cache entry written out has the current timestamp, but that means we can make at most one commit per second, and given that the e-mail patch workflow used by Linus needs to process at least 5 commits per second, it is not an acceptable solution. Linus notes that git-apply is primarily used to update the index while processing e-mailed patches, which is true, and git-apply's up-to-date check is fooled by the same problem but luckily in the other direction, so it is not really a big issue, but still it is disturbing. The function ce_match_stat() is called to bypass the comparison against filesystem data when the stat data recorded in the cache entry matches what stat() returns from the filesystem. This patch tackles the problem by changing it to actually go to the filesystem data for cache entries that have the same mtime as the index file itself. This works as long as the index file and working tree files are on the filesystems that share the same monotonic clock. Files on network mounted filesystems sometimes get skewed timestamps compared to "date" output, but as long as working tree files' timestamps are skewed the same way as the index file's, this approach still works. The only problematic files are the ones that have the same timestamp as the index file's, because two file updates that sandwitch the index file update must happen within the same second to trigger the problem. Signed-off-by: Junio C Hamano <junkio@cox.net>
2005-12-20 11:02:15 +03:00
changed_fs = ce_modified_check_fs(ce, st);
if (changed_fs)
return changed | changed_fs;
return 0;
}
int base_name_compare(const char *name1, int len1, int mode1,
const char *name2, int len2, int mode2)
{
unsigned char c1, c2;
int len = len1 < len2 ? len1 : len2;
int cmp;
cmp = memcmp(name1, name2, len);
if (cmp)
return cmp;
c1 = name1[len];
c2 = name2[len];
if (!c1 && S_ISDIR(mode1))
c1 = '/';
if (!c2 && S_ISDIR(mode2))
c2 = '/';
return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
}
/*
* df_name_compare() is identical to base_name_compare(), except it
* compares conflicting directory/file entries as equal. Note that
* while a directory name compares as equal to a regular file, they
* then individually compare _differently_ to a filename that has
* a dot after the basename (because '\0' < '.' < '/').
*
* This is used by routines that want to traverse the git namespace
* but then handle conflicting entries together when possible.
*/
int df_name_compare(const char *name1, int len1, int mode1,
const char *name2, int len2, int mode2)
{
int len = len1 < len2 ? len1 : len2, cmp;
unsigned char c1, c2;
cmp = memcmp(name1, name2, len);
if (cmp)
return cmp;
/* Directories and files compare equal (same length, same name) */
if (len1 == len2)
return 0;
c1 = name1[len];
if (!c1 && S_ISDIR(mode1))
c1 = '/';
c2 = name2[len];
if (!c2 && S_ISDIR(mode2))
c2 = '/';
if (c1 == '/' && !c2)
return 0;
if (c2 == '/' && !c1)
return 0;
return c1 - c2;
}
int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
{
int len1 = flags1 & CE_NAMEMASK;
int len2 = flags2 & CE_NAMEMASK;
int len = len1 < len2 ? len1 : len2;
int cmp;
cmp = memcmp(name1, name2, len);
if (cmp)
return cmp;
if (len1 < len2)
return -1;
if (len1 > len2)
return 1;
"Assume unchanged" git This adds "assume unchanged" logic, started by this message in the list discussion recently: <Pine.LNX.4.64.0601311807470.7301@g5.osdl.org> This is a workaround for filesystems that do not have lstat() that is quick enough for the index mechanism to take advantage of. On the paths marked as "assumed to be unchanged", the user needs to explicitly use update-index to register the object name to be in the next commit. You can use two new options to update-index to set and reset the CE_VALID bit: git-update-index --assume-unchanged path... git-update-index --no-assume-unchanged path... These forms manipulate only the CE_VALID bit; it does not change the object name recorded in the index file. Nor they add a new entry to the index. When the configuration variable "core.ignorestat = true" is set, the index entries are marked with CE_VALID bit automatically after: - update-index to explicitly register the current object name to the index file. - when update-index --refresh finds the path to be up-to-date. - when tools like read-tree -u and apply --index update the working tree file and register the current object name to the index file. The flag is dropped upon read-tree that does not check out the index entry. This happens regardless of the core.ignorestat settings. Index entries marked with CE_VALID bit are assumed to be unchanged most of the time. However, there are cases that CE_VALID bit is ignored for the sake of safety and usability: - while "git-read-tree -m" or git-apply need to make sure that the paths involved in the merge do not have local modifications. This sacrifices performance for safety. - when git-checkout-index -f -q -u -a tries to see if it needs to checkout the paths. Otherwise you can never check anything out ;-). - when git-update-index --really-refresh (a new flag) tries to see if the index entry is up to date. You can start with everything marked as CE_VALID and run this once to drop CE_VALID bit for paths that are modified. Most notably, "update-index --refresh" honours CE_VALID and does not actively stat, so after you modified a file in the working tree, update-index --refresh would not notice until you tell the index about it with "git-update-index path" or "git-update-index --no-assume-unchanged path". This version is not expected to be perfect. I think diff between index and/or tree and working files may need some adjustment, and there probably needs other cases we should automatically unmark paths that are marked to be CE_VALID. But the basics seem to work, and ready to be tested by people who asked for this feature. Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-02-09 08:15:24 +03:00
/* Compare stages */
flags1 &= CE_STAGEMASK;
flags2 &= CE_STAGEMASK;
"Assume unchanged" git This adds "assume unchanged" logic, started by this message in the list discussion recently: <Pine.LNX.4.64.0601311807470.7301@g5.osdl.org> This is a workaround for filesystems that do not have lstat() that is quick enough for the index mechanism to take advantage of. On the paths marked as "assumed to be unchanged", the user needs to explicitly use update-index to register the object name to be in the next commit. You can use two new options to update-index to set and reset the CE_VALID bit: git-update-index --assume-unchanged path... git-update-index --no-assume-unchanged path... These forms manipulate only the CE_VALID bit; it does not change the object name recorded in the index file. Nor they add a new entry to the index. When the configuration variable "core.ignorestat = true" is set, the index entries are marked with CE_VALID bit automatically after: - update-index to explicitly register the current object name to the index file. - when update-index --refresh finds the path to be up-to-date. - when tools like read-tree -u and apply --index update the working tree file and register the current object name to the index file. The flag is dropped upon read-tree that does not check out the index entry. This happens regardless of the core.ignorestat settings. Index entries marked with CE_VALID bit are assumed to be unchanged most of the time. However, there are cases that CE_VALID bit is ignored for the sake of safety and usability: - while "git-read-tree -m" or git-apply need to make sure that the paths involved in the merge do not have local modifications. This sacrifices performance for safety. - when git-checkout-index -f -q -u -a tries to see if it needs to checkout the paths. Otherwise you can never check anything out ;-). - when git-update-index --really-refresh (a new flag) tries to see if the index entry is up to date. You can start with everything marked as CE_VALID and run this once to drop CE_VALID bit for paths that are modified. Most notably, "update-index --refresh" honours CE_VALID and does not actively stat, so after you modified a file in the working tree, update-index --refresh would not notice until you tell the index about it with "git-update-index path" or "git-update-index --no-assume-unchanged path". This version is not expected to be perfect. I think diff between index and/or tree and working files may need some adjustment, and there probably needs other cases we should automatically unmark paths that are marked to be CE_VALID. But the basics seem to work, and ready to be tested by people who asked for this feature. Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-02-09 08:15:24 +03:00
if (flags1 < flags2)
return -1;
if (flags1 > flags2)
return 1;
return 0;
}
int index_name_pos(const struct index_state *istate, const char *name, int namelen)
{
int first, last;
first = 0;
last = istate->cache_nr;
while (last > first) {
int next = (last + first) >> 1;
struct cache_entry *ce = istate->cache[next];
int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
if (!cmp)
return next;
if (cmp < 0) {
last = next;
continue;
}
first = next+1;
}
return -first-1;
}
/* Remove entry, return true if there are more entries to go.. */
int remove_index_entry_at(struct index_state *istate, int pos)
{
Create pathname-based hash-table lookup into index This creates a hash index of every single file added to the index. Right now that hash index isn't actually used for much: I implemented a "cache_name_exists()" function that uses it to efficiently look up a filename in the index without having to do the O(logn) binary search, but quite frankly, that's not why this patch is interesting. No, the whole and only reason to create the hash of the filenames in the index is that by modifying the hash function, you can fairly easily do things like making it always hash equivalent names into the same bucket. That, in turn, means that suddenly questions like "does this name exist in the index under an _equivalent_ name?" becomes much much cheaper. Guiding principles behind this patch: - it shouldn't be too costly. In fact, my primary goal here was to actually speed up "git commit" with a fully populated kernel tree, by being faster at checking whether a file already existed in the index. I did succeed, but only barely: Best before: [torvalds@woody linux]$ time git commit > /dev/null real 0m0.255s user 0m0.168s sys 0m0.088s Best after: [torvalds@woody linux]$ time ~/git/git commit > /dev/null real 0m0.233s user 0m0.144s sys 0m0.088s so some things are actually faster (~8%). Caveat: that's really the best case. Other things are invariably going to be slightly slower, since we populate that index cache, and quite frankly, few things really use it to look things up. That said, the cost is really quite small. The worst case is probably doing a "git ls-files", which will do very little except puopulate the index, and never actually looks anything up in it, just lists it. Before: [torvalds@woody linux]$ time git ls-files > /dev/null real 0m0.016s user 0m0.016s sys 0m0.000s After: [torvalds@woody linux]$ time ~/git/git ls-files > /dev/null real 0m0.021s user 0m0.012s sys 0m0.008s and while the thing has really gotten relatively much slower, we're still talking about something almost unmeasurable (eg 5ms). And that really should be pretty much the worst case. So we lose 5ms on one "benchmark", but win 22ms on another. Pick your poison - this patch has the advantage that it will _likely_ speed up the cases that are complex and expensive more than it slows down the cases that are already so fast that nobody cares. But if you look at relative speedups/slowdowns, it doesn't look so good. - It should be simple and clean The code may be a bit subtle (the reasons I do hash removal the way I do etc), but it re-uses the existing hash.c files, so it really is fairly small and straightforward apart from a few odd details. Now, this patch on its own doesn't really do much, but I think it's worth looking at, if only because if done correctly, the name hashing really can make an improvement to the whole issue of "do we have a filename that looks like this in the index already". And at least it gets real testing by being used even by default (ie there is a real use-case for it even without any insane filesystems). NOTE NOTE NOTE! The current hash is a joke. I'm ashamed of it, I'm just not ashamed of it enough to really care. I took all the numbers out of my nether regions - I'm sure it's good enough that it works in practice, but the whole point was that you can make a really much fancier hash that hashes characters not directly, but by their upper-case value or something like that, and thus you get a case-insensitive hash, while still keeping the name and the index itself totally case sensitive. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-01-23 05:41:14 +03:00
struct cache_entry *ce = istate->cache[pos];
remove_name_hash(ce);
istate->cache_changed = 1;
istate->cache_nr--;
if (pos >= istate->cache_nr)
return 0;
memmove(istate->cache + pos,
istate->cache + pos + 1,
(istate->cache_nr - pos) * sizeof(struct cache_entry *));
return 1;
}
int remove_file_from_index(struct index_state *istate, const char *path)
{
int pos = index_name_pos(istate, path, strlen(path));
if (pos < 0)
pos = -pos-1;
cache_tree_invalidate_path(istate->cache_tree, path);
while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
remove_index_entry_at(istate, pos);
return 0;
}
static int compare_name(struct cache_entry *ce, const char *path, int namelen)
{
return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
}
static int index_name_pos_also_unmerged(struct index_state *istate,
const char *path, int namelen)
{
int pos = index_name_pos(istate, path, namelen);
struct cache_entry *ce;
if (pos >= 0)
return pos;
/* maybe unmerged? */
pos = -1 - pos;
if (pos >= istate->cache_nr ||
compare_name((ce = istate->cache[pos]), path, namelen))
return -1;
/* order of preference: stage 2, 1, 3 */
if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
ce_stage((ce = istate->cache[pos + 1])) == 2 &&
!compare_name(ce, path, namelen))
pos++;
return pos;
}
static int different_name(struct cache_entry *ce, struct cache_entry *alias)
{
int len = ce_namelen(ce);
return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
}
/*
* If we add a filename that aliases in the cache, we will use the
* name that we already have - but we don't want to update the same
* alias twice, because that implies that there were actually two
* different files with aliasing names!
*
* So we use the CE_ADDED flag to verify that the alias was an old
* one before we accept it as
*/
static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
{
int len;
struct cache_entry *new;
if (alias->ce_flags & CE_ADDED)
die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
/* Ok, create the new entry using the name of the existing alias */
len = ce_namelen(alias);
new = xcalloc(1, cache_entry_size(len));
memcpy(new->name, alias->name, len);
copy_cache_entry(new, ce);
free(ce);
return new;
}
int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
{
int size, namelen, was_same;
mode_t st_mode = st->st_mode;
struct cache_entry *ce, *alias;
git-add: make the entry stat-clean after re-adding the same contents Earlier in commit 0781b8a9b2fe760fc4ed519a3a26e4b9bd6ccffe (add_file_to_index: skip rehashing if the cached stat already matches), add_file_to_index() were taught not to re-add the path if it already matches the index. The change meant well, but was not executed quite right. It used ie_modified() to see if the file on the work tree is really different from the index, and skipped adding the contents if the function says "not modified". This was wrong. There are three possible comparison results between the index and the file in the work tree: - with lstat(2) we _know_ they are different. E.g. if the length or the owner in the cached stat information is different from the length we just obtained from lstat(2), we can tell the file is modified without looking at the actual contents. - with lstat(2) we _know_ they are the same. The same length, the same owner, the same everything (but this has a twist, as described below). - we cannot tell from lstat(2) information alone and need to go to the filesystem to actually compare. The last case arises from what we call 'racy git' situation, that can be caused with this sequence: $ echo hello >file $ git add file $ echo aeiou >file ;# the same length If the second "echo" is done within the same filesystem timestamp granularity as the first "echo", then the timestamp recorded by "git add" and the timestamp we get from lstat(2) will be the same, and we can mistakenly say the file is not modified. The path is called 'racily clean'. We need to reliably detect racily clean paths are in fact modified. To solve this problem, when we write out the index, we mark the index entry that has the same timestamp as the index file itself (that is the time from the point of view of the filesystem) to tell any later code that does the lstat(2) comparison not to trust the cached stat info, and ie_modified() then actually goes to the filesystem to compare the contents for such a path. That's all good, but it should not be used for this "git add" optimization, as the goal of "git add" is to actually update the path in the index and make it stat-clean. With the false optimization, we did _not_ cause any data loss (after all, what we failed to do was only to update the cached stat information), but it made the following sequence leave the file stat dirty: $ echo hello >file $ git add file $ echo hello >file ;# the same contents $ git add file The solution is not to use ie_modified() which goes to the filesystem to see if it is really clean, but instead use ie_match_stat() with "assume racily clean paths are dirty" option, to force re-adding of such a path. There was another problem with "git add -u". The codepath shares the same issue when adding the paths that are found to be modified, but in addition, it asked "git diff-files" machinery run_diff_files() function (which is "git diff-files") to list the paths that are modified. But "git diff-files" machinery uses the same ie_modified() call so that it does not report racily clean _and_ actually clean paths as modified, which is not what we want. The patch allows the callers of run_diff_files() to pass the same "assume racily clean paths are dirty" option, and makes "git-add -u" codepath to use that option, to discover and re-add racily clean _and_ actually clean paths. We could further optimize on top of this patch to differentiate the case where the path really needs re-adding (i.e. the content of the racily clean entry was indeed different) and the case where only the cached stat information needs to be refreshed (i.e. the racily clean entry was actually clean), but I do not think it is worth it. This patch applies to maint and all the way up. Signed-off-by: Junio C Hamano <gitster@pobox.com>
2007-11-10 05:22:52 +03:00
unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
int pretend = flags & ADD_CACHE_PRETEND;
if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
return error("%s: can only add regular files, symbolic links or git-directories", path);
namelen = strlen(path);
if (S_ISDIR(st_mode)) {
while (namelen && path[namelen-1] == '/')
namelen--;
}
size = cache_entry_size(namelen);
ce = xcalloc(1, size);
memcpy(ce->name, path, namelen);
ce->ce_flags = namelen;
fill_stat_cache_info(ce, st);
if (trust_executable_bit && has_symlinks)
ce->ce_mode = create_ce_mode(st_mode);
else {
/* If there is an existing entry, pick the mode bits and type
* from it, otherwise assume unexecutable regular file.
*/
struct cache_entry *ent;
int pos = index_name_pos_also_unmerged(istate, path, namelen);
ent = (0 <= pos) ? istate->cache[pos] : NULL;
ce->ce_mode = ce_mode_from_stat(ent, st_mode);
}
alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
/* Nothing changed, really */
free(ce);
ce_mark_uptodate(alias);
alias->ce_flags |= CE_ADDED;
return 0;
}
if (index_path(ce->sha1, path, st, 1))
return error("unable to index file %s", path);
if (ignore_case && alias && different_name(ce, alias))
ce = create_alias_ce(ce, alias);
ce->ce_flags |= CE_ADDED;
/* It was suspected to be recily clean, but it turns out to be Ok */
was_same = (alias &&
!ce_stage(alias) &&
!hashcmp(alias->sha1, ce->sha1) &&
ce->ce_mode == alias->ce_mode);
if (pretend)
;
else if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
return error("unable to add %s to index",path);
if (verbose && !was_same)
printf("add '%s'\n", path);
return 0;
}
int add_file_to_index(struct index_state *istate, const char *path, int flags)
{
struct stat st;
if (lstat(path, &st))
die("%s: unable to stat (%s)", path, strerror(errno));
return add_to_index(istate, path, &st, flags);
}
struct cache_entry *make_cache_entry(unsigned int mode,
const unsigned char *sha1, const char *path, int stage,
int refresh)
{
int size, len;
struct cache_entry *ce;
if (!verify_path(path))
return NULL;
len = strlen(path);
size = cache_entry_size(len);
ce = xcalloc(1, size);
hashcpy(ce->sha1, sha1);
memcpy(ce->name, path, len);
ce->ce_flags = create_ce_flags(len, stage);
ce->ce_mode = create_ce_mode(mode);
if (refresh)
return refresh_cache_entry(ce, 0);
return ce;
}
int ce_same_name(struct cache_entry *a, struct cache_entry *b)
{
int len = ce_namelen(a);
return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
}
int ce_path_match(const struct cache_entry *ce, const char **pathspec)
{
const char *match, *name;
int len;
if (!pathspec)
return 1;
len = ce_namelen(ce);
name = ce->name;
while ((match = *pathspec++) != NULL) {
int matchlen = strlen(match);
if (matchlen > len)
continue;
if (memcmp(name, match, matchlen))
continue;
if (matchlen && name[matchlen-1] == '/')
return 1;
if (name[matchlen] == '/' || !name[matchlen])
return 1;
if (!matchlen)
return 1;
}
return 0;
}
/*
* We fundamentally don't like some paths: we don't want
* dot or dot-dot anywhere, and for obvious reasons don't
* want to recurse into ".git" either.
*
* Also, we don't want double slashes or slashes at the
* end that can make pathnames ambiguous.
*/
static int verify_dotfile(const char *rest)
{
/*
* The first character was '.', but that
* has already been discarded, we now test
* the rest.
*/
switch (*rest) {
/* "." is not allowed */
case '\0': case '/':
return 0;
/*
* ".git" followed by NUL or slash is bad. This
* shares the path end test with the ".." case.
*/
case 'g':
if (rest[1] != 'i')
break;
if (rest[2] != 't')
break;
rest += 2;
/* fallthrough */
case '.':
if (rest[1] == '\0' || rest[1] == '/')
return 0;
}
return 1;
}
int verify_path(const char *path)
{
char c;
goto inside;
for (;;) {
if (!c)
return 1;
if (c == '/') {
inside:
c = *path++;
switch (c) {
default:
continue;
case '/': case '\0':
break;
case '.':
if (verify_dotfile(path))
continue;
}
return 0;
}
c = *path++;
}
}
/*
* Do we have another file that has the beginning components being a
* proper superset of the name we're trying to add?
*/
static int has_file_name(struct index_state *istate,
const struct cache_entry *ce, int pos, int ok_to_replace)
{
int retval = 0;
int len = ce_namelen(ce);
int stage = ce_stage(ce);
const char *name = ce->name;
while (pos < istate->cache_nr) {
struct cache_entry *p = istate->cache[pos++];
if (len >= ce_namelen(p))
break;
if (memcmp(name, p->name, len))
break;
if (ce_stage(p) != stage)
continue;
if (p->name[len] != '/')
continue;
if (p->ce_flags & CE_REMOVE)
continue;
retval = -1;
if (!ok_to_replace)
break;
remove_index_entry_at(istate, --pos);
}
return retval;
}
/*
* Do we have another file with a pathname that is a proper
* subset of the name we're trying to add?
*/
static int has_dir_name(struct index_state *istate,
const struct cache_entry *ce, int pos, int ok_to_replace)
{
int retval = 0;
int stage = ce_stage(ce);
const char *name = ce->name;
const char *slash = name + ce_namelen(ce);
for (;;) {
int len;
for (;;) {
if (*--slash == '/')
break;
if (slash <= ce->name)
return retval;
}
len = slash - name;
pos = index_name_pos(istate, name, create_ce_flags(len, stage));
if (pos >= 0) {
/*
* Found one, but not so fast. This could
* be a marker that says "I was here, but
* I am being removed". Such an entry is
* not a part of the resulting tree, and
* it is Ok to have a directory at the same
* path.
*/
if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
retval = -1;
if (!ok_to_replace)
break;
remove_index_entry_at(istate, pos);
continue;
}
}
else
pos = -pos-1;
/*
* Trivial optimization: if we find an entry that
* already matches the sub-directory, then we know
* we're ok, and we can exit.
*/
while (pos < istate->cache_nr) {
struct cache_entry *p = istate->cache[pos];
if ((ce_namelen(p) <= len) ||
(p->name[len] != '/') ||
memcmp(p->name, name, len))
break; /* not our subdirectory */
if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
/*
* p is at the same stage as our entry, and
* is a subdirectory of what we are looking
* at, so we cannot have conflicts at our
* level or anything shorter.
*/
return retval;
pos++;
}
}
return retval;
}
/* We may be in a situation where we already have path/file and path
* is being added, or we already have path and path/file is being
* added. Either one would result in a nonsense tree that has path
* twice when git-write-tree tries to write it out. Prevent it.
*
* If ok-to-replace is specified, we remove the conflicting entries
* from the cache so the caller should recompute the insert position.
* When this happens, we return non-zero.
*/
static int check_file_directory_conflict(struct index_state *istate,
const struct cache_entry *ce,
int pos, int ok_to_replace)
{
int retval;
/*
* When ce is an "I am going away" entry, we allow it to be added
*/
if (ce->ce_flags & CE_REMOVE)
return 0;
/*
* We check if the path is a sub-path of a subsequent pathname
* first, since removing those will not change the position
* in the array.
*/
retval = has_file_name(istate, ce, pos, ok_to_replace);
/*
* Then check if the path might have a clashing sub-directory
* before it.
*/
return retval + has_dir_name(istate, ce, pos, ok_to_replace);
}
static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
{
int pos;
int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
"Assume unchanged" git This adds "assume unchanged" logic, started by this message in the list discussion recently: <Pine.LNX.4.64.0601311807470.7301@g5.osdl.org> This is a workaround for filesystems that do not have lstat() that is quick enough for the index mechanism to take advantage of. On the paths marked as "assumed to be unchanged", the user needs to explicitly use update-index to register the object name to be in the next commit. You can use two new options to update-index to set and reset the CE_VALID bit: git-update-index --assume-unchanged path... git-update-index --no-assume-unchanged path... These forms manipulate only the CE_VALID bit; it does not change the object name recorded in the index file. Nor they add a new entry to the index. When the configuration variable "core.ignorestat = true" is set, the index entries are marked with CE_VALID bit automatically after: - update-index to explicitly register the current object name to the index file. - when update-index --refresh finds the path to be up-to-date. - when tools like read-tree -u and apply --index update the working tree file and register the current object name to the index file. The flag is dropped upon read-tree that does not check out the index entry. This happens regardless of the core.ignorestat settings. Index entries marked with CE_VALID bit are assumed to be unchanged most of the time. However, there are cases that CE_VALID bit is ignored for the sake of safety and usability: - while "git-read-tree -m" or git-apply need to make sure that the paths involved in the merge do not have local modifications. This sacrifices performance for safety. - when git-checkout-index -f -q -u -a tries to see if it needs to checkout the paths. Otherwise you can never check anything out ;-). - when git-update-index --really-refresh (a new flag) tries to see if the index entry is up to date. You can start with everything marked as CE_VALID and run this once to drop CE_VALID bit for paths that are modified. Most notably, "update-index --refresh" honours CE_VALID and does not actively stat, so after you modified a file in the working tree, update-index --refresh would not notice until you tell the index about it with "git-update-index path" or "git-update-index --no-assume-unchanged path". This version is not expected to be perfect. I think diff between index and/or tree and working files may need some adjustment, and there probably needs other cases we should automatically unmark paths that are marked to be CE_VALID. But the basics seem to work, and ready to be tested by people who asked for this feature. Signed-off-by: Junio C Hamano <junkio@cox.net>
2006-02-09 08:15:24 +03:00
cache_tree_invalidate_path(istate->cache_tree, ce->name);
pos = index_name_pos(istate, ce->name, ce->ce_flags);
/* existing match? Just replace it. */
if (pos >= 0) {
Create pathname-based hash-table lookup into index This creates a hash index of every single file added to the index. Right now that hash index isn't actually used for much: I implemented a "cache_name_exists()" function that uses it to efficiently look up a filename in the index without having to do the O(logn) binary search, but quite frankly, that's not why this patch is interesting. No, the whole and only reason to create the hash of the filenames in the index is that by modifying the hash function, you can fairly easily do things like making it always hash equivalent names into the same bucket. That, in turn, means that suddenly questions like "does this name exist in the index under an _equivalent_ name?" becomes much much cheaper. Guiding principles behind this patch: - it shouldn't be too costly. In fact, my primary goal here was to actually speed up "git commit" with a fully populated kernel tree, by being faster at checking whether a file already existed in the index. I did succeed, but only barely: Best before: [torvalds@woody linux]$ time git commit > /dev/null real 0m0.255s user 0m0.168s sys 0m0.088s Best after: [torvalds@woody linux]$ time ~/git/git commit > /dev/null real 0m0.233s user 0m0.144s sys 0m0.088s so some things are actually faster (~8%). Caveat: that's really the best case. Other things are invariably going to be slightly slower, since we populate that index cache, and quite frankly, few things really use it to look things up. That said, the cost is really quite small. The worst case is probably doing a "git ls-files", which will do very little except puopulate the index, and never actually looks anything up in it, just lists it. Before: [torvalds@woody linux]$ time git ls-files > /dev/null real 0m0.016s user 0m0.016s sys 0m0.000s After: [torvalds@woody linux]$ time ~/git/git ls-files > /dev/null real 0m0.021s user 0m0.012s sys 0m0.008s and while the thing has really gotten relatively much slower, we're still talking about something almost unmeasurable (eg 5ms). And that really should be pretty much the worst case. So we lose 5ms on one "benchmark", but win 22ms on another. Pick your poison - this patch has the advantage that it will _likely_ speed up the cases that are complex and expensive more than it slows down the cases that are already so fast that nobody cares. But if you look at relative speedups/slowdowns, it doesn't look so good. - It should be simple and clean The code may be a bit subtle (the reasons I do hash removal the way I do etc), but it re-uses the existing hash.c files, so it really is fairly small and straightforward apart from a few odd details. Now, this patch on its own doesn't really do much, but I think it's worth looking at, if only because if done correctly, the name hashing really can make an improvement to the whole issue of "do we have a filename that looks like this in the index already". And at least it gets real testing by being used even by default (ie there is a real use-case for it even without any insane filesystems). NOTE NOTE NOTE! The current hash is a joke. I'm ashamed of it, I'm just not ashamed of it enough to really care. I took all the numbers out of my nether regions - I'm sure it's good enough that it works in practice, but the whole point was that you can make a really much fancier hash that hashes characters not directly, but by their upper-case value or something like that, and thus you get a case-insensitive hash, while still keeping the name and the index itself totally case sensitive. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-01-23 05:41:14 +03:00
replace_index_entry(istate, pos, ce);
return 0;
}
pos = -pos-1;
/*
* Inserting a merged entry ("stage 0") into the index
* will always replace all non-merged entries..
*/
if (pos < istate->cache_nr && ce_stage(ce) == 0) {
while (ce_same_name(istate->cache[pos], ce)) {
ok_to_add = 1;
if (!remove_index_entry_at(istate, pos))
break;
}
}
if (!ok_to_add)
return -1;
if (!verify_path(ce->name))
return -1;
if (!skip_df_check &&
check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
if (!ok_to_replace)
return error("'%s' appears as both a file and as a directory",
ce->name);
pos = index_name_pos(istate, ce->name, ce->ce_flags);
pos = -pos-1;
}
return pos + 1;
}
int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
{
int pos;
if (option & ADD_CACHE_JUST_APPEND)
pos = istate->cache_nr;
else {
int ret;
ret = add_index_entry_with_check(istate, ce, option);
if (ret <= 0)
return ret;
pos = ret - 1;
}
/* Make sure the array is big enough .. */
if (istate->cache_nr == istate->cache_alloc) {
istate->cache_alloc = alloc_nr(istate->cache_alloc);
istate->cache = xrealloc(istate->cache,
istate->cache_alloc * sizeof(struct cache_entry *));
}
/* Add it in.. */
istate->cache_nr++;
if (istate->cache_nr > pos + 1)
memmove(istate->cache + pos + 1,
istate->cache + pos,
(istate->cache_nr - pos - 1) * sizeof(ce));
Create pathname-based hash-table lookup into index This creates a hash index of every single file added to the index. Right now that hash index isn't actually used for much: I implemented a "cache_name_exists()" function that uses it to efficiently look up a filename in the index without having to do the O(logn) binary search, but quite frankly, that's not why this patch is interesting. No, the whole and only reason to create the hash of the filenames in the index is that by modifying the hash function, you can fairly easily do things like making it always hash equivalent names into the same bucket. That, in turn, means that suddenly questions like "does this name exist in the index under an _equivalent_ name?" becomes much much cheaper. Guiding principles behind this patch: - it shouldn't be too costly. In fact, my primary goal here was to actually speed up "git commit" with a fully populated kernel tree, by being faster at checking whether a file already existed in the index. I did succeed, but only barely: Best before: [torvalds@woody linux]$ time git commit > /dev/null real 0m0.255s user 0m0.168s sys 0m0.088s Best after: [torvalds@woody linux]$ time ~/git/git commit > /dev/null real 0m0.233s user 0m0.144s sys 0m0.088s so some things are actually faster (~8%). Caveat: that's really the best case. Other things are invariably going to be slightly slower, since we populate that index cache, and quite frankly, few things really use it to look things up. That said, the cost is really quite small. The worst case is probably doing a "git ls-files", which will do very little except puopulate the index, and never actually looks anything up in it, just lists it. Before: [torvalds@woody linux]$ time git ls-files > /dev/null real 0m0.016s user 0m0.016s sys 0m0.000s After: [torvalds@woody linux]$ time ~/git/git ls-files > /dev/null real 0m0.021s user 0m0.012s sys 0m0.008s and while the thing has really gotten relatively much slower, we're still talking about something almost unmeasurable (eg 5ms). And that really should be pretty much the worst case. So we lose 5ms on one "benchmark", but win 22ms on another. Pick your poison - this patch has the advantage that it will _likely_ speed up the cases that are complex and expensive more than it slows down the cases that are already so fast that nobody cares. But if you look at relative speedups/slowdowns, it doesn't look so good. - It should be simple and clean The code may be a bit subtle (the reasons I do hash removal the way I do etc), but it re-uses the existing hash.c files, so it really is fairly small and straightforward apart from a few odd details. Now, this patch on its own doesn't really do much, but I think it's worth looking at, if only because if done correctly, the name hashing really can make an improvement to the whole issue of "do we have a filename that looks like this in the index already". And at least it gets real testing by being used even by default (ie there is a real use-case for it even without any insane filesystems). NOTE NOTE NOTE! The current hash is a joke. I'm ashamed of it, I'm just not ashamed of it enough to really care. I took all the numbers out of my nether regions - I'm sure it's good enough that it works in practice, but the whole point was that you can make a really much fancier hash that hashes characters not directly, but by their upper-case value or something like that, and thus you get a case-insensitive hash, while still keeping the name and the index itself totally case sensitive. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-01-23 05:41:14 +03:00
set_index_entry(istate, pos, ce);
istate->cache_changed = 1;
return 0;
}
/*
* "refresh" does not calculate a new sha1 file or bring the
* cache up-to-date for mode/content changes. But what it
* _does_ do is to "re-match" the stat information of a file
* with the cache, so that you can refresh the cache for a
* file that hasn't been changed but where the stat entry is
* out of date.
*
* For example, you'd want to do this after doing a "git-read-tree",
* to link up the stat cache details with the proper files.
*/
static struct cache_entry *refresh_cache_ent(struct index_state *istate,
struct cache_entry *ce,
unsigned int options, int *err)
{
struct stat st;
struct cache_entry *updated;
int changed, size;
int ignore_valid = options & CE_MATCH_IGNORE_VALID;
if (ce_uptodate(ce))
return ce;
/*
* CE_VALID means the user promised us that the change to
* the work tree does not matter and told us not to worry.
*/
if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
ce_mark_uptodate(ce);
return ce;
}
if (lstat(ce->name, &st) < 0) {
if (err)
*err = errno;
return NULL;
}
changed = ie_match_stat(istate, ce, &st, options);
if (!changed) {
/*
* The path is unchanged. If we were told to ignore
* valid bit, then we did the actual stat check and
* found that the entry is unmodified. If the entry
* is not marked VALID, this is the place to mark it
* valid again, under "assume unchanged" mode.
*/
if (ignore_valid && assume_unchanged &&
!(ce->ce_flags & CE_VALID))
; /* mark this one VALID again */
else {
/*
* We do not mark the index itself "modified"
* because CE_UPTODATE flag is in-core only;
* we are not going to write this change out.
*/
ce_mark_uptodate(ce);
return ce;
}
}
if (ie_modified(istate, ce, &st, options)) {
if (err)
*err = EINVAL;
return NULL;
}
size = ce_size(ce);
updated = xmalloc(size);
memcpy(updated, ce, size);
fill_stat_cache_info(updated, &st);
/*
* If ignore_valid is not set, we should leave CE_VALID bit
* alone. Otherwise, paths marked with --no-assume-unchanged
* (i.e. things to be edited) will reacquire CE_VALID bit
* automatically, which is not really what we want.
*/
if (!ignore_valid && assume_unchanged &&
!(ce->ce_flags & CE_VALID))
updated->ce_flags &= ~CE_VALID;
return updated;
}
int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
{
int i;
int has_errors = 0;
int really = (flags & REFRESH_REALLY) != 0;
int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
int quiet = (flags & REFRESH_QUIET) != 0;
int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
for (i = 0; i < istate->cache_nr; i++) {
struct cache_entry *ce, *new;
int cache_errno = 0;
ce = istate->cache[i];
if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
continue;
if (ce_stage(ce)) {
while ((i < istate->cache_nr) &&
! strcmp(istate->cache[i]->name, ce->name))
i++;
i--;
if (allow_unmerged)
continue;
printf("%s: needs merge\n", ce->name);
has_errors = 1;
continue;
}
if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
continue;
new = refresh_cache_ent(istate, ce, options, &cache_errno);
if (new == ce)
continue;
if (!new) {
if (not_new && cache_errno == ENOENT)
continue;
if (really && cache_errno == EINVAL) {
/* If we are doing --really-refresh that
* means the index is not valid anymore.
*/
ce->ce_flags &= ~CE_VALID;
istate->cache_changed = 1;
}
if (quiet)
continue;
printf("%s: needs update\n", ce->name);
has_errors = 1;
continue;
}
Create pathname-based hash-table lookup into index This creates a hash index of every single file added to the index. Right now that hash index isn't actually used for much: I implemented a "cache_name_exists()" function that uses it to efficiently look up a filename in the index without having to do the O(logn) binary search, but quite frankly, that's not why this patch is interesting. No, the whole and only reason to create the hash of the filenames in the index is that by modifying the hash function, you can fairly easily do things like making it always hash equivalent names into the same bucket. That, in turn, means that suddenly questions like "does this name exist in the index under an _equivalent_ name?" becomes much much cheaper. Guiding principles behind this patch: - it shouldn't be too costly. In fact, my primary goal here was to actually speed up "git commit" with a fully populated kernel tree, by being faster at checking whether a file already existed in the index. I did succeed, but only barely: Best before: [torvalds@woody linux]$ time git commit > /dev/null real 0m0.255s user 0m0.168s sys 0m0.088s Best after: [torvalds@woody linux]$ time ~/git/git commit > /dev/null real 0m0.233s user 0m0.144s sys 0m0.088s so some things are actually faster (~8%). Caveat: that's really the best case. Other things are invariably going to be slightly slower, since we populate that index cache, and quite frankly, few things really use it to look things up. That said, the cost is really quite small. The worst case is probably doing a "git ls-files", which will do very little except puopulate the index, and never actually looks anything up in it, just lists it. Before: [torvalds@woody linux]$ time git ls-files > /dev/null real 0m0.016s user 0m0.016s sys 0m0.000s After: [torvalds@woody linux]$ time ~/git/git ls-files > /dev/null real 0m0.021s user 0m0.012s sys 0m0.008s and while the thing has really gotten relatively much slower, we're still talking about something almost unmeasurable (eg 5ms). And that really should be pretty much the worst case. So we lose 5ms on one "benchmark", but win 22ms on another. Pick your poison - this patch has the advantage that it will _likely_ speed up the cases that are complex and expensive more than it slows down the cases that are already so fast that nobody cares. But if you look at relative speedups/slowdowns, it doesn't look so good. - It should be simple and clean The code may be a bit subtle (the reasons I do hash removal the way I do etc), but it re-uses the existing hash.c files, so it really is fairly small and straightforward apart from a few odd details. Now, this patch on its own doesn't really do much, but I think it's worth looking at, if only because if done correctly, the name hashing really can make an improvement to the whole issue of "do we have a filename that looks like this in the index already". And at least it gets real testing by being used even by default (ie there is a real use-case for it even without any insane filesystems). NOTE NOTE NOTE! The current hash is a joke. I'm ashamed of it, I'm just not ashamed of it enough to really care. I took all the numbers out of my nether regions - I'm sure it's good enough that it works in practice, but the whole point was that you can make a really much fancier hash that hashes characters not directly, but by their upper-case value or something like that, and thus you get a case-insensitive hash, while still keeping the name and the index itself totally case sensitive. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-01-23 05:41:14 +03:00
replace_index_entry(istate, i, new);
}
return has_errors;
}
struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
{
return refresh_cache_ent(&the_index, ce, really, NULL);
}
static int verify_hdr(struct cache_header *hdr, unsigned long size)
{
SHA_CTX c;
unsigned char sha1[20];
if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
return error("bad signature");
if (hdr->hdr_version != htonl(2))
return error("bad index version");
SHA1_Init(&c);
SHA1_Update(&c, hdr, size - 20);
SHA1_Final(sha1, &c);
if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
return error("bad index file sha1 signature");
return 0;
}
static int read_index_extension(struct index_state *istate,
const char *ext, void *data, unsigned long sz)
{
switch (CACHE_EXT(ext)) {
case CACHE_EXT_TREE:
istate->cache_tree = cache_tree_read(data, sz);
break;
default:
if (*ext < 'A' || 'Z' < *ext)
return error("index uses %.4s extension, which we do not understand",
ext);
fprintf(stderr, "ignoring %.4s extension\n", ext);
break;
}
return 0;
}
int read_index(struct index_state *istate)
{
return read_index_from(istate, get_index_file());
}
static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
{
size_t len;
ce->ce_ctime = ntohl(ondisk->ctime.sec);
ce->ce_mtime = ntohl(ondisk->mtime.sec);
ce->ce_dev = ntohl(ondisk->dev);
ce->ce_ino = ntohl(ondisk->ino);
ce->ce_mode = ntohl(ondisk->mode);
ce->ce_uid = ntohl(ondisk->uid);
ce->ce_gid = ntohl(ondisk->gid);
ce->ce_size = ntohl(ondisk->size);
/* On-disk flags are just 16 bits */
ce->ce_flags = ntohs(ondisk->flags);
hashcpy(ce->sha1, ondisk->sha1);
len = ce->ce_flags & CE_NAMEMASK;
if (len == CE_NAMEMASK)
len = strlen(ondisk->name);
/*
* NEEDSWORK: If the original index is crafted, this copy could
* go unchecked.
*/
memcpy(ce->name, ondisk->name, len + 1);
}
Create pathname-based hash-table lookup into index This creates a hash index of every single file added to the index. Right now that hash index isn't actually used for much: I implemented a "cache_name_exists()" function that uses it to efficiently look up a filename in the index without having to do the O(logn) binary search, but quite frankly, that's not why this patch is interesting. No, the whole and only reason to create the hash of the filenames in the index is that by modifying the hash function, you can fairly easily do things like making it always hash equivalent names into the same bucket. That, in turn, means that suddenly questions like "does this name exist in the index under an _equivalent_ name?" becomes much much cheaper. Guiding principles behind this patch: - it shouldn't be too costly. In fact, my primary goal here was to actually speed up "git commit" with a fully populated kernel tree, by being faster at checking whether a file already existed in the index. I did succeed, but only barely: Best before: [torvalds@woody linux]$ time git commit > /dev/null real 0m0.255s user 0m0.168s sys 0m0.088s Best after: [torvalds@woody linux]$ time ~/git/git commit > /dev/null real 0m0.233s user 0m0.144s sys 0m0.088s so some things are actually faster (~8%). Caveat: that's really the best case. Other things are invariably going to be slightly slower, since we populate that index cache, and quite frankly, few things really use it to look things up. That said, the cost is really quite small. The worst case is probably doing a "git ls-files", which will do very little except puopulate the index, and never actually looks anything up in it, just lists it. Before: [torvalds@woody linux]$ time git ls-files > /dev/null real 0m0.016s user 0m0.016s sys 0m0.000s After: [torvalds@woody linux]$ time ~/git/git ls-files > /dev/null real 0m0.021s user 0m0.012s sys 0m0.008s and while the thing has really gotten relatively much slower, we're still talking about something almost unmeasurable (eg 5ms). And that really should be pretty much the worst case. So we lose 5ms on one "benchmark", but win 22ms on another. Pick your poison - this patch has the advantage that it will _likely_ speed up the cases that are complex and expensive more than it slows down the cases that are already so fast that nobody cares. But if you look at relative speedups/slowdowns, it doesn't look so good. - It should be simple and clean The code may be a bit subtle (the reasons I do hash removal the way I do etc), but it re-uses the existing hash.c files, so it really is fairly small and straightforward apart from a few odd details. Now, this patch on its own doesn't really do much, but I think it's worth looking at, if only because if done correctly, the name hashing really can make an improvement to the whole issue of "do we have a filename that looks like this in the index already". And at least it gets real testing by being used even by default (ie there is a real use-case for it even without any insane filesystems). NOTE NOTE NOTE! The current hash is a joke. I'm ashamed of it, I'm just not ashamed of it enough to really care. I took all the numbers out of my nether regions - I'm sure it's good enough that it works in practice, but the whole point was that you can make a really much fancier hash that hashes characters not directly, but by their upper-case value or something like that, and thus you get a case-insensitive hash, while still keeping the name and the index itself totally case sensitive. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-01-23 05:41:14 +03:00
static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
{
long per_entry;
per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
/*
* Alignment can cause differences. This should be "alignof", but
* since that's a gcc'ism, just use the size of a pointer.
*/
per_entry += sizeof(void *);
return ondisk_size + entries*per_entry;
}
/* remember to discard_cache() before reading a different cache! */
int read_index_from(struct index_state *istate, const char *path)
{
int fd, i;
struct stat st;
unsigned long src_offset, dst_offset;
struct cache_header *hdr;
void *mmap;
size_t mmap_size;
errno = EBUSY;
if (istate->alloc)
return istate->cache_nr;
errno = ENOENT;
istate->timestamp = 0;
fd = open(path, O_RDONLY);
if (fd < 0) {
if (errno == ENOENT)
return 0;
die("index file open failed (%s)", strerror(errno));
}
if (fstat(fd, &st))
die("cannot stat the open index (%s)", strerror(errno));
errno = EINVAL;
mmap_size = xsize_t(st.st_size);
if (mmap_size < sizeof(struct cache_header) + 20)
die("index file smaller than expected");
mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
close(fd);
if (mmap == MAP_FAILED)
die("unable to map index file");
hdr = mmap;
if (verify_hdr(hdr, mmap_size) < 0)
goto unmap;
istate->cache_nr = ntohl(hdr->hdr_entries);
istate->cache_alloc = alloc_nr(istate->cache_nr);
istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
/*
* The disk format is actually larger than the in-memory format,
* due to space for nsec etc, so even though the in-memory one
* has room for a few more flags, we can allocate using the same
* index size
*/
Create pathname-based hash-table lookup into index This creates a hash index of every single file added to the index. Right now that hash index isn't actually used for much: I implemented a "cache_name_exists()" function that uses it to efficiently look up a filename in the index without having to do the O(logn) binary search, but quite frankly, that's not why this patch is interesting. No, the whole and only reason to create the hash of the filenames in the index is that by modifying the hash function, you can fairly easily do things like making it always hash equivalent names into the same bucket. That, in turn, means that suddenly questions like "does this name exist in the index under an _equivalent_ name?" becomes much much cheaper. Guiding principles behind this patch: - it shouldn't be too costly. In fact, my primary goal here was to actually speed up "git commit" with a fully populated kernel tree, by being faster at checking whether a file already existed in the index. I did succeed, but only barely: Best before: [torvalds@woody linux]$ time git commit > /dev/null real 0m0.255s user 0m0.168s sys 0m0.088s Best after: [torvalds@woody linux]$ time ~/git/git commit > /dev/null real 0m0.233s user 0m0.144s sys 0m0.088s so some things are actually faster (~8%). Caveat: that's really the best case. Other things are invariably going to be slightly slower, since we populate that index cache, and quite frankly, few things really use it to look things up. That said, the cost is really quite small. The worst case is probably doing a "git ls-files", which will do very little except puopulate the index, and never actually looks anything up in it, just lists it. Before: [torvalds@woody linux]$ time git ls-files > /dev/null real 0m0.016s user 0m0.016s sys 0m0.000s After: [torvalds@woody linux]$ time ~/git/git ls-files > /dev/null real 0m0.021s user 0m0.012s sys 0m0.008s and while the thing has really gotten relatively much slower, we're still talking about something almost unmeasurable (eg 5ms). And that really should be pretty much the worst case. So we lose 5ms on one "benchmark", but win 22ms on another. Pick your poison - this patch has the advantage that it will _likely_ speed up the cases that are complex and expensive more than it slows down the cases that are already so fast that nobody cares. But if you look at relative speedups/slowdowns, it doesn't look so good. - It should be simple and clean The code may be a bit subtle (the reasons I do hash removal the way I do etc), but it re-uses the existing hash.c files, so it really is fairly small and straightforward apart from a few odd details. Now, this patch on its own doesn't really do much, but I think it's worth looking at, if only because if done correctly, the name hashing really can make an improvement to the whole issue of "do we have a filename that looks like this in the index already". And at least it gets real testing by being used even by default (ie there is a real use-case for it even without any insane filesystems). NOTE NOTE NOTE! The current hash is a joke. I'm ashamed of it, I'm just not ashamed of it enough to really care. I took all the numbers out of my nether regions - I'm sure it's good enough that it works in practice, but the whole point was that you can make a really much fancier hash that hashes characters not directly, but by their upper-case value or something like that, and thus you get a case-insensitive hash, while still keeping the name and the index itself totally case sensitive. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-01-23 05:41:14 +03:00
istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
src_offset = sizeof(*hdr);
dst_offset = 0;
for (i = 0; i < istate->cache_nr; i++) {
struct ondisk_cache_entry *disk_ce;
struct cache_entry *ce;
disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
convert_from_disk(disk_ce, ce);
Create pathname-based hash-table lookup into index This creates a hash index of every single file added to the index. Right now that hash index isn't actually used for much: I implemented a "cache_name_exists()" function that uses it to efficiently look up a filename in the index without having to do the O(logn) binary search, but quite frankly, that's not why this patch is interesting. No, the whole and only reason to create the hash of the filenames in the index is that by modifying the hash function, you can fairly easily do things like making it always hash equivalent names into the same bucket. That, in turn, means that suddenly questions like "does this name exist in the index under an _equivalent_ name?" becomes much much cheaper. Guiding principles behind this patch: - it shouldn't be too costly. In fact, my primary goal here was to actually speed up "git commit" with a fully populated kernel tree, by being faster at checking whether a file already existed in the index. I did succeed, but only barely: Best before: [torvalds@woody linux]$ time git commit > /dev/null real 0m0.255s user 0m0.168s sys 0m0.088s Best after: [torvalds@woody linux]$ time ~/git/git commit > /dev/null real 0m0.233s user 0m0.144s sys 0m0.088s so some things are actually faster (~8%). Caveat: that's really the best case. Other things are invariably going to be slightly slower, since we populate that index cache, and quite frankly, few things really use it to look things up. That said, the cost is really quite small. The worst case is probably doing a "git ls-files", which will do very little except puopulate the index, and never actually looks anything up in it, just lists it. Before: [torvalds@woody linux]$ time git ls-files > /dev/null real 0m0.016s user 0m0.016s sys 0m0.000s After: [torvalds@woody linux]$ time ~/git/git ls-files > /dev/null real 0m0.021s user 0m0.012s sys 0m0.008s and while the thing has really gotten relatively much slower, we're still talking about something almost unmeasurable (eg 5ms). And that really should be pretty much the worst case. So we lose 5ms on one "benchmark", but win 22ms on another. Pick your poison - this patch has the advantage that it will _likely_ speed up the cases that are complex and expensive more than it slows down the cases that are already so fast that nobody cares. But if you look at relative speedups/slowdowns, it doesn't look so good. - It should be simple and clean The code may be a bit subtle (the reasons I do hash removal the way I do etc), but it re-uses the existing hash.c files, so it really is fairly small and straightforward apart from a few odd details. Now, this patch on its own doesn't really do much, but I think it's worth looking at, if only because if done correctly, the name hashing really can make an improvement to the whole issue of "do we have a filename that looks like this in the index already". And at least it gets real testing by being used even by default (ie there is a real use-case for it even without any insane filesystems). NOTE NOTE NOTE! The current hash is a joke. I'm ashamed of it, I'm just not ashamed of it enough to really care. I took all the numbers out of my nether regions - I'm sure it's good enough that it works in practice, but the whole point was that you can make a really much fancier hash that hashes characters not directly, but by their upper-case value or something like that, and thus you get a case-insensitive hash, while still keeping the name and the index itself totally case sensitive. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-01-23 05:41:14 +03:00
set_index_entry(istate, i, ce);
src_offset += ondisk_ce_size(ce);
dst_offset += ce_size(ce);
}
istate->timestamp = st.st_mtime;
while (src_offset <= mmap_size - 20 - 8) {
/* After an array of active_nr index entries,
* there can be arbitrary number of extended
* sections, each of which is prefixed with
* extension name (4-byte) and section length
* in 4-byte network byte order.
*/
unsigned long extsize;
memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
extsize = ntohl(extsize);
if (read_index_extension(istate,
(const char *) mmap + src_offset,
(char *) mmap + src_offset + 8,
extsize) < 0)
goto unmap;
src_offset += 8;
src_offset += extsize;
}
munmap(mmap, mmap_size);
return istate->cache_nr;
unmap:
munmap(mmap, mmap_size);
errno = EINVAL;
die("index file corrupt");
}
int discard_index(struct index_state *istate)
{
istate->cache_nr = 0;
istate->cache_changed = 0;
istate->timestamp = 0;
Create pathname-based hash-table lookup into index This creates a hash index of every single file added to the index. Right now that hash index isn't actually used for much: I implemented a "cache_name_exists()" function that uses it to efficiently look up a filename in the index without having to do the O(logn) binary search, but quite frankly, that's not why this patch is interesting. No, the whole and only reason to create the hash of the filenames in the index is that by modifying the hash function, you can fairly easily do things like making it always hash equivalent names into the same bucket. That, in turn, means that suddenly questions like "does this name exist in the index under an _equivalent_ name?" becomes much much cheaper. Guiding principles behind this patch: - it shouldn't be too costly. In fact, my primary goal here was to actually speed up "git commit" with a fully populated kernel tree, by being faster at checking whether a file already existed in the index. I did succeed, but only barely: Best before: [torvalds@woody linux]$ time git commit > /dev/null real 0m0.255s user 0m0.168s sys 0m0.088s Best after: [torvalds@woody linux]$ time ~/git/git commit > /dev/null real 0m0.233s user 0m0.144s sys 0m0.088s so some things are actually faster (~8%). Caveat: that's really the best case. Other things are invariably going to be slightly slower, since we populate that index cache, and quite frankly, few things really use it to look things up. That said, the cost is really quite small. The worst case is probably doing a "git ls-files", which will do very little except puopulate the index, and never actually looks anything up in it, just lists it. Before: [torvalds@woody linux]$ time git ls-files > /dev/null real 0m0.016s user 0m0.016s sys 0m0.000s After: [torvalds@woody linux]$ time ~/git/git ls-files > /dev/null real 0m0.021s user 0m0.012s sys 0m0.008s and while the thing has really gotten relatively much slower, we're still talking about something almost unmeasurable (eg 5ms). And that really should be pretty much the worst case. So we lose 5ms on one "benchmark", but win 22ms on another. Pick your poison - this patch has the advantage that it will _likely_ speed up the cases that are complex and expensive more than it slows down the cases that are already so fast that nobody cares. But if you look at relative speedups/slowdowns, it doesn't look so good. - It should be simple and clean The code may be a bit subtle (the reasons I do hash removal the way I do etc), but it re-uses the existing hash.c files, so it really is fairly small and straightforward apart from a few odd details. Now, this patch on its own doesn't really do much, but I think it's worth looking at, if only because if done correctly, the name hashing really can make an improvement to the whole issue of "do we have a filename that looks like this in the index already". And at least it gets real testing by being used even by default (ie there is a real use-case for it even without any insane filesystems). NOTE NOTE NOTE! The current hash is a joke. I'm ashamed of it, I'm just not ashamed of it enough to really care. I took all the numbers out of my nether regions - I'm sure it's good enough that it works in practice, but the whole point was that you can make a really much fancier hash that hashes characters not directly, but by their upper-case value or something like that, and thus you get a case-insensitive hash, while still keeping the name and the index itself totally case sensitive. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2008-01-23 05:41:14 +03:00
free_hash(&istate->name_hash);
cache_tree_free(&(istate->cache_tree));
free(istate->alloc);
istate->alloc = NULL;
/* no need to throw away allocated active_cache */
return 0;
}
int unmerged_index(const struct index_state *istate)
{
int i;
for (i = 0; i < istate->cache_nr; i++) {
if (ce_stage(istate->cache[i]))
return 1;
}
return 0;
}
#define WRITE_BUFFER_SIZE 8192
static unsigned char write_buffer[WRITE_BUFFER_SIZE];
static unsigned long write_buffer_len;
static int ce_write_flush(SHA_CTX *context, int fd)
{
unsigned int buffered = write_buffer_len;
if (buffered) {
SHA1_Update(context, write_buffer, buffered);
if (write_in_full(fd, write_buffer, buffered) != buffered)
return -1;
write_buffer_len = 0;
}
return 0;
}
static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
{
while (len) {
unsigned int buffered = write_buffer_len;
unsigned int partial = WRITE_BUFFER_SIZE - buffered;
if (partial > len)
partial = len;
memcpy(write_buffer + buffered, data, partial);
buffered += partial;
if (buffered == WRITE_BUFFER_SIZE) {
write_buffer_len = buffered;
if (ce_write_flush(context, fd))
return -1;
buffered = 0;
}
write_buffer_len = buffered;
len -= partial;
data = (char *) data + partial;
}
return 0;
}
static int write_index_ext_header(SHA_CTX *context, int fd,
unsigned int ext, unsigned int sz)
{
ext = htonl(ext);
sz = htonl(sz);
return ((ce_write(context, fd, &ext, 4) < 0) ||
(ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
}
static int ce_flush(SHA_CTX *context, int fd)
{
unsigned int left = write_buffer_len;
if (left) {
write_buffer_len = 0;
SHA1_Update(context, write_buffer, left);
}
/* Flush first if not enough space for SHA1 signature */
if (left + 20 > WRITE_BUFFER_SIZE) {
if (write_in_full(fd, write_buffer, left) != left)
return -1;
left = 0;
}
/* Append the SHA1 signature at the end */
SHA1_Final(write_buffer + left, context);
left += 20;
return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
}
static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
{
/*
* The only thing we care about in this function is to smudge the
* falsely clean entry due to touch-update-touch race, so we leave
* everything else as they are. We are called for entries whose
* ce_mtime match the index file mtime.
*/
struct stat st;
if (lstat(ce->name, &st) < 0)
return;
if (ce_match_stat_basic(ce, &st))
return;
if (ce_modified_check_fs(ce, &st)) {
/* This is "racily clean"; smudge it. Note that this
* is a tricky code. At first glance, it may appear
* that it can break with this sequence:
*
* $ echo xyzzy >frotz
* $ git-update-index --add frotz
* $ : >frotz
* $ sleep 3
* $ echo filfre >nitfol
* $ git-update-index --add nitfol
*
* but it does not. When the second update-index runs,
* it notices that the entry "frotz" has the same timestamp
* as index, and if we were to smudge it by resetting its
* size to zero here, then the object name recorded
* in index is the 6-byte file but the cached stat information
* becomes zero --- which would then match what we would
* obtain from the filesystem next time we stat("frotz").
*
* However, the second update-index, before calling
* this function, notices that the cached size is 6
* bytes and what is on the filesystem is an empty
* file, and never calls us, so the cached size information
* for "frotz" stays 6 which does not match the filesystem.
*/
ce->ce_size = 0;
}
}
static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)
{
int size = ondisk_ce_size(ce);
struct ondisk_cache_entry *ondisk = xcalloc(1, size);
ondisk->ctime.sec = htonl(ce->ce_ctime);
ondisk->ctime.nsec = 0;
ondisk->mtime.sec = htonl(ce->ce_mtime);
ondisk->mtime.nsec = 0;
ondisk->dev = htonl(ce->ce_dev);
ondisk->ino = htonl(ce->ce_ino);
ondisk->mode = htonl(ce->ce_mode);
ondisk->uid = htonl(ce->ce_uid);
ondisk->gid = htonl(ce->ce_gid);
ondisk->size = htonl(ce->ce_size);
hashcpy(ondisk->sha1, ce->sha1);
ondisk->flags = htons(ce->ce_flags);
memcpy(ondisk->name, ce->name, ce_namelen(ce));
return ce_write(c, fd, ondisk, size);
}
int write_index(const struct index_state *istate, int newfd)
{
SHA_CTX c;
struct cache_header hdr;
int i, err, removed;
struct cache_entry **cache = istate->cache;
int entries = istate->cache_nr;
for (i = removed = 0; i < entries; i++)
if (cache[i]->ce_flags & CE_REMOVE)
removed++;
hdr.hdr_signature = htonl(CACHE_SIGNATURE);
hdr.hdr_version = htonl(2);
hdr.hdr_entries = htonl(entries - removed);
SHA1_Init(&c);
if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
return -1;
for (i = 0; i < entries; i++) {
struct cache_entry *ce = cache[i];
if (ce->ce_flags & CE_REMOVE)
continue;
if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
ce_smudge_racily_clean_entry(ce);
if (ce_write_entry(&c, newfd, ce) < 0)
return -1;
}
/* Write extension data here */
if (istate->cache_tree) {
struct strbuf sb;
strbuf_init(&sb, 0);
cache_tree_write(&sb, istate->cache_tree);
err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
|| ce_write(&c, newfd, sb.buf, sb.len) < 0;
strbuf_release(&sb);
if (err)
return -1;
}
return ce_flush(&c, newfd);
}