nick
/
hb
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

fixed parallelism

main
nick 2024-03-13 17:40:46 -04:00
parent c170f4baa6
commit 158aaea2f3
1 changed files with 82 additions and 57 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

139
src/directory.rs
View File

@ -1,8 +1,10 @@
use std::ffi::OsString;
use std::fs::read_dir;
use std::hint::unreachable_unchecked;
use std::path::{Path, PathBuf};
use std::io::{Result, ErrorKind};
use std::io::ErrorKind;
use anyhow::{Context, Result};
use crate::args::Args;
use crate::unit::Unit;
use rayon::prelude::*;
@ -19,38 +21,55 @@ impl Directory {
self.size
}
#[inline]
pub fn scale(&self, unit: Unit) -> String {
unit.convert(self.size)
}
#[inline]
pub fn path(&self) -> &Path {
self.name.as_ref()
}
pub fn new<P>(path: P) -> Result<Self>
where
P: AsRef<Path>
{
pub fn new< P: AsRef<Path> >(path: P, args: &Args) -> Result<Option<Self>> {
let path = path.as_ref();
let name = path.canonicalize()?
.file_name()
// file_name() returns None if and only if the path ends in "..".
// due to the call to canonicalize(), this can never be the case,
// so this can be safely unwrapped
.unwrap_or_else(|| unsafe { unreachable_unchecked() })
let name = path.file_name()
.map_or_else(|| OsString::from("/"), ToOwned::to_owned)
.into();
// symlink_metadata() is the same as metadata() but it doesn't
// traverse symlinks, so that we can exclude them if necessary
let meta = match (path.symlink_metadata(), args.persistant()) {
(Ok(md), _) => md,
(Err(_), true) => return Ok(None),
(Err(e), false) => return Err(e.into()),
};
if args.should_exclude(path, &meta) {
// Ok(None) is only meant to arise from an error
// while persistant. When that happens, a no-op
// entry is substituted, which is precisely
// we want to happen when we hit an excluded file.
// NOTE: return Ok(None) is *not* equivalent, because
// that's only produced when an error occurs but
// the program is running in persistant mode. I used
// to return that, but that causes incredibly
// bizarre and wrong behaviour.
return Ok(Some( Self { name, size: 0, children: Vec::new() } ))
}
let dir = match read_dir(path) {
Ok(dir) => dir,
Err(io_error) => match io_error.kind() {
ErrorKind::NotADirectory => return Ok(Self {
name,
size: path.metadata()?.len(),
children: Vec::new()
}),
_ => return Err(io_error),
ErrorKind::NotADirectory => {
return Ok(Some(
Self {
name,
size: meta.len(),
children: Vec::new()
}
))
},
other => return Result::context(
Err(io_error),
format!("{}: {}", path.display(), other)
),
}
};
@ -62,16 +81,21 @@ impl Directory {
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()))
.map(|entry| Self::new(entry?.path(), args))
// 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?;
|(mut size, mut children), dir| -> Result<(u64, Vec<Self>)> {
let Some(dir) = Result::from(dir)?
else {
// some intermediate operation failed, but we
// are persistant, so just skip
return Result::Ok((0, Vec::new()))
};
size += dir.size;
children.push(dir);
// have to specify std::io::Result::Ok otherwise it complains
// have to specify anyhow::Result::Ok otherwise it complains
// that it can't infer the E in Result<T, E>
Result::Ok((size, children))
}
@ -82,56 +106,57 @@ impl Directory {
|| (0, Vec::new()),
|(asize, mut avec), (bsize, bvec)| {
avec.extend(bvec);
Ok((asize + bsize, avec))
Result::Ok((asize + bsize, avec))
}
)?;
// ^ note the Try, because of course any of these operations could
// fail
// 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
// 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
// knows how to simplify it I'm all ears, but 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,
})
Ok(Some(
Self {
name,
size,
children,
}
))
}
pub fn display(self, unit: Unit) -> String {
pub fn tree(self, args: &Args) -> String {
// since self.size is definitionally the greatest value, the tab length
// is just the length of self.len, plus two for a tab width
let tab_size = self.size.to_string().len() + 2;
let mut result = self.vectorise(unit).iter().map(|e| e.stringify_tabbed(tab_size) + "\n").collect::<String>();
if ! result.is_empty() {
// 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);
}
result
self.vectorise(args)
.iter()
.map(|e| e.stringify_tabbed(tab_size))
.reduce(|s1, s2| s1 + "\n" + &s2)
.unwrap_or_default()
}
/// TODO: make not recursive, take &self if possible,
/// and maybe write directly to stdout to not use so much mem
fn vectorise(self, unit: Unit) -> Vec<TreeEntry> {
fn vectorise(mut self, args: &Args) -> Vec<TreeEntry> {
let mut result = Vec::new();
result.push(TreeEntry::new(
Vec::new(), self.name.display().to_string(), self.size, unit
self.name.display().to_string(), self.size, args.unit()
));
let mut new_entry_part = TreePart::First;
let mut continue_part = TreePart::Wait;
self.children.retain(|dir| args.should_print(dir.path()));
let len = self.children.len();
for (idx, child) in self.children.into_iter().enumerate() {
@ -140,7 +165,7 @@ impl Directory {
continue_part = TreePart::Blank;
}
let subtree = child.vectorise(unit);
let subtree = child.vectorise(args);
for mut item in subtree {
if item.parts.is_empty() {
@ -164,9 +189,9 @@ struct TreeEntry {
unit: Unit
}
impl TreeEntry {
fn new(parts: Vec<TreePart>, path: String, size: u64, unit: Unit) -> Self {
fn new(path: String, size: u64, unit: Unit) -> Self {
Self {
parts, path, size, unit
parts: Vec::new(), path, size, unit
}
}