cli: Add an option to record consecutive snapshots as a single op

CHANGELOG.md

@@ -51,6 +51,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 * When creating a new workspace, the sparse patterns are now copied over from
   the current workspace.
 
+* Set config `snapshot.squash-consecutive-snapshots = true` to make consecutive
+  snapshots result in a single operation being recorded, effectively
+  automatically abandoning the intermediate snapshot operations, which helps
+  GC, especially if you run `jj log` in a loop for example.
+
 ### Fixed bugs
 
 * On Windows, symlinks in the repo are now materialized as regular files in the
@@ -70,7 +75,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   copy commit on top of a single specified revision, i.e. with one parent.
   `merge` creates a new working copy commit on top of *at least* two specified
   revisions, i.e. with two or more parents.
-  
+
   The only difference between these commands and `jj new`, which *also* creates
   a new working copy commit, is that `new` can create a working copy commit on
   top of any arbitrary number of revisions, so it can handle both the previous
@@ -222,7 +227,7 @@ Thanks to the people who made this release happen!
 
 * `jj branch set` no longer creates a new branch. Use `jj branch create`
   instead.
-  
+
 * `jj init --git` in an existing Git repository now errors and exits rather than
   creating a second Git store.
 
@@ -386,8 +391,8 @@ Thanks to the people who made this release happen!
 
 ### New features
 
-* The `ancestors()` revset function now takes an optional `depth` argument 
-  to limit the depth of the ancestor set. For example, use `jj log -r 
+* The `ancestors()` revset function now takes an optional `depth` argument
+  to limit the depth of the ancestor set. For example, use `jj log -r
   'ancestors(@, 5)` to view the last 5 commits.
 
 * Support for the Watchman filesystem monitor is now bundled by default. Set
@@ -552,13 +557,13 @@ Thanks to the people who made this release happen!
   respectively.
 
 * `jj log` timestamp format now accepts `.utc()` to convert a timestamp to UTC.
- 
+
 * templates now support additional string methods `.starts_with(x)`, `.ends_with(x)`
   `.remove_prefix(x)`, `.remove_suffix(x)`, and `.substr(start, end)`.
 
 * `jj next` and `jj prev` are added, these allow you to traverse the history
   in a linear style. For people coming from Sapling and `git-branchles`
-  see [#2126](https://github.com/martinvonz/jj/issues/2126) for 
+  see [#2126](https://github.com/martinvonz/jj/issues/2126) for
   further pending improvements.
 
 * `jj diff --stat` has been implemented. It shows a histogram of the changes,

cli/src/cli_util.rs

@@ -1433,7 +1433,21 @@ See https://github.com/martinvonz/jj/blob/main/docs/working-copy.md#stale-workin
                 print_failed_git_export(ui, &failed_branches)?;
             }
 
-            self.user_repo = ReadonlyUserRepo::new(tx.commit("snapshot working copy"));
+            const DESCRIPTION: &str = "snapshot working copy";
+            const TAG: &str = "snapshots";
+
+            if self.settings.squash_consecutive_snapshots() {
+                if let [parent_op] = tx.parent_ops() {
+                    let meta = &parent_op.store_operation().metadata;
+                    if meta.description == DESCRIPTION {
+                        let tag = meta.tags.get(TAG).and_then(|s| s.parse().ok()).unwrap_or(1);
+                        tx.replace_parent()?;
+                        tx.set_tag(TAG.into(), (tag + 1).to_string())
+                    }
+                }
+            }
+
+            self.user_repo = ReadonlyUserRepo::new(tx.commit(DESCRIPTION));
         }
         locked_ws.finish(self.user_repo.repo.op_id().clone())?;
         Ok(())

cli/src/config-schema.json

@@ -357,6 +357,11 @@
                     ],
                     "description": "New files with a size in bytes above this threshold are not snapshotted, unless the threshold is 0",
                     "default": "1MiB"
+                },
+                "squash-consecutive-snapshots": {
+                    "type": "boolean",
+                    "description": "Should snapshot operation happening after another snapshot operation replace it, effectively abandoning the first snapshot and releasing it's view for garbage collection",
+                    "default": false
                 }
             }
         },

cli/tests/test_working_copy.rs

@@ -30,3 +30,31 @@ fn test_snapshot_large_file() {
     want this file to be snapshotted. Otherwise add it to your `.gitignore` file.
     "###);
 }
+
+#[test]
+fn test_consecutive_snapshots() {
+    let test_env = TestEnvironment::default();
+    test_env.jj_cmd_ok(test_env.env_root(), &["init", "repo", "--git"]);
+    let repo_path = test_env.env_root().join("repo");
+
+    test_env.add_config(r#"snapshot.squash-consecutive-snapshots = true"#);
+
+    std::fs::write(repo_path.join("a"), "").unwrap();
+    test_env.jj_cmd_success(&repo_path, &["log"]);
+
+    std::fs::write(repo_path.join("b"), "").unwrap();
+    test_env.jj_cmd_success(&repo_path, &["log"]);
+
+    let stdout = test_env.jj_cmd_success(&repo_path, &["operation", "log"]);
+    insta::assert_snapshot!(stdout, @r###"
+    @  a0f9fa640725 [email protected] 2001-02-03 04:05:09.000 +07:00 - 2001-02-03 04:05:09.000 +07:00
+    │  snapshot working copy
+    │  args: jj log
+    │  snapshots: 2
+    ◉  6ac4339ad699 [email protected] 2001-02-03 04:05:07.000 +07:00 - 2001-02-03 04:05:07.000 +07:00
+    │  add workspace 'default'
+    ◉  1b0049c19762 [email protected] 2001-02-03 04:05:07.000 +07:00 - 2001-02-03 04:05:07.000 +07:00
+    │  initialize repo
+    ◉  000000000000 root()
+    "###);
+}

lib/src/op_store.rs

@@ -388,7 +388,7 @@ impl Operation {
             description: "".to_string(),
             hostname: "".to_string(),
             username: "".to_string(),
-            tags: HashMap::new(),
+            tags: BTreeMap::new(),
         };
         Operation {
             view_id: empty_view_id,
@@ -407,7 +407,7 @@ content_hash! {
         pub description: String,
         pub hostname: String,
         pub username: String,
-        pub tags: HashMap<String, String>,
+        pub tags: BTreeMap<String, String>,
     }
 }
 

lib/src/settings.rs

@@ -261,6 +261,12 @@ impl UserSettings {
         }
     }
 
+    pub fn squash_consecutive_snapshots(&self) -> bool {
+        self.config
+            .get_bool("snapshot.squash-consecutive-snapshots")
+            .unwrap_or(false)
+    }
+
     // separate from sign_settings as those two are needed in pretty different
     // places
     pub fn signing_backend(&self) -> Option<String> {

lib/src/simple_op_store.rs

@@ -356,7 +356,7 @@ fn operation_metadata_to_proto(
         description: metadata.description.clone(),
         hostname: metadata.hostname.clone(),
         username: metadata.username.clone(),
-        tags: metadata.tags.clone(),
+        tags: metadata.tags.clone().into_iter().collect(),
     }
 }
 
@@ -371,7 +371,7 @@ fn operation_metadata_from_proto(
         description: proto.description,
         hostname: proto.hostname,
         username: proto.username,
-        tags: proto.tags,
+        tags: proto.tags.into_iter().collect(),
     }
 }
 
@@ -757,7 +757,7 @@ mod tests {
                 description: "check out foo".to_string(),
                 hostname: "some.host.example.com".to_string(),
                 username: "someone".to_string(),
-                tags: hashmap! {
+                tags: btreemap! {
                     "key1".to_string() => "value1".to_string(),
                     "key2".to_string() => "value2".to_string(),
                 },

lib/src/transaction.rs

@@ -20,7 +20,7 @@ use itertools::Itertools as _;
 
 use crate::backend::Timestamp;
 use crate::index::ReadonlyIndex;
-use crate::op_store::OperationMetadata;
+use crate::op_store::{OperationId, OperationMetadata};
 use crate::operation::Operation;
 use crate::repo::{MutableRepo, ReadonlyRepo, Repo, RepoLoader, RepoLoaderError};
 use crate::settings::UserSettings;
@@ -32,25 +32,32 @@ pub struct Transaction {
     parent_ops: Vec<Operation>,
     op_metadata: OperationMetadata,
     end_time: Option<Timestamp>,
+    old_ops: Vec<OperationId>,
 }
 
 impl Transaction {
     pub fn new(mut_repo: MutableRepo, user_settings: &UserSettings) -> Transaction {
         let parent_ops = vec![mut_repo.base_repo().operation().clone()];
         let op_metadata = create_op_metadata(user_settings, "".to_string());
         let end_time = user_settings.operation_timestamp();
+        let old_ops = parent_ops.iter().map(|op| op.id().clone()).collect();
         Transaction {
             mut_repo,
             parent_ops,
             op_metadata,
             end_time,
+            old_ops,
         }
     }
 
     pub fn base_repo(&self) -> &Arc<ReadonlyRepo> {
         self.mut_repo.base_repo()
     }
 
+    pub fn parent_ops(&self) -> &[Operation] {
+        &self.parent_ops
+    }
+
     pub fn set_tag(&mut self, key: String, value: String) {
         self.op_metadata.tags.insert(key, value);
     }
@@ -74,12 +81,25 @@ impl Transaction {
         let repo_loader = self.base_repo().loader();
         let base_repo = repo_loader.load_at(&ancestor_op)?;
         let other_repo = repo_loader.load_at(&other_op)?;
+        self.old_ops.push(other_op.id().clone());
         self.parent_ops.push(other_op);
         let merged_repo = self.mut_repo();
         merged_repo.merge(&base_repo, &other_repo);
         Ok(())
     }
 
+    /// Replaces the parent ops with grandparent ops, effectively rewriting the
+    /// parent with self.
+    ///
+    /// Panics if self does not have exactly one parent.
+    pub fn replace_parent(&mut self) -> Result<(), RepoLoaderError> {
+        let [parent]: [Operation; 1] = std::mem::take(&mut self.parent_ops)
+            .try_into()
+            .expect("Merge or root operations cannot replace their parent");
+        self.parent_ops = parent.parents().collect::<Result<_, _>>()?;
+        Ok(())
+    }
+
     /// Writes the transaction to the operation store and publishes it.
     pub fn commit(self, description: impl Into<String>) -> Arc<ReadonlyRepo> {
         self.write(description).publish()
@@ -117,7 +137,7 @@ impl Transaction {
             .index_store()
             .write_index(mut_index, operation.id())
             .unwrap();
-        UnpublishedOperation::new(base_repo.loader(), operation, view, index)
+        UnpublishedOperation::new(base_repo.loader(), operation, view, index, self.old_ops)
     }
 }
 
@@ -148,6 +168,7 @@ pub struct UnpublishedOperation {
     repo_loader: RepoLoader,
     data: Option<NewRepoData>,
     closed: bool,
+    old_ops: Vec<OperationId>,
 }
 
 impl UnpublishedOperation {
@@ -156,6 +177,7 @@ impl UnpublishedOperation {
         operation: Operation,
         view: View,
         index: Box<dyn ReadonlyIndex>,
+        old_ops: Vec<OperationId>,
     ) -> Self {
         let data = Some(NewRepoData {
             operation,
@@ -166,6 +188,7 @@ impl UnpublishedOperation {
             repo_loader,
             data,
             closed: false,
+            old_ops,
         }
     }
 
@@ -179,7 +202,7 @@ impl UnpublishedOperation {
             let _lock = self.repo_loader.op_heads_store().lock();
             self.repo_loader
                 .op_heads_store()
-                .update_op_heads(data.operation.parent_ids(), data.operation.id());
+                .update_op_heads(&self.old_ops, data.operation.id());
         }
         let repo = self
             .repo_loader