Skip to content

Commit

Permalink
cli: git sync - Proof of concept
Browse files Browse the repository at this point in the history
This is not meant to be merged in this form.
I am bashing everything on the head with a large bat, until it works.
Once everything works nicely in this draft, I will break it up into
smaller tested pieces that are intended to be merged.

* Grab current heads and build a set of [(Parent, Child)...]
* Start a transaction.
* Fetch in the transaction:
    It's extremely  important to fetch *all* branches from *all* remotes.
    This is because if we don't fetch all branches, when we start rebasing,
    a root branch may be merged upstream with contents of a branch we didn't
    fetch, so we end up with conflicts. It's safest to always fetch all branches
    and then only rebase what we need.
* Grab current heads on transaction:
  * Transaction MutableRepo should be updated after fetch.
  * Build a Map where old heads are the keys and the new heads are values.
    * If old == new, skip it.
    * This way if nothing changed, we end up with an empty map and avoid bugs
      downstream (index.is_ancestor returns true if old == new).
  * Relationship is figured out by index.is_ancestor(old_head_id, new_head_id) in a loop.
* Check if rebase is needed.
  * old_heads.set_diffence(new_heads) gives old heads that are no longer heads, which would
    mean, their descendants need to be rebased.
* Find children needing rebase:
  * Build list of rebase specs with: (commit, new_parent)
    * commit is the immdiate child of old_heads needing rebase.
* call repo_mut.transform_descendants with a rewritter closure.
  * simplify_merge
  * rewrite commits matching the rebase_specs
  * for all other commits, if any new_parents match old_heads in our old -> new mapping,
    update those to point at the updated heads.
  * rebase with emptying newly emptied commits.


Issue: #1039
  • Loading branch information
essiene committed Nov 12, 2024
1 parent 845892d commit 344f0ae
Show file tree
Hide file tree
Showing 4 changed files with 340 additions and 0 deletions.
5 changes: 5 additions & 0 deletions cli/src/commands/git/mod.rs
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,7 @@ pub mod init;
pub mod push;
pub mod remote;
pub mod submodule;
pub mod sync;

use clap::Subcommand;

Expand All @@ -39,6 +40,8 @@ use self::remote::cmd_git_remote;
use self::remote::RemoteCommand;
use self::submodule::cmd_git_submodule;
use self::submodule::GitSubmoduleCommand;
use self::sync::cmd_git_sync;
use self::sync::GitSyncArgs;
use crate::cli_util::CommandHelper;
use crate::cli_util::WorkspaceCommandHelper;
use crate::command_error::user_error_with_message;
Expand All @@ -61,6 +64,7 @@ pub enum GitCommand {
Remote(RemoteCommand),
#[command(subcommand, hide = true)]
Submodule(GitSubmoduleCommand),
Sync(GitSyncArgs),
}

pub fn cmd_git(
Expand All @@ -77,6 +81,7 @@ pub fn cmd_git(
GitCommand::Push(args) => cmd_git_push(ui, command, args),
GitCommand::Remote(args) => cmd_git_remote(ui, command, args),
GitCommand::Submodule(args) => cmd_git_submodule(ui, command, args),
GitCommand::Sync(args) => cmd_git_sync(ui, command, args),
}
}

Expand Down
306 changes: 306 additions & 0 deletions cli/src/commands/git/sync.rs
Original file line number Diff line number Diff line change
@@ -0,0 +1,306 @@
use std::collections::BTreeMap;
use std::collections::BTreeSet;
use std::fmt;

use itertools::Itertools;
use jj_lib::backend::CommitId;
use jj_lib::commit::Commit;
use jj_lib::repo::Repo;
use jj_lib::revset::FailingSymbolResolver;
use jj_lib::revset::RevsetExpression;
use jj_lib::revset::RevsetIteratorExt;
use jj_lib::rewrite::EmptyBehaviour;
use jj_lib::str_util::StringPattern;

use crate::cli_util::short_change_hash;
use crate::cli_util::short_commit_hash;
use crate::cli_util::CommandHelper;
use crate::cli_util::WorkspaceCommandTransaction;
use crate::commands::CommandError;
use crate::git_util::get_fetch_remotes;
use crate::git_util::get_git_repo;
use crate::git_util::git_fetch;
use crate::git_util::FetchArgs;
use crate::ui::Ui;

/// Sync the local JJ repo to specified Git remote branch(es).
///
/// The sync command will first fetch from the Git remote, then
/// rebase all local changes onto the appropriate updated
/// heads that were fetched.
///
/// Changes that are made empty by the rebase are dropped.
#[derive(clap::Args, Clone, Debug)]
pub struct GitSyncArgs {
#[command(flatten)]
fetch: FetchArgs,
}

pub fn cmd_git_sync(
ui: &mut Ui,
command: &CommandHelper,
args: &GitSyncArgs,
) -> Result<(), CommandError> {
let mut workspace_command = command.workspace_helper(ui)?;

let git_repo = get_git_repo(workspace_command.repo().store())?;
let remotes = get_fetch_remotes(
ui,
command.settings(),
&git_repo,
&FetchArgs {
branch: vec![StringPattern::everything()],
remotes: vec![],
all_remotes: true,
},
)?;

let remote_patterns = remotes
.iter()
.map(|p| StringPattern::Exact(p.to_string()))
.collect_vec();

let mut tx = workspace_command.start_transaction();

let old_heads = get_branch_heads(tx.repo(), &args.fetch.branch)?;
let parent_child_pairs = get_parent_child_commit_pairs(
tx.repo(),
&old_heads.iter().map(|c| c.clone()).collect_vec(),
&args.fetch.branch,
&remote_patterns,
)?;

// prep to git fetch
git_fetch(
ui,
&mut tx,
&git_repo,
&FetchArgs {
branch: vec![StringPattern::everything()],
remotes: remotes.clone(),
all_remotes: true,
},
)?;

// is rebase needed?
let updated_heads = get_branch_heads(tx.repo(), &args.fetch.branch)?;
let new_heads = updated_heads
.difference(&old_heads)
.map(|c| c.clone())
.collect_vec();
if new_heads.is_empty() {
// rebase not needed
tx.finish(ui, format!("sync completed; no rebase"))?;
return Ok(());
}

// find rebase targets
let needs_rebase = old_heads
.difference(&updated_heads)
.map(|c| c.clone())
.collect_vec();

let old_to_new = map_old_to_new_heads(&tx, &needs_rebase[..], &new_heads[..])?;

let rebase_specs = parent_child_pairs
.iter()
.filter_map(|pair| {
if needs_rebase.contains(&pair.parent) {
if let Some(new) = old_to_new.get(&pair.parent.id()) {
Some(RebaseSpec {
commit: pair.child.id().clone(),
new_parent: new.clone(),
})
} else {
None
}
} else {
None
}
})
.collect_vec();

let settings = tx.settings().clone();

tx.repo_mut().transform_descendants(
&settings,
rebase_specs
.iter()
.map(|spec| spec.commit.clone())
.collect::<Vec<_>>(),
|mut rewriter| {
rewriter.simplify_ancestor_merge();
for spec in &rebase_specs {
// move children of updated branches to the latest branch head
if rewriter.old_commit().id() == &spec.commit {
rewriter.set_new_parents(vec![spec.new_parent.clone()]);
} else {
// if new parents point at old heads, update them to point
// at the updated heads.
let mut updated_parents: Vec<CommitId> = vec![];

let old_parents = rewriter.new_parents().to_vec();

for parent in &old_parents {
if let Some(updated) = maybe_update_commit(
rewriter.repo(),
&parent,
&new_heads.iter().map(|c| c.id().clone()).collect_vec(),
) {
updated_parents.push(updated.clone());
} else {
updated_parents.push(parent.clone());
}
}

rewriter.set_new_parents(updated_parents);
}
}

if let Some(builder) =
rewriter.rebase_with_empty_behavior(&settings, EmptyBehaviour::AbandonNewlyEmpty)?
{
builder.write()?;
}

Ok(())
},
)?;

tx.finish(ui, format!("sync completed; commits rebased to new heads"))?;

Ok(())
}

fn get_branch_heads(
repo: &dyn Repo,
branches: &[StringPattern],
) -> Result<BTreeSet<Commit>, CommandError> {
let mut commits: BTreeSet<Commit> = BTreeSet::from([]);

for branch in branches {
let mut branch_commits: BTreeSet<Commit> = RevsetExpression::bookmarks(branch.clone())
.heads()
.resolve_user_expression(repo, &FailingSymbolResolver)?
.evaluate(repo)?
.iter()
.commits(repo.store())
.try_collect()?;

commits.append(&mut branch_commits);
}

Ok(commits)
}

fn maybe_update_commit(
repo: &dyn Repo,
commit: &CommitId,
new_heads: &[CommitId],
) -> Option<CommitId> {
new_heads
.iter()
.filter_map(|new| {
if new != commit && repo.index().is_ancestor(commit, new) {
Some(new.clone())
} else {
None
}
})
.next()
}

fn map_old_to_new_heads(
tx: &WorkspaceCommandTransaction,
old_heads: &[Commit],
new_heads: &[Commit],
) -> Result<BTreeMap<CommitId, CommitId>, CommandError> {
let mut out: BTreeMap<CommitId, CommitId> = BTreeMap::from([]);
let index = tx.repo().index();
new_heads.iter().for_each(|new| {
old_heads.iter().for_each(|old| {
if old != new && index.is_ancestor(old.id(), new.id()) {
out.insert(old.id().clone(), new.id().clone());
}
});
});

Ok(out)
}

fn format_commit(commit: &Commit) -> String {
let change_hash = short_change_hash(commit.change_id());
let commit_hash = short_commit_hash(commit.id());

format!("{commit_hash}:{change_hash}")
}

#[derive(Eq, Ord, PartialEq, PartialOrd)]
pub struct CommitPair {
parent: Commit,
child: Commit,
}

impl fmt::Display for CommitPair {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
let parent = format_commit(&self.parent);
let child = format_commit(&self.child);
write!(f, "=> {parent} --> {child}")
}
}

pub struct RebaseSpec {
commit: CommitId,
new_parent: CommitId,
}

fn get_parent_child_commit_pairs(
repo: &dyn Repo,
start: &[Commit],
branches: &[StringPattern],
remotes: &[StringPattern],
) -> Result<Vec<CommitPair>, CommandError> {
let store = repo.store();
let mut pairs: Vec<CommitPair> = vec![];

for commit in start {
for parent_id in commit.parent_ids() {
pairs.push(CommitPair {
parent: store.get_commit(parent_id)?,
child: commit.clone(),
});
}
}

let start = start.iter().map(|c| c.id().clone()).collect_vec();

for remote in remotes {
for branch in branches {
let commits: Vec<Commit> = RevsetExpression::commits(start.to_vec())
.descendants()
.minus(&RevsetExpression::commits(start.to_vec()))
.minus(&RevsetExpression::remote_bookmarks(
branch.clone(),
remote.clone(),
None,
))
.resolve_user_expression(repo, &FailingSymbolResolver)?
.evaluate(repo)?
.iter()
.commits(repo.store())
.try_collect()?;

for commit in commits {
for parent_id in commit.parent_ids() {
pairs.push(CommitPair {
parent: store.get_commit(parent_id)?,
child: commit.clone(),
});
}
}
}
}

pairs.sort();
Ok(pairs)
}
24 changes: 24 additions & 0 deletions cli/tests/[email protected]
Original file line number Diff line number Diff line change
Expand Up @@ -60,6 +60,7 @@ This document contains the help content for the `jj` command-line program.
* [`jj git remote remove`↴](#jj-git-remote-remove)
* [`jj git remote rename`↴](#jj-git-remote-rename)
* [`jj git remote set-url`↴](#jj-git-remote-set-url)
* [`jj git sync`↴](#jj-git-sync)
* [`jj help`↴](#jj-help)
* [`jj init`↴](#jj-init)
* [`jj interdiff`↴](#jj-interdiff)
Expand Down Expand Up @@ -1040,6 +1041,7 @@ For a comparison with Git, including a table of commands, see https://martinvonz
* `init` — Create a new Git backed repo
* `push` — Push to a Git remote
* `remote` — Manage Git remotes
* `sync` — Sync the local JJ repo to specified Git remote branch(es)
Expand Down Expand Up @@ -1247,6 +1249,28 @@ Set the URL of a Git remote
## `jj git sync`
Sync the local JJ repo to specified Git remote branch(es).
The sync command will first fetch from the Git remote, then rebase all local changes onto the appropriate updated heads that were fetched.
Changes that are made empty by the rebase are dropped.
**Usage:** `jj git sync [OPTIONS]`
###### **Options:**
* `-b`, `--branch <BRANCH>` — Fetch only some of the branches
By default, the specified name matches exactly. Use `glob:` prefix to expand `*` as a glob. The other wildcard characters aren't supported.
Default value: `glob:*`
* `--remote <REMOTE>` — The remote to fetch from (only named remotes are supported, can be repeated)
* `--all-remotes` — Fetch from all remotes
## `jj help`
Print this message or the help of the given subcommand(s)
Expand Down
5 changes: 5 additions & 0 deletions lib/src/rewrite.rs
Original file line number Diff line number Diff line change
Expand Up @@ -155,6 +155,11 @@ impl<'repo> CommitRewriter<'repo> {
self.mut_repo
}

/// Returns immutable reference to the `MutableRepo`.
pub fn repo(&mut self) -> &MutableRepo {
&self.mut_repo
}

/// The commit we're rewriting.
pub fn old_commit(&self) -> &Commit {
&self.old_commit
Expand Down

0 comments on commit 344f0ae

Please sign in to comment.