From fd292a53be38e25b286585a3955dea2f060fe699 Mon Sep 17 00:00:00 2001
From: nick <nick@paco.to>
Date: Mon, 11 Mar 2024 22:11:11 -0400
Subject: [PATCH] added parallelism

---
 src/directory.rs | 57 ++++++++++++++++++++++++++++++++++++++++--------
 1 file changed, 48 insertions(+), 9 deletions(-)

diff --git a/src/directory.rs b/src/directory.rs
index 34707a3..6bfdfdc 100644
--- a/src/directory.rs
+++ b/src/directory.rs
@@ -5,6 +5,8 @@ use std::io::{Result, ErrorKind};
 
 use crate::unit::Unit;
 
+use rayon::prelude::*;
+
 #[derive(Debug, Clone)]
 pub struct Directory {
     name: PathBuf,
@@ -52,15 +54,51 @@ impl Directory {
             }
         };
 
-        let mut size = 0;
-        let mut children = Vec::new();
-        for entry in dir {
-            let child = Self::new(entry?.path())?;
-            size += child.size;
-            children.push(child);
-        }
+        // this is a compicated iterator pattern. I'll do my best to explain.
+        // 1. the end result is that we `reduce()` the iterator to a single
+        //    (u64, Vec<Directory>) tuple to return. this is done by...
+        let (size, children) =
+            // 2. taking the iterator over the directory and parallelising it...
+            dir.par_bridge()
+            // 3, this is the recursive step: try to create new Directory
+            //    objects from each item in the iterator
+            .map(|entry| Self::new(entry?.path()))
+            // 4. the fold (this is try_fold because we're iterating over Result.).
+            //    each fold adds a directory as a child and increases the total size
+            .try_fold(
+                || (0, Vec::new()),
+                |(mut size, mut children), dir| {
+                    let dir = dir?;
+                    size += dir.size;
+                    children.push(dir);
+                    // have to specify std::io::Result::Ok otherwise it complains
+                    // that it can't infer the E in Result<T, E>
+                    Result::Ok((size, children))
+                }
+            )
+            // 5. the final step is to reduce, which is as simple as concatenating
+            //    every vector and summing up their sizes.
+            .try_reduce(
+                || (0, Vec::new()),
+                |(asize, mut avec), (bsize, bvec)| {
+                    avec.extend(bvec);
+                    Ok((asize + bsize, avec))
+                }
+            )?;
 
-        Ok(Self{
+        // final notes:
+        // 1. I am unsure if it is better to do a bunch of partial sums
+        // during the fold() and reduce() steps, or if it is best to
+        // have them only do data collection and sum the lengths
+        // later. intuitively we would want to do everything in
+        // parallel but I have no data to support this
+        // 2. this is a super complicated iterator pattern, If anyone
+        // knows how to simplify it I'm all ears, but this being
+        // parallel is the main advantage it has over du so I don't
+        // want to abandon that, even though a serial for loop is
+        // *incredibly* clearer
+
+        Ok(Self {
             name,
             size,
             children,
@@ -74,7 +112,8 @@ impl Directory {
         let mut result = self.vectorise(unit).iter().map(|e| e.stringify_tabbed(tab_size) + "\n").collect::<String>();
 
         if ! result.is_empty() {
-            let final_newline_char_range = result.len()-2 .. result.len();
+            // unless there was no output, remove the final "\n"
+            let final_newline_char_range = result.len()-"\n".len() .. result.len();
             result.drain(final_newline_char_range);
         }