tablejohn/src/recurring/repo.rs

203 lines
6.2 KiB
Rust

//! Add new commits to the database and update the tracked refs.
// TODO Think about whether ref hashes should be tracked in the db
use std::collections::HashSet;
use futures::TryStreamExt;
use gix::{objs::Kind, traverse::commit::Info, ObjectId, Repository};
use sqlx::{Acquire, SqliteConnection, SqlitePool};
use tracing::{debug, info};
use crate::somehow;
async fn get_all_commits_from_db(
conn: &mut SqliteConnection,
) -> somehow::Result<HashSet<ObjectId>> {
let hashes = sqlx::query!("SELECT hash FROM commits")
.fetch(conn)
.err_into::<somehow::Error>()
.and_then(|r| async move { r.hash.parse::<ObjectId>().map_err(|e| e.into()) })
.try_collect::<HashSet<_>>()
.await?;
Ok(hashes)
}
fn get_new_commits_from_repo(
repo: &Repository,
old: &HashSet<ObjectId>,
) -> somehow::Result<Vec<Info>> {
// Collect all references starting with "refs"
let mut all_references: Vec<ObjectId> = vec![];
for reference in repo.references()?.prefixed("refs")? {
let reference = reference.map_err(|e| somehow::Error(anyhow::anyhow!(e)))?;
let id = reference.into_fully_peeled_id()?;
// Some repos *cough*linuxkernel*cough* have refs that don't point to
// commits. This makes the rev walk choke and die. We don't want that.
if id.object()?.kind != Kind::Commit {
continue;
}
all_references.push(id.into());
}
// Walk from those until hitting old references
let new_commits = repo
.rev_walk(all_references)
.selected(|c| !old.contains(c))?
.map(|r| r.map(|i| i.detach()))
.collect::<Result<Vec<_>, _>>()?;
Ok(new_commits)
}
async fn insert_new_commits(conn: &mut SqliteConnection, new: &[Info]) -> somehow::Result<()> {
for commit in new {
let hash = commit.id.to_string();
sqlx::query!("INSERT OR IGNORE INTO commits (hash) VALUES (?)", hash)
.execute(&mut *conn)
.await?;
}
Ok(())
}
async fn insert_new_commit_links(conn: &mut SqliteConnection, new: &[Info]) -> somehow::Result<()> {
for commit in new {
let child = commit.id.to_string();
for parent in &commit.parent_ids {
let parent = parent.to_string();
// Commits *cough*linuxkernel*cough* may list the same parent
// multiple times, so we just ignore duplicates during insert.
sqlx::query!(
"INSERT OR IGNORE INTO commit_links (parent, child) VALUES (?, ?)",
parent,
child,
)
.execute(&mut *conn)
.await?;
}
}
Ok(())
}
async fn mark_all_commits_as_old(conn: &mut SqliteConnection) -> somehow::Result<()> {
sqlx::query!("UPDATE commits SET new = 0")
.execute(conn)
.await?;
Ok(())
}
async fn track_main_branch(conn: &mut SqliteConnection, repo: &Repository) -> somehow::Result<()> {
let Some(head) = repo.head_ref()? else { return Ok(()); };
let name = head.inner.name.to_string();
let hash = head.into_fully_peeled_id()?.to_string();
sqlx::query!(
"INSERT OR IGNORE INTO tracked_refs (name, hash) VALUES (?, ?)",
name,
hash,
)
.execute(conn)
.await?;
Ok(())
}
// TODO Write all refs to DB, not just tracked ones
async fn update_tracked_refs(
conn: &mut SqliteConnection,
repo: &Repository,
) -> somehow::Result<()> {
let tracked_refs = sqlx::query!("SELECT name, hash FROM tracked_refs")
.fetch_all(&mut *conn)
.await?;
for tracked_ref in tracked_refs {
if let Some(reference) = repo.try_find_reference(&tracked_ref.name)? {
let hash = reference.id().to_string();
if hash != tracked_ref.hash {
debug!("Updated tracked ref {}", tracked_ref.name);
sqlx::query!(
"UPDATE tracked_refs SET hash = ? WHERE name = ?",
hash,
tracked_ref.name
)
.execute(&mut *conn)
.await?;
}
} else {
debug!("Deleted tracked ref {}", tracked_ref.name);
sqlx::query!("DELETE FROM tracked_refs WHERE name = ?", tracked_ref.name)
.execute(&mut *conn)
.await?;
}
}
Ok(())
}
// TODO tracked -> reachable, 0 = unreachable, 1 = reachable, 2 = reachable from tracked ref
async fn update_commit_tracked_status(conn: &mut SqliteConnection) -> somehow::Result<()> {
sqlx::query!(
"
WITH RECURSIVE reachable(hash) AS (
SELECT hash FROM tracked_refs
UNION
SELECT parent FROM commit_links
JOIN reachable ON hash = child
)
UPDATE commits
SET tracked = (hash IN reachable)
"
)
.execute(conn)
.await?;
Ok(())
}
pub async fn update(db: &SqlitePool, repo: &Repository) -> somehow::Result<()> {
debug!("Updating repo");
let mut tx = db.begin().await?;
let conn = tx.acquire().await?;
let old = get_all_commits_from_db(&mut *conn).await?;
debug!("Loaded {} commits from the db", old.len());
let repo_is_new = old.is_empty();
if repo_is_new {
info!("Initializing new repo");
}
let new = get_new_commits_from_repo(repo, &old)?;
debug!("Found {} new commits in repo", new.len());
// Defer foreign key checks until the end of the transaction to improve
// insert performance.
sqlx::query!("PRAGMA defer_foreign_keys=1")
.execute(&mut *conn)
.await?;
// Inserts are grouped by table so sqlite can process them *a lot* faster
// than if they were grouped by commit (insert commit and parents, then next
// commit and so on).
insert_new_commits(conn, &new).await?;
insert_new_commit_links(conn, &new).await?;
debug!("Inserted {} new commits into db", new.len());
if repo_is_new {
mark_all_commits_as_old(conn).await?;
track_main_branch(conn, repo).await?;
debug!("Prepared new repo");
}
update_tracked_refs(conn, repo).await?;
update_commit_tracked_status(conn).await?;
debug!("Updated tracked refs");
tx.commit().await?;
if repo_is_new {
info!("Initialized new repo");
}
debug!("Updated repo");
Ok(())
}