This avoids https://github.com/rust-lang/futures-rs/issues/2090. I
don't think we need to worry about reading legacy conflicts
asynchronously - async is really only useful for Google's backend
right now, and we don't use the legacy format at Google. In
particular, I don't want `MergedTree::value()` to have to be async.
It seems we'll end up using `block_on()` quite a bit, at least until
we're done transitioning to async, and the function name doesn't
conflict with anything else, so let's always import it when we need
it.
The commit backend at Google is cloud-based (and so are the other
backends); it reads and writes commits from/to a server, which stores
them in a database. That makes latency much higher than for disk-based
backends. To reduce the latency, we have a local daemon process that
caches and prefetches objects. There are still many cases where
latency is high, such as when diffing two uncached commits. We can
improve that by changing some of our (jj's) algorithms to read many
objects concurrently from the backend. In the case of tree-diffing, we
can fetch one level (depth) of the tree at a time. There are several
ways of doing that:
* Make the backend methods `async`
* Use many threads for reading from the backend
* Add backend methods for batch reading
I don't think we typically need CPU parallelism, so it's wasteful to
have hundreds of threads running in order to fetch hundreds of objects
in parallel (especially when using a synchronous backend like the Git
backend). Batching would work well for the tree-diffing case, but it's
not as composable as `async`. For example, if we wanted to fetch some
commits at the same time as we were doing a diff, it's hard to see how
to do that with batching. Using async seems like our best bet.
I didn't make the backend interface's write functions async because
writes are already async with the daemon we have at Google. That
daemon will hash the object and immediately return, and then send the
object to the server in the background. I think any cloud-based
solution will need a similar daemon process. However, we may need to
reconsider this if/when jj gets used on a server with a custom backend
that writes directly to a database (i.e. no async daemon in between).
I've tried to measure the performance impact. That's the largest
difference I've been able to measure was on `jj diff
--ignore-working-copy -s --from v5.0 --to v6.0` in the Linux repo,
which increases from 749 ms to 773 ms (3.3%). In most cases I've
tested, there's no measurable difference. I've tried diffing from the
root commit, as well as `jj --ignore-working-copy log --no-graph -r
'::v3.0 & author(torvalds)' -T 'commit_id ++ "\n"'` (to test a
commit-heavy load).
One problematic scenario is that we have commits imported by old jj, and all
of their descendant commits are created by jj. Therefore import_head_commits()
wouldn't reach the old ancestor commits.
This change might bury a real bug, but I don't have a better alternative. Maybe
we can remove this hack after a couple of jj releases, and add a debug command
that imports all reachable Git commits from all historical heads.
Closes#2343
While debugging git issues, I often ended up creating a deadlock by adding
debug prints. It's also not obvious that git::export_refs() works even if the
git_repo() has already been locked, whereas git::import_refs() wouldn't. Let's
consolidate lock handling to the backend implementation.
The main goal of this change is to enable tree-level conflict format, but it
also allows us to bulk-import commits on clone/init. I think a separate method
will help if we want to provide progress information, enable check for
.jjconflict entries under certain condition, etc.
Since git::import_refs() now depends on GitBackend type, it might be better to
remove git_repo from the function arguments.
We currently represent the root tree id in a commit by `Merge<TreeId>`
plus a boolean `uses_tree_conflict_format`. It's better to use an enum
for that. That makes it harder to forget to check which type of tree
it is, and it makes it impossible to store a legacy tree with multiple
ids (as we could with `uses_tree_conflict_format=false`,
`root_tree=Merge::new(...)`).
Maybe more importantly, we're also going to want to pass around this
information in most places where we currently pass a single `TreeId`,
and passing two separate values would be annoying.
One of the error types that I later created embedded `BackendError`, but `clippy` complained that the size of the type was too large. This helps address that.
Since `Conflict<T>` can also represent a non-conflict state (a single
term), `Merge<T>` seems like better name.
Thanks to @ilyagr for the suggestion in
https://github.com/martinvonz/jj/pull/1774#discussion_r1257547709
Sorry about the churn. It would have been better if I thought of this
name before I introduced `Conflict<T>`.
Tree-level conflicts (#1624) will be stored as multiple trees
associated with a single commit. This patch adds support for that in
`backend::Commit` and in the backends.
When the Git backend writes a tree conflict, it creates a special root
tree for the commit. That tree has only the individual trees from the
conflict as subtrees. That way we prevent the trees from getting
GC'd. We also write the tree ids to the extra metadata table
(i.e. outside of the Git repo) so we don't need to load the tree
object to determine if there are conflicts.
I also added new flag to `backend::Commit` indicating whether the
commit is a new-style commit (with support for tree-level
conflicts). That will help with the migration. We will remove it once
we no longer care about old repos. When the flag is set, we know that
a commit with a single tree cannot have conflicts. When the flag is
not set, it's an old-style commit where we have to walk the whole tree
to find conflicts.
Errors that may occur while loading backend would vary per backends, and
it's unlikely that these errors could be mapped to BackendError variants
other than BackendError::Other. So let's extract Other(_) of that kind as
a separate type to clarify there would be no other error variants.
Perhaps, Backend/Error will be renamed to CommitBackend/Error or
CommitStore/Error?, whereas I think BackendInit/LoadError can be shared
among store factories.
This helps to map initialization error to BackendError without too general
From impl. I don't think io::Error (or our PathError) should be automatically
translated to BackendError::Other because BackendError has more specific
variants depending on context. If the error is specific to initialization,
it makes sense to translate it to Other variant.
It was convenient that what the git backend stored in its "extras"
table is exactly a subset of the fields that local backend stores, but
it's bit ugly and limiting. For example, it makes it possible to
populate the `author` field in the git extras, but that would have no
effect. It's better that it's not possible to do that (we store the
author field in the git commit, of course).
What made me notice this now was that I'm working on tree-level
conflicts (#1624) and I'm thinking of adding a field to the git extras
saying "this commit has single tree, but it's still a new-style
commit", so we can know not to walking such trees to find path-level
conflicts. That's only needed for the git backend because we don't
care about compatibility for the local backend.
This commit fixes#1305
Before this commit, running `jj init --git-repo=./` in a folder that
does not have a .git would cause jj to panick and leave an unfinished corrupted jj repo.
This commit fixes that by changing the call chain to return an error
instead of calling .unwrap() and panicking. This commit also adds logic to delete the unfinished jj
repository when the git backend initialization failed.
Before this commit, running the above command would result in the following
```
Running `jj/target/debug/jj init --git-repo=./`
thread 'main' panicked at 'called `Result::unwrap()` on an `Err` value: Error { code: -3, klass: 2, message: "failed to resolve path '/Users/kevincliao/github/jj/test-repo/.jj/repo/store/../../../.git': No such file or directory" }', lib/src/git_backend.rs:83:75
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
```
After this commit, the result is the following and the jj repo is deleted:
```
Running `jj/target/debug/jj init --git-repo=./`
Error: Failed to access the repository: Error: Failed to open git repository: failed to resolve path '/Users/kevincliao/github/jj/test-repo/.jj/repo/store/../../../.git': No such file or directory; class=Os (2); code=NotFound (-3)
```
Otherwise, "jj init --git-repo ." would create extra table files per commit,
and merge them.
I considered adding an explicit GitBackend method to be called from
git::import_refs(), but the call order matters. The method should be invoked
before calling store.get_commit(..) or mut_repo.add_head(..). Since commits
are likely to be loaded from the head, we can instead make read_commit()
import ancestor metadata at all.
Alternatively, we could make a Git commit hidden until it's inserted into
the extra table. It's rather big change, and I wouldn't like to do that
without thinking more thoroughly.
I'm going to extract a helper function that converts git2::Commit to
backend::Commit struct, and the commit id can also be obtained from the
git2::Commit object.
My first attempt was to fix up corrupted index when merging, but it turned
out to be not easy because the self side may contain corrupted data. It's
also possible that two concurrent commit operations have exactly the same
view state (because change id isn't hashed into commit id), and only the
table heads diverge.
#924
