Sort commits topologically

.sqlx/query-3620648a1003211c184423a3c13b6eae3500b888490430fbaf20517830db508e.json

@@ -0,0 +1,26 @@
+{
+  "db_name": "SQLite",
+  "query": "SELECT parent, child FROM commit_links JOIN commits AS p ON p.hash = parent JOIN commits AS c ON c.hash = child ORDER BY unixepoch(p.committer_date) ASC, p.hash ASC, unixepoch(c.committer_date) ASC, c.hash ASC ",
+  "describe": {
+    "columns": [
+      {
+        "name": "parent",
+        "ordinal": 0,
+        "type_info": "Text"
+      },
+      {
+        "name": "child",
+        "ordinal": 1,
+        "type_info": "Text"
+      }
+    ],
+    "parameters": {
+      "Right": 0
+    },
+    "nullable": [
+      false,
+      false
+    ]
+  },
+  "hash": "3620648a1003211c184423a3c13b6eae3500b888490430fbaf20517830db508e"
+}

.sqlx/query-639b559608466d60bc9cbfb9f543db10b596edee0e9bcb590c6139a7e8927945.json

@@ -0,0 +1,20 @@
+{
+  "db_name": "SQLite",
+  "query": "SELECT committer_date AS \"time: OffsetDateTime\" FROM commits ORDER BY unixepoch(committer_date) ASC, hash ASC ",
+  "describe": {
+    "columns": [
+      {
+        "name": "time: OffsetDateTime",
+        "ordinal": 0,
+        "type_info": "Text"
+      }
+    ],
+    "parameters": {
+      "Right": 0
+    },
+    "nullable": [
+      false
+    ]
+  },
+  "hash": "639b559608466d60bc9cbfb9f543db10b596edee0e9bcb590c6139a7e8927945"
+}

.sqlx/query-a718c407e9620bf99ed97247b89dd0afb7ef657ee0a302433641f651e72b8cbe.json

@@ -0,0 +1,20 @@
+{
+  "db_name": "SQLite",
+  "query": "SELECT hash FROM commits ORDER BY unixepoch(committer_date) ASC, hash ASC ",
+  "describe": {
+    "columns": [
+      {
+        "name": "hash",
+        "ordinal": 0,
+        "type_info": "Text"
+      }
+    ],
+    "parameters": {
+      "Right": 0
+    },
+    "nullable": [
+      false
+    ]
+  },
+  "hash": "a718c407e9620bf99ed97247b89dd0afb7ef657ee0a302433641f651e72b8cbe"
+}

.sqlx/query-ac7a2e727b2ff96ae6d6be2c05e9991a7a27708a69fbf89b43231927fc85519f.json

@@ -1,26 +0,0 @@
-{
-  "db_name": "SQLite",
-  "query": "SELECT hash, committer_date AS \"committer_date: time::OffsetDateTime\" FROM commits ORDER BY unixepoch(committer_date) ASC, hash ASC ",
-  "describe": {
-    "columns": [
-      {
-        "name": "hash",
-        "ordinal": 0,
-        "type_info": "Text"
-      },
-      {
-        "name": "committer_date: time::OffsetDateTime",
-        "ordinal": 1,
-        "type_info": "Text"
-      }
-    ],
-    "parameters": {
-      "Right": 0
-    },
-    "nullable": [
-      false,
-      false
-    ]
-  },
-  "hash": "ac7a2e727b2ff96ae6d6be2c05e9991a7a27708a69fbf89b43231927fc85519f"
-}

scripts/graph.ts

@@ -3,7 +3,7 @@ import uPlot from "./uPlot.js";
 interface GraphData {
     hashes: string[];
     times: number[];
-    metrics: { [key: string]: (number | null)[]; };
+    measurements: { [key: string]: (number | null)[]; };
 }
 
 let opts = {
@@ -29,7 +29,7 @@ fetch("data?metric=wall-clock/build")
         console.log(data);
         plot.setData([
             data.times,
-            data.metrics["wall-clock/build"]!,
+            data.measurements["wall-clock/build"]!,
         ]);
     });
 

src/server/web/pages/graph.rs

@@ -1,10 +1,14 @@
+mod util;
+
 use std::collections::HashMap;
 
 use askama::Template;
 use axum::{extract::State, response::IntoResponse, Json};
 use axum_extra::extract::Query;
+use futures::{StreamExt, TryStreamExt};
 use serde::{Deserialize, Serialize};
 use sqlx::{Acquire, SqlitePool};
+use time::OffsetDateTime;
 
 use crate::{
     config::Config,
@@ -45,9 +49,11 @@ pub struct QueryGraphData {
 #[derive(Serialize)]
 struct GraphData {
     hashes: Vec<String>,
+    parents: HashMap<usize, Vec<usize>>,
     times: Vec<i64>,
+
     // TODO f32 for smaller transmission size?
-    metrics: HashMap<String, Vec<Option<f64>>>,
+    measurements: HashMap<String, Vec<Option<f64>>>,
 }
 
 pub async fn get_graph_data(
@@ -58,30 +64,76 @@ pub async fn get_graph_data(
     let mut tx = db.begin().await?;
     let conn = tx.acquire().await?;
 
-    let rows = sqlx::query!(
+    // The SQL queries that return one result per commit *must* return the same
+    // amount of rows in the same order!
+
+    let unsorted_hashes = sqlx::query_scalar!(
         "\
-        SELECT \
-            hash, \
-            committer_date AS \"committer_date: time::OffsetDateTime\" \
-        FROM commits \
+        SELECT hash FROM commits \
         ORDER BY unixepoch(committer_date) ASC, hash ASC \
         "
     )
     .fetch_all(&mut *conn)
     .await?;
 
-    let mut hashes = Vec::with_capacity(rows.len());
-    let mut times = Vec::with_capacity(rows.len());
-    for row in rows {
-        hashes.push(row.hash);
-        times.push(row.committer_date.unix_timestamp());
+    let parent_child_pairs = sqlx::query!(
+        "\
+        SELECT parent, child \
+        FROM commit_links \
+        JOIN commits AS p ON p.hash = parent \
+        JOIN commits AS c ON c.hash = child \
+        ORDER BY \
+            unixepoch(p.committer_date) ASC, p.hash ASC, \
+            unixepoch(c.committer_date) ASC, c.hash ASC \
+        "
+    )
+    .fetch(&mut *conn)
+    .map_ok(|r| (r.parent, r.child))
+    .try_collect::<Vec<_>>()
+    .await?;
+
+    let sorted_hashes = util::sort_topologically(&unsorted_hashes, &parent_child_pairs);
+
+    let sorted_hash_indices = sorted_hashes
+        .iter()
+        .cloned()
+        .enumerate()
+        .map(|(i, hash)| (hash, i))
+        .collect::<HashMap<_, _>>();
+
+    let mut parents = HashMap::<usize, Vec<usize>>::new();
+    for (parent, child) in &parent_child_pairs {
+        let parent_idx = sorted_hash_indices[parent];
+        let child_idx = sorted_hash_indices[child];
+        parents.entry(parent_idx).or_default().push(child_idx);
     }
 
-    // TODO Topological sort (s. velcom)
-    // TODO Redo indices once queries are finalized
-    let mut metrics = HashMap::new();
+    // permutation[unsorted_index] = sorted_index
+    let permutation = unsorted_hashes
+        .iter()
+        .map(|h| sorted_hash_indices[h])
+        .collect::<Vec<_>>();
+
+    // Collect and permutate commit times
+    let mut times = vec![0; sorted_hashes.len()];
+    let mut rows = sqlx::query_scalar!(
+        "\
+        SELECT committer_date AS \"time: OffsetDateTime\" FROM commits \
+        ORDER BY unixepoch(committer_date) ASC, hash ASC \
+        "
+    )
+    .fetch(&mut *conn)
+    .enumerate();
+    while let Some((i, time)) = rows.next().await {
+        times[permutation[i]] = time?.unix_timestamp();
+    }
+    drop(rows);
+
+    // Collect and permutate measurements
+    let mut measurements = HashMap::new();
     for metric in form.metric {
-        let values = sqlx::query_scalar!(
+        let mut values = vec![None; sorted_hashes.len()];
+        let mut rows = sqlx::query_scalar!(
             "\
             WITH \
             measurements AS ( \
@@ -99,15 +151,20 @@ pub async fn get_graph_data(
             ",
             metric,
         )
-        .fetch_all(&mut *conn)
-        .await?;
+        .fetch(&mut *conn)
+        .enumerate();
+        while let Some((i, value)) = rows.next().await {
+            values[permutation[i]] = value?;
+        }
+        drop(rows);
 
-        metrics.insert(metric, values);
+        measurements.insert(metric, values);
     }
 
     Ok(Json(GraphData {
-        hashes,
+        hashes: sorted_hashes,
+        parents,
         times,
-        metrics,
+        measurements,
     }))
 }

src/server/web/pages/graph/util.rs

@@ -0,0 +1,87 @@
+use std::collections::{HashMap, HashSet};
+
+/// Sort commits topologically such that parents come before their children.
+///
+/// Assumes that `parent_child_pairs` contains no duplicates and is in the
+/// desired order (see below for more info on the order).
+///
+/// The algorithm used is a version of [Kahn's algorithm][0] that starts at the
+/// nodes with no parents. It uses a stack for the set of parentless nodes,
+/// meaning the resulting commit order is depth-first-y, not breadth-first-y.
+/// For example, this commit graph (where children are ordered top to bottom)
+/// results in the order `A, B, C, D, E, F` and not an interleaved order like
+/// `A, B, D, C, E, F` (which a queue would produce):
+///
+/// ```text
+/// A - B - C
+///  \       \
+///   D - E - F
+/// ```
+///
+/// When a node is visited and added to the list of sorted nodes, it is removed
+/// as parent from all its children. Those who had no other parents are added to
+/// the stack in reverse order. In the final list, the children appear in the
+/// order they appeared in the parent child pairs, if possible. This means that
+/// the order of the commits and of the pairs matters and should probably be
+/// deterministic.
+///
+/// [0]: https://en.wikipedia.org/wiki/Topological_sorting#Kahn's_algorithm
+pub fn sort_topologically(
+    commits: &[String],
+    parent_child_pairs: &[(String, String)],
+) -> Vec<String> {
+    // These maps have entries for each commit hash we might want to inspect, so
+    // we know `.get()`, `.get_mut()` and `.remove()` must always succeed.
+    let mut parent_child_map = commits
+        .iter()
+        .map(|hash| (hash.clone(), Vec::<String>::new()))
+        .collect::<HashMap<_, _>>();
+    let mut child_parent_map = commits
+        .iter()
+        .map(|hash| (hash.clone(), HashSet::<String>::new()))
+        .collect::<HashMap<_, _>>();
+    for (parent, child) in parent_child_pairs {
+        parent_child_map
+            .get_mut(parent)
+            .unwrap()
+            .push(child.clone());
+        child_parent_map
+            .get_mut(child)
+            .unwrap()
+            .insert(parent.clone());
+    }
+
+    // Initialize parentless stack using commit list, in reverse order so that
+    // the order is right when popping.
+    let mut parentless = Vec::<String>::new();
+    for commit in commits.iter().rev() {
+        if child_parent_map[commit].is_empty() {
+            // A (quadratic-time) linear scan here is OK since the number of
+            // parentless commits is usually fairly small.
+            if !parentless.contains(commit) {
+                parentless.push(commit.clone());
+            }
+        }
+    }
+
+    let mut sorted = Vec::<String>::new();
+    while let Some(hash) = parentless.pop() {
+        // Inspect children in reverse order so that the order is right when
+        // popping off the parentless stack.
+        for child in parent_child_map.remove(&hash).unwrap().into_iter().rev() {
+            let child_parents = child_parent_map.get_mut(&child).unwrap();
+            child_parents.remove(&hash);
+            if child_parents.is_empty() {
+                parentless.push(child);
+            }
+        }
+
+        sorted.push(hash);
+    }
+
+    assert!(parent_child_map.is_empty());
+    assert!(child_parent_map.values().all(|v| v.is_empty()));
+    assert!(parentless.is_empty());
+    assert_eq!(commits.len(), sorted.len());
+    sorted
+}