-
Notifications
You must be signed in to change notification settings - Fork 156
Fix a few split-index bugs #1497
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
This commit adds a new test case that demonstrates a bug in the split-index code that is triggered under certain circumstances when the FSMonitor is enabled, and its symptom manifests in the form of one of the following error messages: BUG: fsmonitor.c:20: fsmonitor_dirty has more entries than the index (2 > 1) BUG: unpack-trees.c:776: pos <n> doesn't point to the first entry of <dir>/ in index error: invalid path '' error: The following untracked working tree files would be overwritten by reset: initial.t Which of these error messages appears depends on timing-dependent conditions. Technically the root cause lies with a bug in the split-index code that has nothing to do with FSMonitor, but for the sake of this new test case it was the easiest way to trigger the bug. The bug is this: Under specific conditions, Git needs to skip writing the "link" extension (which is the index extension containing the information pertaining to the split-index). To do that, the `base_oid` attribute of the `split_index` structure in the in-memory index is zeroed out, and `do_write_index()` specifically checks for a "null" `base_oid` to understand that the "link" extension should not be written. However, this violates the consistency of the in-memory index structure, but that does not cause problems in most cases because the process exits without using the in-memory index structure anymore, anyway. But: _When_ the in-memory index is still used (which is the case e.g. in `git rebase`), subsequent writes of `the_index` are at risk of writing out a bogus index file, one that _should_ have a "link" extension but does not. In many cases, the `SPLIT_INDEX_ORDERED` flag _happens_ to be set for subsequent writes, forcing the shared index to be written, which re-initializes `base_oid` to a non-bogus state, and all is good. When it is _not_ set, however, all kinds of mayhem ensue, resulting in above-mentioned error messages, and often enough putting worktrees in a totally broken state where the only recourse is to manually delete the `index` and the `index.lock` files and then call `git reset` manually. Not something to ask users to do. The reason why it is comparatively easy to trigger the bug with FSMonitor is that there is _another_ bug in the FSMonitor code: `mark_fsmonitor_valid()` sets `cache_changed` to 1, i.e. treating that variable as a Boolean. But it is a bit field, and 1 happens to be the `SOMETHING_CHANGED` bit that forces the "link" extension to be skipped when writing the index, among other things. "Comparatively easy" is a relative term in this context, for sure. The essence of how the new test case triggers the bug is as following: 1. The `git rebase` invocation will first reset the worktree to a commit that contains only the `one.t` file, and then execute a rebase script that starts with the following commands (commit hashes skipped): label onto reset initial pick two label two reset two pick three [...] 2. Before executing the `label` command, a split index is written, as well as the shared index. 3. The `reset initial` command in the rebase script writes out a new split index but skips writing the shared index, as intended. 4. The `pick two` command updates the worktree and refreshes the index, marking the `two.t` entry as valid via the FSMonitor, which sets the `SOMETHING_CHANGED` bit in `cache_changed`, which in turn causes the `base_oid` attribute to be zeroed out and a full (non-split) index to be written (making sure _not_ to write the "link" extension). 5. Now, the `reset two` command will leave the worktree alone, but still write out a new split index, not writing the shared index (because `base_oid` is still zeroed out, and there is no index entry update requiring it to be written, either). 6. When it is turn to run `pick three`, the index is read, but it is too short: It only contains a single entry when there should be two, because the "link" extension is missing from the written-out index file. There are three bugs at play, actually, which will be fixed over the course of the next commits: - The `base_oid` attribute should not be zeroed out to indicate when the "link" extension should not be written, as it puts the in-memory index structure into an inconsistent state. - The FSMonitor should not overwrite bits in `cache_changed`. - The `unpack_trees()` function tries to reuse the `split_index` structure from the source index, if any, but does not propagate the `SPLIT_INDEX_ORDERED` flag. While a fix for the second bug would let this test case pass, there are other conditions where the `SOMETHING_CHANGED` bit is set. Therefore, the bug that most crucially needs to be fixed is the first one. Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Note: The build failures are a well-known issue with the target branch, |
Even though you're based on |
TBH I had not even thought about checking for conflicts with newer branches, that you for reminding me ;-) Turns out there are trivial merge conflicts with |
/submit |
Submitted as pull.1497.git.1679500859.gitgitgadget@gmail.com To fetch this version into
To fetch this version to local tag
|
On the Git mailing list, Junio C Hamano wrote (reply to this): "Johannes Schindelin via GitGitGadget" <gitgitgadget@gmail.com>
writes:
> Note: While the Git maintainer has stated a strong preference to introduce
> regression tests in the same patch that fixes the corresponding regression,
> this patch series starts with a stand-alone patch that demonstrates a
> problematic scenario via a new test_expect_failure test case.
It is fine, especially to show existing/old bugs that need extensive
explanation.
> This patch series is based on maint-2.37, the oldest maintenance branch it
> applies without merge conflicts. When merging with next, there are only
> trivial conflicts in unpack-trees.c due to en/dir-api-cleanup where
> o->result is now o->internal.result.
Thanks for digging into old and important case. Maintenance of the
index data structure is a crucial part of the health of the system. |
* written by default if the user explicitly requested | ||
* threaded index reads. | ||
*/ | ||
return !git_config_get_index_threads(&val) && val != 1; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
On the Git mailing list, Junio C Hamano wrote (reply to this):
"Johannes Schindelin via GitGitGadget" <gitgitgadget@gmail.com>
writes:
> From: Johannes Schindelin <johannes.schindelin@gmx.de>
>
> When a split-index is in effect, the `$GIT_DIR/index` file needs to
> contain a "link" extension that contains all the information about the
> split-index, including the information about the shared index.
> ...
> Let's stop zeroing out the `base_oid` to indicate that the "link"
> extension should not be written.
Nicely explained.
> One might be tempted to simply call `discard_split_index()` instead,
> under the assumption that Git decided to write a non-split index and
> therefore the the `split_index` structure might no longer be wanted.
"the the".
> +enum strip_extensions {
> + WRITE_ALL_EXTENSIONS = 0,
> + STRIP_ALL_EXTENSIONS = 1,
> + STRIP_LINK_EXTENSION_ONLY = 2
> +};
We do not need to spell out the specific values for this enum; the
users' (i.e. the callers of do_write_index()) sole requirement is
for these symbols to have different values.
Also do we envision that (1) we would need to keep STRIP_LINK_ONLY
to be with the largest value among the enum values, or (2) we would
never add new value to the set? Otherwise let's end the last one
with a trailing comma.
Looking at the way strip_extensions variable is used in
do_write_index(), an alternative design might be to make it a set of
bits (e.g. unsigned write_extension) and give one bit to each
extension. But such a clean-up is better left outside the topic, I
would imagine, as we do not have any need to skip an arbitrary set
of extensions right now.
> +/*
> + * Write the Git index into a `.lock` file
> + *
> + * If `strip_link_extension` is non-zero, avoid writing any "link" extension
> + * (used by the split-index feature).
> + */
Not exposing "enum strip_extensions" to the caller of this function,
like this patch does, is probably a very safe and sensible thing to
do. We do not have a reason to allow its callers to (perhaps
mistakenly) pass STRIP_ALL_EXTENSIONS to it.
> static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
> - unsigned flags)
> + unsigned flags, int strip_link_extension)
> {
> int ret;
> int was_full = istate->sparse_index == INDEX_EXPANDED;
> @@ -3185,7 +3197,7 @@ static int do_write_locked_index(struct index_state *istate, struct lock_file *l
> */
> trace2_region_enter_printf("index", "do_write_index", the_repository,
> "%s", get_lock_file_path(lock));
> - ret = do_write_index(istate, lock->tempfile, 0, flags);
> + ret = do_write_index(istate, lock->tempfile, strip_link_extension ? STRIP_LINK_EXTENSION_ONLY : 0, flags);
> trace2_region_leave_printf("index", "do_write_index", the_repository,
> "%s", get_lock_file_path(lock));
>
OK.
Very nicely done.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
On the Git mailing list, Jeff Hostetler wrote (reply to this):
On 3/22/23 5:24 PM, Junio C Hamano wrote:
> "Johannes Schindelin via GitGitGadget" <gitgitgadget@gmail.com>
> writes:
> >> From: Johannes Schindelin <johannes.schindelin@gmx.de>
>>
>> When a split-index is in effect, the `$GIT_DIR/index` file needs to
>> contain a "link" extension that contains all the information about the
>> split-index, including the information about the shared index.
>> ...
>> Let's stop zeroing out the `base_oid` to indicate that the "link"
>> extension should not be written.
> > Nicely explained.
> >> One might be tempted to simply call `discard_split_index()` instead,
>> under the assumption that Git decided to write a non-split index and
>> therefore the the `split_index` structure might no longer be wanted.
> > "the the".
> >> +enum strip_extensions {
>> + WRITE_ALL_EXTENSIONS = 0,
>> + STRIP_ALL_EXTENSIONS = 1,
>> + STRIP_LINK_EXTENSION_ONLY = 2
>> +};
> > We do not need to spell out the specific values for this enum; the
> users' (i.e. the callers of do_write_index()) sole requirement is
> for these symbols to have different values.
There are several calls to do_write_locked_index() that pass 0 or 1
as the new final arg. If we update them to use these enum values,
then we don't need integer values here.
> > Also do we envision that (1) we would need to keep STRIP_LINK_ONLY
> to be with the largest value among the enum values, or (2) we would
> never add new value to the set? Otherwise let's end the last one
> with a trailing comma.
> > Looking at the way strip_extensions variable is used in
> do_write_index(), an alternative design might be to make it a set of
> bits (e.g. unsigned write_extension) and give one bit to each
> extension. But such a clean-up is better left outside the topic, I
> would imagine, as we do not have any need to skip an arbitrary set
> of extensions right now.
Agreed, I thought about suggesting a set of bits too, but right now
we only need to strip all of them or just this one.
> >> +/*
>> + * Write the Git index into a `.lock` file
>> + *
>> + * If `strip_link_extension` is non-zero, avoid writing any "link" extension
>> + * (used by the split-index feature).
>> + */
> > Not exposing "enum strip_extensions" to the caller of this function,
> like this patch does, is probably a very safe and sensible thing to
> do. We do not have a reason to allow its callers to (perhaps
> mistakenly) pass STRIP_ALL_EXTENSIONS to it.
> >> static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
>> - unsigned flags)
>> + unsigned flags, int strip_link_extension)
>> {
>> int ret;
>> int was_full = istate->sparse_index == INDEX_EXPANDED;
>> @@ -3185,7 +3197,7 @@ static int do_write_locked_index(struct index_state *istate, struct lock_file *l
>> */
>> trace2_region_enter_printf("index", "do_write_index", the_repository,
>> "%s", get_lock_file_path(lock));
>> - ret = do_write_index(istate, lock->tempfile, 0, flags);
>> + ret = do_write_index(istate, lock->tempfile, strip_link_extension ? STRIP_LINK_EXTENSION_ONLY : 0, flags);
In the else of the ?: operator, could we use the WRITE_ALL_EXTENSIONS
instead of 0?
>> trace2_region_leave_printf("index", "do_write_index", the_repository,
>> "%s", get_lock_file_path(lock));
>> > > OK.
> > Very nicely done.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
On the Git mailing list, Junio C Hamano wrote (reply to this):
Jeff Hostetler <git@jeffhostetler.com> writes:
>>> +enum strip_extensions {
>>> + WRITE_ALL_EXTENSIONS = 0,
>>> + STRIP_ALL_EXTENSIONS = 1,
>>> + STRIP_LINK_EXTENSION_ONLY = 2
>>> +};
>> We do not need to spell out the specific values for this enum; the
>> users' (i.e. the callers of do_write_index()) sole requirement is
>> for these symbols to have different values.
>
> There are several calls to do_write_locked_index() that pass 0 or 1
> as the new final arg. If we update them to use these enum values,
> then we don't need integer values here.
Good eyes. Yes, the new caller that selectively passes
STRIP_LINK_EXTENSION_ONLY should pass WRITE_ALL_EXTENSIONS, not 0,
on the other side of ?: as you pointed out.
Thanks.
This branch is now known as |
This patch series was integrated into seen via git@12567f3. |
There was a status update in the "New Topics" section about the branch The index files can become corrupt under certain conditions when the split-index feature is in use, especially together with fsmonitor, which have been corrected. Will merge to 'next'? source: <pull.1497.git.1679500859.gitgitgadget@gmail.com> |
There are issues in commit acbe657: |
# We should have seen explicit NFD from OS. | ||
# We should have synthesized an NFC event. | ||
egrep "^event: nfd/d_${utf8_nfd}/?$" ./unicode.trace && | ||
egrep "^event: nfd/d_${utf8_nfc}/?$" ./unicode.trace |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
On the Git mailing list, Jeff Hostetler wrote (reply to this):
On 3/22/23 12:00 PM, Johannes Schindelin via GitGitGadget wrote:
> From: Johannes Schindelin <johannes.schindelin@gmx.de>
> > This commit adds a new test case that demonstrates a bug in the
> split-index code that is triggered under certain circumstances when the
> FSMonitor is enabled, and its symptom manifests in the form of one of
> the following error messages:
> > BUG: fsmonitor.c:20: fsmonitor_dirty has more entries than the index (2 > 1)
> > BUG: unpack-trees.c:776: pos <n> doesn't point to the first entry of <dir>/ in index
> > error: invalid path ''
> error: The following untracked working tree files would be overwritten by reset:
> initial.t
> > Which of these error messages appears depends on timing-dependent
> conditions.
> > Technically the root cause lies with a bug in the split-index code that
> has nothing to do with FSMonitor, but for the sake of this new test case
> it was the easiest way to trigger the bug.
> > The bug is this: Under specific conditions, Git needs to skip writing
> the "link" extension (which is the index extension containing the
> information pertaining to the split-index). To do that, the `base_oid`
> attribute of the `split_index` structure in the in-memory index is
> zeroed out, and `do_write_index()` specifically checks for a "null"
> `base_oid` to understand that the "link" extension should not be
> written. However, this violates the consistency of the in-memory index
> structure, but that does not cause problems in most cases because the
> process exits without using the in-memory index structure anymore,
> anyway.
> > But: _When_ the in-memory index is still used (which is the case e.g. in
> `git rebase`), subsequent writes of `the_index` are at risk of writing
> out a bogus index file, one that _should_ have a "link" extension but
> does not. In many cases, the `SPLIT_INDEX_ORDERED` flag _happens_ to be
> set for subsequent writes, forcing the shared index to be written, which
> re-initializes `base_oid` to a non-bogus state, and all is good.
> > When it is _not_ set, however, all kinds of mayhem ensue, resulting in
> above-mentioned error messages, and often enough putting worktrees in a
> totally broken state where the only recourse is to manually delete the
> `index` and the `index.lock` files and then call `git reset` manually.
> Not something to ask users to do.
> > The reason why it is comparatively easy to trigger the bug with
> FSMonitor is that there is _another_ bug in the FSMonitor code:
> `mark_fsmonitor_valid()` sets `cache_changed` to 1, i.e. treating that
> variable as a Boolean. But it is a bit field, and 1 happens to be the
> `SOMETHING_CHANGED` bit that forces the "link" extension to be skipped
> when writing the index, among other things.
> > "Comparatively easy" is a relative term in this context, for sure. The
> essence of how the new test case triggers the bug is as following:
> > 1. The `git rebase` invocation will first reset the worktree to
> a commit that contains only the `one.t` file, and then execute a
> rebase script that starts with the following commands (commit hashes
> skipped):
> > label onto
> > reset initial
> pick two
> label two
> > reset two
> pick three
> [...]
> > 2. Before executing the `label` command, a split index is written, as
> well as the shared index.
> > 3. The `reset initial` command in the rebase script writes out a new
> split index but skips writing the shared index, as intended.
> > 4. The `pick two` command updates the worktree and refreshes the index,
> marking the `two.t` entry as valid via the FSMonitor, which sets the
> `SOMETHING_CHANGED` bit in `cache_changed`, which in turn causes the
> `base_oid` attribute to be zeroed out and a full (non-split) index
> to be written (making sure _not_ to write the "link" extension).
> > 5. Now, the `reset two` command will leave the worktree alone, but
> still write out a new split index, not writing the shared index
> (because `base_oid` is still zeroed out, and there is no index entry
> update requiring it to be written, either).
> > 6. When it is turn to run `pick three`, the index is read, but it is
> too short: It only contains a single entry when there should be two,
> because the "link" extension is missing from the written-out index
> file.
> > There are three bugs at play, actually, which will be fixed over the
> course of the next commits:
> > - The `base_oid` attribute should not be zeroed out to indicate when
> the "link" extension should not be written, as it puts the in-memory
> index structure into an inconsistent state.
> > - The FSMonitor should not overwrite bits in `cache_changed`.
> > - The `unpack_trees()` function tries to reuse the `split_index`
> structure from the source index, if any, but does not propagate the
> `SPLIT_INDEX_ORDERED` flag.
> > While a fix for the second bug would let this test case pass, there are
> other conditions where the `SOMETHING_CHANGED` bit is set. Therefore,
> the bug that most crucially needs to be fixed is the first one.
> > Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Very well said. Thank you!!!
> ---
> t/t7527-builtin-fsmonitor.sh | 37 ++++++++++++++++++++++++++++++++++++
> 1 file changed, 37 insertions(+)
> > diff --git a/t/t7527-builtin-fsmonitor.sh b/t/t7527-builtin-fsmonitor.sh
> index d419085379c..cbafdd69602 100755
> --- a/t/t7527-builtin-fsmonitor.sh
> +++ b/t/t7527-builtin-fsmonitor.sh
> @@ -1003,4 +1003,41 @@ test_expect_success !UNICODE_COMPOSITION_SENSITIVE 'Unicode nfc/nfd' '
> egrep "^event: nfd/d_${utf8_nfc}/?$" ./unicode.trace
> '
> > +test_expect_failure 'split-index and FSMonitor work well together' '
> + git init split-index &&
> + test_when_finished "git -C \"$PWD/split-index\" \
> + fsmonitor--daemon stop" &&
> + (
> + cd split-index &&
> + git config core.splitIndex true &&
> + # force split-index in most cases
> + git config splitIndex.maxPercentChange 99 &&
> + git config core.fsmonitor true &&
> +
> + # Create the following commit topology:
> + #
> + # * merge three
> + # |\
> + # | * three
> + # * | merge two
> + # |\|
> + # | * two
> + # * | one
> + # |/
> + # * 5a5efd7 initial
> +
> + test_commit initial &&
> + test_commit two &&
> + test_commit three &&
> + git reset --hard initial &&
> + test_commit one &&
> + test_tick &&
> + git merge two &&
> + test_tick &&
> + git merge three &&
> +
> + git rebase --force-rebase -r one
> + )
> +'
> +
> test_done
On the Git mailing list, Jeff Hostetler wrote (reply to this), regarding 3963d3e (outdated): On 3/22/23 12:00 PM, Johannes Schindelin via GitGitGadget wrote:
> From: Johannes Schindelin <johannes.schindelin@gmx.de>
> > When copying the `split_index` structure from one index structure to
> another, we need to propagate the `SPLIT_INDEX_ORDERED` flag, too, if it
> is set, otherwise Git might forget to write the shared index when that
> is actually needed.
> > It just so _happens_ that in many instances when `unpack_trees()` is
> called, the result causes the shared index to be written anyway, but
> there are edge cases when that is not so.
> > Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
> ---
> unpack-trees.c | 2 ++
> 1 file changed, 2 insertions(+)
> > diff --git a/unpack-trees.c b/unpack-trees.c
> index 90b92114be8..ca5e47c77c0 100644
> --- a/unpack-trees.c
> +++ b/unpack-trees.c
> @@ -1916,6 +1916,8 @@ int unpack_trees(unsigned len, struct tree_desc *t, struct unpack_trees_options
> * create a new one.
> */
> o->result.split_index = o->src_index->split_index;
> + if (o->src_index->cache_changed & SPLIT_INDEX_ORDERED)
> + o->result.cache_changed |= SPLIT_INDEX_ORDERED;
Nice find!
> o->result.split_index->refcount++;
> } else {
> o->result.split_index = init_split_index(&o->result); |
On the Git mailing list, Jeff Hostetler wrote (reply to this), regarding f1897b8 (outdated): On 3/22/23 12:00 PM, Johannes Schindelin via GitGitGadget wrote:
> From: Johannes Schindelin <johannes.schindelin@gmx.de>
> > When a split-index is in effect, the `$GIT_DIR/index` file needs to
> contain a "link" extension that contains all the information about the
> split-index, including the information about the shared index.
> > However, in some cases Git needs to suppress writing that "link"
> extension (i.e. to fall back to writing a full index) even if the
> in-memory index structure _has_ a `split_index` configured. This is the
> case e.g. when "too many not shared" index entries exist.
> > In such instances, the current code sets the `base_oid` field of said
> `split_index` structure to all-zero to indicate that `do_write_index()`
> should skip writing the "link" extension.
> > This can lead to problems later on, when the in-memory index is still
> used to perform other operations and eventually wants to write a
> split-index, detects the presence of the `split_index` and reuses that,
> too (under the assumption that it has been initialized correctly and
> still has a non-null `base_oid`).
> > Let's stop zeroing out the `base_oid` to indicate that the "link"
> extension should not be written.
> > One might be tempted to simply call `discard_split_index()` instead,
> under the assumption that Git decided to write a non-split index and
> therefore the the `split_index` structure might no longer be wanted.
> However, that is not possible because that would release index entries
> in `split_index->base` that are likely to still be in use. Therefore we
> cannot do that.
> > The next best thing we _can_ do is to introduce a flag, specifically
> indicating when the "link" extension should be skipped. So that's what
> we do here.
> > Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
> ---
> read-cache.c | 37 ++++++++++++++++++++++--------------
> t/t7527-builtin-fsmonitor.sh | 2 +-
> 2 files changed, 24 insertions(+), 15 deletions(-)
> > diff --git a/read-cache.c b/read-cache.c
> index b09128b1884..8fcb2d54c05 100644
> --- a/read-cache.c
> +++ b/read-cache.c
> @@ -2868,6 +2868,12 @@ static int record_ieot(void)
> return !git_config_get_index_threads(&val) && val != 1;
> }
> > +enum strip_extensions {
> + WRITE_ALL_EXTENSIONS = 0,
> + STRIP_ALL_EXTENSIONS = 1,
> + STRIP_LINK_EXTENSION_ONLY = 2
> +};
Earlier (in a response to Junio's response on this commit) I said
that I didn't think we needed to make a bit set here, but I want
to re-think that or at least walk thru the change and talk out loud.
I'll explain in-line below.
> +
> /*
> * On success, `tempfile` is closed. If it is the temporary file
> * of a `struct lock_file`, we will therefore effectively perform
> @@ -2876,7 +2882,7 @@ static int record_ieot(void)
> * rely on it.
> */
> static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
> - int strip_extensions, unsigned flags)
> + enum strip_extensions strip_extensions, unsigned flags)
> {
> uint64_t start = getnanotime();
> struct hashfile *f;
> @@ -3045,7 +3051,7 @@ static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
> return -1;
> }
> > - if (!strip_extensions && istate->split_index &&
> + if (strip_extensions == WRITE_ALL_EXTENSIONS && istate->split_index &&
> !is_null_oid(&istate->split_index->base_oid)) {
(I hate all of this double negative logic...)
Here we only want the extension if we have WRITE_ALL, so that is
NOT STRIP_ALL and NOT STRIP_LINK_ONLY, so that is OK.
> struct strbuf sb = STRBUF_INIT;
> > @@ -3060,7 +3066,7 @@ static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
> if (err)
> return -1;
> }
> - if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
> + if (strip_extensions != STRIP_ALL_EXTENSIONS && !drop_cache_tree && istate->cache_tree) {
Here we only want the extension when NOT STRIP_ALL, so this is
either WRITE_ALL or STRIP_LINK_ONLY, so this is OK. The rest are
the same, so I'll omit them.
[...]
All of this looks correct, but I stumbled over things on my first
or second reading. I wonder if it would it simplify things to define
this as:
enum strip_extensions {
WRITE_ALL_EXTENSIONS = 0,
STRIP_LINK_EXTENSION = (1<0),
STRIP_OTHER_EXTENSIONS = (1<1),
STRIP_ALL_EXTENSIONS = (STRIP_LINK_EXTENSION
| STRIP_OTHER_EXTENSIONS),
};
Then the link test becomes:
if ( ! (strip_extensions & STRIP_LINK_EXTENSION) &&
istate->split_index &&
...) {
and the others become:
if ( ! (strip_extensions & STRIP_OTHER_EXTENSIONS) &&
...) {
If we need to add the ability later to strip an individual,
we can easily add a bit to the enum and update the _ALL_ mask
and the corresponding `if` test.
In a later commit (probably in another series), we can invert
these double negatives to improve readability.
> +/*
> + * Write the Git index into a `.lock` file
> + *
> + * If `strip_link_extension` is non-zero, avoid writing any "link" extension
> + * (used by the split-index feature).
> + */
> static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
> - unsigned flags)
> + unsigned flags, int strip_link_extension)
> {
> int ret;
> int was_full = istate->sparse_index == INDEX_EXPANDED;
> @@ -3185,7 +3197,7 @@ static int do_write_locked_index(struct index_state *istate, struct lock_file *l
> */
> trace2_region_enter_printf("index", "do_write_index", the_repository,
> "%s", get_lock_file_path(lock));
> - ret = do_write_index(istate, lock->tempfile, 0, flags);
> + ret = do_write_index(istate, lock->tempfile, strip_link_extension ? STRIP_LINK_EXTENSION_ONLY : 0, flags);
> trace2_region_leave_printf("index", "do_write_index", the_repository,
> "%s", get_lock_file_path(lock));
> > @@ -3214,7 +3226,7 @@ static int write_split_index(struct index_state *istate,
> {
> int ret;
> prepare_to_write_split_index(istate);
> - ret = do_write_locked_index(istate, lock, flags);
> + ret = do_write_locked_index(istate, lock, flags, 0);
could we use the enum values here instead of 0 ?
> finish_writing_split_index(istate);
> return ret;
> }
> @@ -3366,9 +3378,7 @@ int write_locked_index(struct index_state *istate, struct lock_file *lock,
> if ((!si && !test_split_index_env) ||
> alternate_index_output ||
> (istate->cache_changed & ~EXTMASK)) {
> - if (si)
> - oidclr(&si->base_oid);
> - ret = do_write_locked_index(istate, lock, flags);
> + ret = do_write_locked_index(istate, lock, flags, 1);
and here
> goto out;
> }
> > @@ -3394,8 +3404,7 @@ int write_locked_index(struct index_state *istate, struct lock_file *lock,
> /* Same initial permissions as the main .git/index file */
> temp = mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
> if (!temp) {
> - oidclr(&si->base_oid);
> - ret = do_write_locked_index(istate, lock, flags);
> + ret = do_write_locked_index(istate, lock, flags, 1);
and here
> goto out;
> }
> ret = write_shared_index(istate, &temp, flags);
> diff --git a/t/t7527-builtin-fsmonitor.sh b/t/t7527-builtin-fsmonitor.sh
> index cbafdd69602..9fab9a2ab38 100755
> --- a/t/t7527-builtin-fsmonitor.sh
> +++ b/t/t7527-builtin-fsmonitor.sh
> @@ -1003,7 +1003,7 @@ test_expect_success !UNICODE_COMPOSITION_SENSITIVE 'Unicode nfc/nfd' '
> egrep "^event: nfd/d_${utf8_nfc}/?$" ./unicode.trace
> '
> > -test_expect_failure 'split-index and FSMonitor work well together' '
> +test_expect_success 'split-index and FSMonitor work well together' '
> git init split-index &&
> test_when_finished "git -C \"$PWD/split-index\" \
> fsmonitor--daemon stop" &&
Thanks
Jeff |
This patch series was integrated into seen via git@c6cb969. |
This patch series was integrated into seen via git@b8c158f. |
When a split-index is in effect, the `$GIT_DIR/index` file needs to contain a "link" extension that contains all the information about the split-index, including the information about the shared index. However, in some cases Git needs to suppress writing that "link" extension (i.e. to fall back to writing a full index) even if the in-memory index structure _has_ a `split_index` configured. This is the case e.g. when "too many not shared" index entries exist. In such instances, the current code sets the `base_oid` field of said `split_index` structure to all-zero to indicate that `do_write_index()` should skip writing the "link" extension. This can lead to problems later on, when the in-memory index is still used to perform other operations and eventually wants to write a split-index, detects the presence of the `split_index` and reuses that, too (under the assumption that it has been initialized correctly and still has a non-null `base_oid`). Let's stop zeroing out the `base_oid` to indicate that the "link" extension should not be written. One might be tempted to simply call `discard_split_index()` instead, under the assumption that Git decided to write a non-split index and therefore the `split_index` structure might no longer be wanted. However, that is not possible because that would release index entries in `split_index->base` that are likely to still be in use. Therefore we cannot do that. The next best thing we _can_ do is to introduce a bit field to indicate specifically which index extensions (not) to write. So that's what we do here. Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
As of e636a7b (read-cache: be specific what part of the index has changed, 2014-06-13), the paradigm `cache_changed = 1` fell out of fashion and it became a bit field instead. This is important because some bits have specific meaning and should not be unset without care, e.g. `SPLIT_INDEX_ORDERED`. However, b5a8169 (mark_fsmonitor_valid(): mark the index as changed if needed, 2019-05-24) did use the `cache_changed` attribute as if it were a Boolean instead of a bit field. That not only would override the `SPLIT_INDEX_ORDERED` bit when marking index entries as valid via the FSMonitor, but worse: it would set the `SOMETHING_OTHER` bit (whose value is 1). This means that Git would unnecessarily force a full index to be written out when a split index was asked for. Let's instead use the bit that is specifically intended to indicate FSMonitor-triggered changes, allowing the split-index feature to work as designed. Noticed-by: Jeff Hostetler <jeffhost@microsoft.com> Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
When copying the `split_index` structure from one index structure to another, we need to propagate the `SPLIT_INDEX_ORDERED` flag, too, if it is set, otherwise Git might forget to write the shared index when that is actually needed. It just so _happens_ that in many instances when `unpack_trees()` is called, the result causes the shared index to be written anyway, but there are edge cases when that is not so. Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
/submit |
Submitted as pull.1497.v2.git.1679870743.gitgitgadget@gmail.com To fetch this version into
To fetch this version to local tag
|
On the Git mailing list, Jeff Hostetler wrote (reply to this): On 3/26/23 6:45 PM, Johannes Schindelin via GitGitGadget wrote:
> I received an internal bug report that after upgrading from v2.39.2 to
> v2.40.0, some users ran into the following error message:
> ...
Very nice. I like the new bit mask and getting rid of the double
negative strip/write logic.
Signed-off-by: Jeff Hostetler <jeffhostetler@github.com>
Thanks!
Jeff
|
On the Git mailing list, Junio C Hamano wrote (reply to this): Jeff Hostetler <git@jeffhostetler.com> writes:
> On 3/26/23 6:45 PM, Johannes Schindelin via GitGitGadget wrote:
>> I received an internal bug report that after upgrading from v2.39.2 to
>> v2.40.0, some users ran into the following error message:
>>
> ...
>
> Very nice. I like the new bit mask and getting rid of the double
> negative strip/write logic.
Yup, it indeed made the logic easier to see to split them into
individual bits per extensions even though an initial "gut" reaction
may have been "you ain't gonna need such a flexibility" ;-).
Thanks, both. Will replace. |
This patch series was integrated into seen via git@40b90e9. |
This patch series was integrated into seen via git@6abfe93. |
This patch series was integrated into next via git@3a7b7da. |
This patch series was integrated into seen via git@b124a96. |
This patch series was integrated into seen via git@e155395. |
There was a status update in the "Cooking" section about the branch The index files can become corrupt under certain conditions when the split-index feature is in use, especially together with fsmonitor, which have been corrected. Will merge to 'master'. source: <pull.1497.v2.git.1679870743.gitgitgadget@gmail.com> |
This patch series was integrated into seen via git@c809c5f. |
This patch series was integrated into seen via git@7646f49. |
This patch series was integrated into seen via git@baf81d5. |
This patch series was integrated into seen via git@224dc8b. |
This patch series was integrated into seen via git@5e15a7f. |
This patch series was integrated into seen via git@c05f43e. |
This patch series was integrated into seen via git@f315a8b. |
This patch series was integrated into master via git@f315a8b. |
This patch series was integrated into next via git@f315a8b. |
Closed via f315a8b. |
I received an internal bug report that after upgrading from v2.39.2 to v2.40.0, some users ran into the following error message:
It sounds very much like the report we received in https://lore.kernel.org/git/CAC7ZvybvykKQyMWcZoKXxFDu_amnkxZCDq2C6KHoyhmHN2tcKw@mail.gmail.com/, but sadly that thread petered out when the reporter stopped being able to reproduce the problem.
After a few days of investigating, I am convinced that this is due to some old bugs, and not actually a regression in v2.40.0 (although I can believe that some improvements in v2.40.0 might make it easier to run into these bugs).
This patch series addresses those bugs.
Note: While the Git maintainer has stated a strong preference to introduce regression tests in the same patch that fixes the corresponding regression, this patch series starts with a stand-alone patch that demonstrates a problematic scenario via a new
test_expect_failure
test case. The reason why I specifically split out the test into its own commit is that there is a lot of information to unpack in the commit message that is larger than any of the subsequent bug fixes. Besides, it motivates not only the second patch (which marks the test case astest_expect_success
) but paints the larger picture necessary to understand also the need for the remaining two patches.This patch series is based on
maint-2.37
, the oldest maintenance branch it applies without merge conflicts. When merging withnext
, there are only trivial conflicts inunpack-trees.c
due toen/dir-api-cleanup
whereo->result
is nowo->internal.result
.Changes since v1:
enum strip_extensions
by the bit fieldenum write_extensions
, inverting the meaning of the values to avoid double negativesenum
valuesCc: Jeff Hostetler git@jeffhostetler.com