803 lines
33 KiB
Rust
803 lines
33 KiB
Rust
// TODO when using GUI all or most println!() should be used as variables passed by argument
|
|
use humansize::{file_size_opts as options, FileSize};
|
|
use std::collections::{BTreeMap, HashMap};
|
|
use std::fs;
|
|
use std::fs::{File, Metadata};
|
|
use std::io::prelude::*;
|
|
use std::path::Path;
|
|
use std::time::{SystemTime, UNIX_EPOCH};
|
|
|
|
use crate::common::Common;
|
|
|
|
#[derive(PartialEq)]
|
|
pub enum CheckingMethod {
|
|
SIZE,
|
|
HASH,
|
|
}
|
|
|
|
#[derive(Eq, PartialEq)]
|
|
pub enum DeleteMethod {
|
|
None,
|
|
AllExceptNewest,
|
|
AllExceptOldest,
|
|
OneOldest,
|
|
OneNewest,
|
|
}
|
|
|
|
#[derive(Clone)]
|
|
struct FileEntry {
|
|
pub path: String,
|
|
pub size: u64,
|
|
pub created_date: SystemTime,
|
|
pub modified_date: SystemTime,
|
|
}
|
|
|
|
/// Struct with required information's to work
|
|
pub struct DuplicateFinder {
|
|
infos: Info,
|
|
files_with_identical_size: HashMap<u64, Vec<FileEntry>>,
|
|
files_with_identical_hashes: BTreeMap<u64, Vec<Vec<FileEntry>>>,
|
|
allowed_extensions: Vec<String>, // jpg, jpeg, mp4
|
|
// excluded_items: Vec<String>, // TODO, support for e.g. */.git/*
|
|
excluded_directories: Vec<String>,
|
|
included_directories: Vec<String>,
|
|
min_file_size: u64,
|
|
}
|
|
|
|
/// Info struck with helpful information's about results
|
|
pub struct Info {
|
|
pub errors: Vec<String>,
|
|
pub warnings: Vec<String>,
|
|
pub messages: Vec<String>,
|
|
pub number_of_checked_files: usize,
|
|
pub number_of_checked_folders: usize,
|
|
pub number_of_ignored_files: usize,
|
|
pub number_of_ignored_things: usize,
|
|
pub number_of_duplicated_files: usize,
|
|
pub lost_space: u64,
|
|
pub number_of_removed_files: usize,
|
|
pub number_of_failed_to_remove_files: usize,
|
|
pub gained_space: u64,
|
|
}
|
|
impl Info {
|
|
pub fn new() -> Info {
|
|
Info {
|
|
errors: vec![],
|
|
warnings: vec![],
|
|
messages: vec![],
|
|
number_of_checked_files: 0,
|
|
number_of_ignored_files: 0,
|
|
number_of_checked_folders: 0,
|
|
number_of_ignored_things: 0,
|
|
number_of_duplicated_files: 0,
|
|
lost_space: 0,
|
|
number_of_removed_files: 0,
|
|
number_of_failed_to_remove_files: 0,
|
|
gained_space: 0,
|
|
}
|
|
}
|
|
}
|
|
|
|
impl Default for Info {
|
|
fn default() -> Self {
|
|
Self::new()
|
|
}
|
|
}
|
|
|
|
impl DuplicateFinder {
|
|
pub fn new() -> DuplicateFinder {
|
|
DuplicateFinder {
|
|
infos: Info::new(),
|
|
files_with_identical_size: Default::default(),
|
|
files_with_identical_hashes: Default::default(),
|
|
// excluded_items: vec![],
|
|
excluded_directories: vec![],
|
|
included_directories: vec![],
|
|
min_file_size: 1024,
|
|
allowed_extensions: vec![],
|
|
}
|
|
}
|
|
pub fn get_infos(&self) -> &Info {
|
|
&self.infos
|
|
}
|
|
|
|
pub fn find_duplicates(&mut self, check_method: &CheckingMethod, delete_method: &DeleteMethod) {
|
|
self.optimize_directories();
|
|
self.check_files_size();
|
|
self.remove_files_with_unique_size();
|
|
if *check_method == CheckingMethod::HASH {
|
|
self.check_files_hash();
|
|
}
|
|
self.calculate_lost_space(check_method);
|
|
self.delete_files(check_method, delete_method);
|
|
self.debug_print();
|
|
}
|
|
|
|
pub fn set_min_file_size(&mut self, min_size: u64) {
|
|
self.min_file_size = min_size;
|
|
}
|
|
|
|
pub fn set_excluded_items(&mut self, _excluded_items: String) {
|
|
// TODO Still don't know how to exactly parse this
|
|
// Things like /.git/ should be by default hidden with help of this *.git*
|
|
}
|
|
pub fn set_allowed_extensions(&mut self, mut allowed_extensions: String) {
|
|
if allowed_extensions.is_empty() {
|
|
self.infos.messages.push("No allowed extension was provided, so all are allowed".to_string());
|
|
return;
|
|
}
|
|
allowed_extensions = allowed_extensions.replace("IMAGE", "jpg,kra,gif,png,bmp,tiff,webp,hdr,svg");
|
|
allowed_extensions = allowed_extensions.replace("VIDEO", "mp4,flv,mkv,webm,vob,ogv,gifv,avi,mov,wmv,mpg,m4v,m4p,mpeg,3gp");
|
|
allowed_extensions = allowed_extensions.replace("MUSIC", "mp3,flac,ogg,tta,wma,webm");
|
|
allowed_extensions = allowed_extensions.replace("TEXT", "txt,doc,docx,odt,rtf");
|
|
|
|
let extensions: Vec<String> = allowed_extensions.split(',').map(String::from).collect();
|
|
for mut extension in extensions {
|
|
if extension == "" {
|
|
continue;
|
|
}
|
|
|
|
if extension.contains('.') {
|
|
if !extension.starts_with('.') {
|
|
self.infos.warnings.push(extension + " is not valid extension(valid extension doesn't have dot inside)");
|
|
continue;
|
|
}
|
|
extension = extension.replace('.', "");
|
|
}
|
|
if !self.allowed_extensions.contains(&extension.trim().to_string()) {
|
|
self.allowed_extensions.push(extension.trim().to_string());
|
|
}
|
|
}
|
|
|
|
if self.allowed_extensions.is_empty() {
|
|
self.infos.messages.push("No valid extensions were provided, so allowing all extensions by default.".to_string());
|
|
}
|
|
}
|
|
pub fn set_include_directory(&mut self, mut include_directory: String) -> bool {
|
|
// let start_time: SystemTime = SystemTime::now();
|
|
|
|
if include_directory.is_empty() {
|
|
self.infos.errors.push("At least one directory must be provided".to_string());
|
|
return false;
|
|
}
|
|
|
|
include_directory = include_directory.replace("\"", "");
|
|
let directories: Vec<String> = include_directory.split(',').map(String::from).collect();
|
|
let mut checked_directories: Vec<String> = Vec::new();
|
|
|
|
for directory in directories {
|
|
let directory: String = directory.trim().to_string();
|
|
|
|
if directory == "" {
|
|
continue;
|
|
}
|
|
if directory.contains('*') {
|
|
self.infos.warnings.push("Include Directory Warning: Wildcards in path are not supported, ignoring ".to_string() + &*directory);
|
|
continue;
|
|
}
|
|
if !directory.starts_with('/') {
|
|
self.infos.warnings.push("Include Directory Warning: Relative path are not supported, ignoring ".to_string() + &*directory);
|
|
continue;
|
|
}
|
|
if !Path::new(&directory).exists() {
|
|
self.infos.warnings.push("Include Directory Warning: Provided folder path must exits, ignoring ".to_string() + &*directory);
|
|
continue;
|
|
}
|
|
if !Path::new(&directory).is_dir() {
|
|
self.infos.warnings.push("Include Directory Warning: Provided path must point at the directory, ignoring ".to_string() + &*directory);
|
|
continue;
|
|
}
|
|
|
|
// directory must end with /, due to possibility of incorrect assumption, that e.g. /home/rafal is top folder to /home/rafalinho
|
|
if !directory.ends_with('/') {
|
|
checked_directories.push(directory + "/");
|
|
} else {
|
|
checked_directories.push(directory);
|
|
}
|
|
}
|
|
|
|
if checked_directories.is_empty() {
|
|
self.infos.errors.push("Include Directory ERROR: Not found even one correct path to include which is required.".to_string());
|
|
return false;
|
|
}
|
|
|
|
self.included_directories = checked_directories;
|
|
|
|
//Common::print_time(start_time, SystemTime::now(), "set_include_directory".to_string());
|
|
true
|
|
}
|
|
|
|
pub fn set_exclude_directory(&mut self, mut exclude_directory: String) {
|
|
//let start_time: SystemTime = SystemTime::now();
|
|
if exclude_directory.is_empty() {
|
|
return;
|
|
}
|
|
|
|
exclude_directory = exclude_directory.replace("\"", "");
|
|
let directories: Vec<String> = exclude_directory.split(',').map(String::from).collect();
|
|
let mut checked_directories: Vec<String> = Vec::new();
|
|
|
|
for directory in directories {
|
|
let directory: String = directory.trim().to_string();
|
|
|
|
if directory == "" {
|
|
continue;
|
|
}
|
|
if directory == "/" {
|
|
self.infos.errors.push("Exclude Directory ERROR: Excluding / is pointless, because it means that no files will be scanned.".to_string());
|
|
break;
|
|
}
|
|
if directory.contains('*') {
|
|
self.infos.warnings.push("Exclude Directory Warning: Wildcards in path are not supported, ignoring ".to_string() + &*directory);
|
|
continue;
|
|
}
|
|
if !directory.starts_with('/') {
|
|
self.infos.warnings.push("Exclude Directory Warning: Relative path are not supported, ignoring ".to_string() + &*directory);
|
|
continue;
|
|
}
|
|
if !Path::new(&directory).exists() {
|
|
self.infos.warnings.push("Exclude Directory Warning: Provided folder path must exits, ignoring ".to_string() + &*directory);
|
|
continue;
|
|
}
|
|
if !Path::new(&directory).is_dir() {
|
|
self.infos.warnings.push("Exclude Directory Warning: Provided path must point at the directory, ignoring ".to_string() + &*directory);
|
|
continue;
|
|
}
|
|
|
|
// directory must end with /, due to possibility of incorrect assumption, that e.g. /home/rafal is top folder to /home/rafalinho
|
|
if !directory.ends_with('/') {
|
|
checked_directories.push(directory.trim().to_string() + "/");
|
|
} else {
|
|
checked_directories.push(directory.trim().to_string());
|
|
}
|
|
}
|
|
self.excluded_directories = checked_directories;
|
|
|
|
//Common::print_time(start_time, SystemTime::now(), "set_exclude_directory".to_string());
|
|
}
|
|
fn calculate_lost_space(&mut self, check_method: &CheckingMethod) {
|
|
let mut bytes: u64 = 0;
|
|
|
|
match check_method {
|
|
CheckingMethod::SIZE => {
|
|
for i in &self.files_with_identical_size {
|
|
bytes += i.0 * (i.1.len() as u64 - 1);
|
|
}
|
|
}
|
|
CheckingMethod::HASH => {
|
|
for i in &self.files_with_identical_hashes {
|
|
for j in i.1 {
|
|
bytes += i.0 * (j.len() as u64 - 1);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
self.infos.lost_space = bytes;
|
|
}
|
|
|
|
fn check_files_size(&mut self) {
|
|
// TODO maybe add multithreading checking for file hash
|
|
let start_time: SystemTime = SystemTime::now();
|
|
let mut folders_to_check: Vec<String> = Vec::with_capacity(1024 * 2); // This should be small enough too not see to big difference and big enough to store most of paths without needing to resize vector
|
|
|
|
// Add root folders for finding
|
|
for id in &self.included_directories {
|
|
folders_to_check.push(id.to_string());
|
|
}
|
|
|
|
let mut current_folder: String;
|
|
let mut next_folder: String;
|
|
while !folders_to_check.is_empty() {
|
|
current_folder = folders_to_check.pop().unwrap();
|
|
|
|
let read_dir = match fs::read_dir(¤t_folder) {
|
|
Ok(t) => t,
|
|
Err(_) => {
|
|
self.infos.warnings.push("Cannot open dir ".to_string() + &*current_folder);
|
|
continue;
|
|
} // Permissions denied
|
|
};
|
|
for entry in read_dir {
|
|
let entry_data = match entry {
|
|
Ok(t) => t,
|
|
Err(_) => {
|
|
self.infos.warnings.push("Cannot read entry in dir ".to_string() + &*current_folder);
|
|
continue;
|
|
} //Permissions denied
|
|
};
|
|
let metadata: Metadata = match entry_data.metadata() {
|
|
Ok(t) => t,
|
|
Err(_) => {
|
|
self.infos.warnings.push("Cannot read metadata in dir ".to_string() + &*current_folder);
|
|
continue;
|
|
} //Permissions denied
|
|
};
|
|
if metadata.is_dir() {
|
|
// if entry_data.file_name().into_string().is_err() { // Probably this can be removed, if crash still will be happens, then uncomment this line
|
|
// self.infos.warnings.push("Cannot read folder name in dir ".to_string() + &*current_folder);
|
|
// continue; // Permissions denied
|
|
// }
|
|
|
|
let mut is_excluded_dir = false;
|
|
next_folder = "".to_owned() + ¤t_folder + &entry_data.file_name().into_string().unwrap() + "/";
|
|
for ed in &self.excluded_directories {
|
|
if next_folder == *ed {
|
|
is_excluded_dir = true;
|
|
break;
|
|
}
|
|
}
|
|
if !is_excluded_dir {
|
|
folders_to_check.push(next_folder);
|
|
}
|
|
self.infos.number_of_checked_folders += 1;
|
|
} else if metadata.is_file() {
|
|
let mut have_valid_extension: bool;
|
|
let file_name_lowercase: String = entry_data.file_name().into_string().unwrap().to_lowercase();
|
|
|
|
if !self.allowed_extensions.is_empty() {
|
|
have_valid_extension = false;
|
|
for i in &self.allowed_extensions {
|
|
if file_name_lowercase.ends_with(&(".".to_string() + &*i.to_lowercase().to_string())) {
|
|
have_valid_extension = true;
|
|
break;
|
|
}
|
|
}
|
|
} else {
|
|
have_valid_extension = true;
|
|
}
|
|
|
|
if metadata.len() >= self.min_file_size && have_valid_extension {
|
|
let current_file_name = "".to_owned() + ¤t_folder + &entry_data.file_name().into_string().unwrap();
|
|
|
|
let fe: FileEntry = FileEntry {
|
|
path: current_file_name.clone(),
|
|
size: metadata.len(),
|
|
created_date: match metadata.created() {
|
|
Ok(t) => t,
|
|
Err(_) => {
|
|
self.infos.warnings.push("Unable to get creation date from file ".to_string() + &*current_file_name);
|
|
SystemTime::now()
|
|
} // Permissions Denied
|
|
},
|
|
modified_date: match metadata.modified() {
|
|
Ok(t) => t,
|
|
Err(_) => {
|
|
self.infos.warnings.push("Unable to get modification date from file ".to_string() + &*current_file_name);
|
|
SystemTime::now()
|
|
} // Permissions Denied
|
|
},
|
|
};
|
|
|
|
self.files_with_identical_size.entry(metadata.len()).or_insert_with(Vec::new);
|
|
self.files_with_identical_size.get_mut(&metadata.len()).unwrap().push(fe);
|
|
|
|
self.infos.number_of_checked_files += 1;
|
|
} else {
|
|
self.infos.number_of_ignored_files += 1;
|
|
}
|
|
} else {
|
|
// Probably this is symbolic links so we are free to ignore this
|
|
self.infos.number_of_ignored_things += 1;
|
|
}
|
|
}
|
|
}
|
|
Common::print_time(start_time, SystemTime::now(), "check_files_size".to_string());
|
|
//println!("Duration of finding duplicates {:?}", end_time.duration_since(start_time).expect("a"));
|
|
}
|
|
// pub fn save_results_to_file(&self) {}
|
|
|
|
/// Remove files which have unique size
|
|
fn remove_files_with_unique_size(&mut self) {
|
|
let start_time: SystemTime = SystemTime::now();
|
|
let mut new_hashmap: HashMap<u64, Vec<FileEntry>> = Default::default();
|
|
|
|
self.infos.number_of_duplicated_files = 0;
|
|
|
|
for entry in &self.files_with_identical_size {
|
|
if entry.1.len() > 1 {
|
|
self.infos.number_of_duplicated_files += entry.1.len() - 1;
|
|
new_hashmap.insert(*entry.0, entry.1.clone());
|
|
}
|
|
}
|
|
|
|
self.files_with_identical_size = new_hashmap;
|
|
|
|
Common::print_time(start_time, SystemTime::now(), "remove_files_with_unique_size".to_string());
|
|
}
|
|
|
|
/// Should be slower than checking in different ways, but still needs to be checked
|
|
fn check_files_hash(&mut self) {
|
|
let start_time: SystemTime = SystemTime::now();
|
|
let mut file_handler: File;
|
|
let mut hashmap_with_hash: HashMap<String, Vec<FileEntry>>;
|
|
|
|
for entry in &self.files_with_identical_size {
|
|
hashmap_with_hash = Default::default();
|
|
|
|
for file_entry in entry.1.iter().enumerate() {
|
|
file_handler = match File::open(&file_entry.1.path) {
|
|
Ok(t) => t,
|
|
Err(_) => {
|
|
self.infos.warnings.push("Unable to check hash of file ".to_string() + &*file_entry.1.path);
|
|
continue;
|
|
}
|
|
};
|
|
|
|
let mut error_reading_file: bool = false;
|
|
|
|
let mut hasher: blake3::Hasher = blake3::Hasher::new();
|
|
let mut buffer = [0u8; 16384];
|
|
loop {
|
|
let n = match file_handler.read(&mut buffer) {
|
|
Ok(t) => t,
|
|
Err(_) => {
|
|
self.infos.warnings.push("Error happened when checking hash of file ".to_string() + &*file_entry.1.path);
|
|
error_reading_file = true;
|
|
break;
|
|
}
|
|
}; //.unwrap();
|
|
if n == 0 {
|
|
break;
|
|
}
|
|
hasher.update(&buffer[..n]);
|
|
}
|
|
if !error_reading_file {
|
|
let hash_string: String = hasher.finalize().to_hex().to_string();
|
|
hashmap_with_hash.entry(hash_string.to_string()).or_insert_with(Vec::new);
|
|
hashmap_with_hash.get_mut(&*hash_string).unwrap().push(file_entry.1.to_owned());
|
|
}
|
|
}
|
|
for hash_entry in hashmap_with_hash {
|
|
if hash_entry.1.len() > 1 {
|
|
self.files_with_identical_hashes.entry(*entry.0).or_insert_with(Vec::new);
|
|
self.files_with_identical_hashes.get_mut(entry.0).unwrap().push(hash_entry.1);
|
|
}
|
|
}
|
|
}
|
|
Common::print_time(start_time, SystemTime::now(), "check_files_hash".to_string());
|
|
}
|
|
|
|
#[allow(dead_code)]
|
|
/// Setting include directories, panics when there is not directories available
|
|
fn debug_print(&self) {
|
|
println!("---------------DEBUG PRINT---------------");
|
|
println!("Number of checked files - {}", self.infos.number_of_checked_files);
|
|
println!("Number of checked folders - {}", self.infos.number_of_checked_folders);
|
|
println!("Number of ignored files - {}", self.infos.number_of_ignored_files);
|
|
println!("Number of ignored things(like symbolic links) - {}", self.infos.number_of_ignored_things);
|
|
println!("Number of duplicated files - {}", self.infos.number_of_duplicated_files);
|
|
let mut file_size: u64 = 0;
|
|
for i in &self.files_with_identical_size {
|
|
file_size += i.1.len() as u64;
|
|
}
|
|
println!("Files list size - {} ({})", self.files_with_identical_size.len(), file_size);
|
|
let mut hashed_file_size: u64 = 0;
|
|
for i in &self.files_with_identical_hashes {
|
|
for j in i.1 {
|
|
hashed_file_size += j.len() as u64;
|
|
}
|
|
}
|
|
println!("Hashed Files list size - {} ({})", self.files_with_identical_hashes.len(), hashed_file_size);
|
|
println!("Number of removed files - {}", self.infos.number_of_removed_files);
|
|
println!("Number of failed to remove files - {}", self.infos.number_of_failed_to_remove_files);
|
|
println!("Lost space - {} ({} bytes)", self.infos.lost_space.file_size(options::BINARY).unwrap(), self.infos.lost_space);
|
|
println!("Gained space by removing duplicated entries - {} ({} bytes)", self.infos.gained_space.file_size(options::BINARY).unwrap(), self.infos.gained_space);
|
|
println!("Excluded directories - {:?}", self.excluded_directories);
|
|
println!("Included directories - {:?}", self.included_directories);
|
|
println!("-----------------------------------------");
|
|
}
|
|
|
|
/// Print information about duplicated entries
|
|
pub fn print_duplicated_entries(&self, check_method: &CheckingMethod) {
|
|
let start_time: SystemTime = SystemTime::now();
|
|
let mut number_of_files: u64 = 0;
|
|
let mut number_of_groups: u64 = 0;
|
|
|
|
match check_method {
|
|
CheckingMethod::HASH => {
|
|
for (_size, vector) in self.files_with_identical_hashes.iter() {
|
|
for j in vector {
|
|
number_of_files += j.len() as u64;
|
|
number_of_groups += 1;
|
|
}
|
|
}
|
|
println!(
|
|
"Found {} duplicated files in {} groups with same content which took {}:",
|
|
number_of_files,
|
|
number_of_groups,
|
|
self.infos.lost_space.file_size(options::BINARY).unwrap()
|
|
);
|
|
for (key, vector) in self.files_with_identical_hashes.iter().rev() {
|
|
println!("Size - {}", key.file_size(options::BINARY).unwrap());
|
|
for j in vector {
|
|
for k in j {
|
|
println!("{}", k.path);
|
|
}
|
|
println!("----");
|
|
}
|
|
println!();
|
|
}
|
|
}
|
|
CheckingMethod::SIZE => {
|
|
for i in &self.files_with_identical_size {
|
|
number_of_files += i.1.len() as u64;
|
|
number_of_groups += 1;
|
|
}
|
|
println!(
|
|
"Found {} files in {} groups with same size(may have different content) which took {}:",
|
|
number_of_files,
|
|
number_of_groups,
|
|
self.infos.lost_space.file_size(options::BINARY).unwrap()
|
|
);
|
|
for i in &self.files_with_identical_size {
|
|
println!("Size - {}", i.0);
|
|
for j in i.1 {
|
|
println!("{}", j.path);
|
|
}
|
|
println!();
|
|
}
|
|
}
|
|
}
|
|
Common::print_time(start_time, SystemTime::now(), "print_duplicated_entries".to_string());
|
|
}
|
|
/// Remove unused entries when included or excluded overlaps with each other or are duplicated
|
|
/// ```
|
|
// let df : DuplicateFinder = saf
|
|
/// ```
|
|
fn optimize_directories(&mut self) -> bool {
|
|
let start_time: SystemTime = SystemTime::now();
|
|
|
|
let mut optimized_included: Vec<String> = Vec::<String>::new();
|
|
let mut optimized_excluded: Vec<String> = Vec::<String>::new();
|
|
// Remove duplicated entries like: "/", "/"
|
|
|
|
self.excluded_directories.sort();
|
|
self.included_directories.sort();
|
|
|
|
self.excluded_directories.dedup();
|
|
self.included_directories.dedup();
|
|
|
|
// Optimize for duplicated included directories - "/", "/home". "/home/Pulpit" to "/"
|
|
let mut is_inside: bool;
|
|
for ed_checked in &self.excluded_directories {
|
|
is_inside = false;
|
|
for ed_help in &self.excluded_directories {
|
|
if ed_checked == ed_help {
|
|
// We checking same element
|
|
continue;
|
|
}
|
|
if ed_checked.starts_with(ed_help) {
|
|
is_inside = true;
|
|
break;
|
|
}
|
|
}
|
|
if !is_inside {
|
|
optimized_excluded.push(ed_checked.to_string());
|
|
}
|
|
}
|
|
|
|
for id_checked in &self.included_directories {
|
|
is_inside = false;
|
|
for id_help in &self.included_directories {
|
|
if id_checked == id_help {
|
|
// We checking same element
|
|
continue;
|
|
}
|
|
if id_checked.starts_with(id_help) {
|
|
is_inside = true;
|
|
break;
|
|
}
|
|
}
|
|
if !is_inside {
|
|
optimized_included.push(id_checked.to_string());
|
|
}
|
|
}
|
|
|
|
self.included_directories = optimized_included;
|
|
optimized_included = Vec::<String>::new();
|
|
self.excluded_directories = optimized_excluded;
|
|
optimized_excluded = Vec::<String>::new();
|
|
|
|
// Remove include directories which are inside any exclude directory
|
|
for id in &self.included_directories {
|
|
let mut is_inside: bool = false;
|
|
for ed in &self.excluded_directories {
|
|
if id.starts_with(ed) {
|
|
is_inside = true;
|
|
break;
|
|
}
|
|
}
|
|
if !is_inside {
|
|
optimized_included.push(id.to_string());
|
|
}
|
|
}
|
|
self.included_directories = optimized_included;
|
|
optimized_included = Vec::<String>::new();
|
|
|
|
// Remove non existed directories
|
|
for id in &self.included_directories {
|
|
let path = Path::new(id);
|
|
if path.exists() {
|
|
optimized_included.push(id.to_string());
|
|
}
|
|
}
|
|
|
|
for ed in &self.excluded_directories {
|
|
let path = Path::new(ed);
|
|
if path.exists() {
|
|
optimized_excluded.push(ed.to_string());
|
|
}
|
|
}
|
|
|
|
self.included_directories = optimized_included;
|
|
// optimized_included = Vec::<String>::new();
|
|
self.excluded_directories = optimized_excluded;
|
|
optimized_excluded = Vec::<String>::new();
|
|
|
|
// Excluded paths must are inside include path, because
|
|
for ed in &self.excluded_directories {
|
|
let mut is_inside: bool = false;
|
|
for id in &self.included_directories {
|
|
if ed.starts_with(id) {
|
|
is_inside = true;
|
|
break;
|
|
}
|
|
}
|
|
if is_inside {
|
|
optimized_excluded.push(ed.to_string());
|
|
}
|
|
}
|
|
|
|
self.excluded_directories = optimized_excluded;
|
|
// optimized_excluded = Vec::<String>::new();
|
|
|
|
if self.included_directories.is_empty() {
|
|
self.infos.errors.push("Optimize Directories ERROR: Excluded directories overlaps all included directories.".to_string());
|
|
return false;
|
|
}
|
|
|
|
// Not needed, but better is to have sorted everything
|
|
self.excluded_directories.sort();
|
|
self.included_directories.sort();
|
|
Common::print_time(start_time, SystemTime::now(), "optimize_directories".to_string());
|
|
true
|
|
}
|
|
|
|
fn delete_files(&mut self, check_method: &CheckingMethod, delete_method: &DeleteMethod) {
|
|
if *delete_method == DeleteMethod::None {
|
|
return;
|
|
}
|
|
let start_time: SystemTime = SystemTime::now();
|
|
|
|
match check_method {
|
|
CheckingMethod::HASH => {
|
|
for entry in &self.files_with_identical_hashes {
|
|
for vector in entry.1 {
|
|
let tuple: (u64, usize, usize) = delete_files(&vector, &delete_method, &mut self.infos.warnings);
|
|
self.infos.gained_space += tuple.0;
|
|
self.infos.number_of_removed_files += tuple.1;
|
|
self.infos.number_of_failed_to_remove_files += tuple.2;
|
|
}
|
|
}
|
|
}
|
|
CheckingMethod::SIZE => {
|
|
for entry in &self.files_with_identical_size {
|
|
let tuple: (u64, usize, usize) = delete_files(&entry.1, &delete_method, &mut self.infos.warnings);
|
|
self.infos.gained_space += tuple.0;
|
|
self.infos.number_of_removed_files += tuple.1;
|
|
self.infos.number_of_failed_to_remove_files += tuple.2;
|
|
}
|
|
}
|
|
}
|
|
|
|
Common::print_time(start_time, SystemTime::now(), "delete_files".to_string());
|
|
}
|
|
}
|
|
impl Default for DuplicateFinder {
|
|
fn default() -> Self {
|
|
Self::new()
|
|
}
|
|
}
|
|
|
|
fn delete_files(vector: &[FileEntry], delete_method: &DeleteMethod, warnings: &mut Vec<String>) -> (u64, usize, usize) {
|
|
assert!(vector.len() > 1, "Vector length must be bigger than 1(This should be done in previous steps).");
|
|
let mut q_index: usize = 0;
|
|
let mut q_time: u64 = 0;
|
|
|
|
let mut gained_space: u64 = 0;
|
|
let mut removed_files: usize = 0;
|
|
let mut failed_to_remove_files: usize = 0;
|
|
|
|
match delete_method {
|
|
DeleteMethod::OneOldest => {
|
|
for files in vector.iter().enumerate() {
|
|
let time_since_epoch = files.1.created_date.duration_since(UNIX_EPOCH).expect("Invalid file date").as_secs();
|
|
if q_time == 0 || q_time > time_since_epoch {
|
|
q_time = time_since_epoch;
|
|
q_index = files.0;
|
|
}
|
|
}
|
|
match fs::remove_file(vector[q_index].path.clone()) {
|
|
Ok(_) => {
|
|
removed_files += 1;
|
|
gained_space += vector[q_index].size;
|
|
}
|
|
Err(_) => {
|
|
failed_to_remove_files += 1;
|
|
warnings.push("Failed to delete".to_string() + &*vector[q_index].path);
|
|
}
|
|
};
|
|
}
|
|
DeleteMethod::OneNewest => {
|
|
for files in vector.iter().enumerate() {
|
|
let time_since_epoch = files.1.created_date.duration_since(UNIX_EPOCH).expect("Invalid file date").as_secs();
|
|
if q_time == 0 || q_time < time_since_epoch {
|
|
q_time = time_since_epoch;
|
|
q_index = files.0;
|
|
}
|
|
}
|
|
match fs::remove_file(vector[q_index].path.clone()) {
|
|
Ok(_) => {
|
|
removed_files += 1;
|
|
gained_space += vector[q_index].size;
|
|
}
|
|
Err(_) => {
|
|
failed_to_remove_files += 1;
|
|
warnings.push("Failed to delete".to_string() + &*vector[q_index].path);
|
|
}
|
|
};
|
|
}
|
|
DeleteMethod::AllExceptOldest => {
|
|
for files in vector.iter().enumerate() {
|
|
let time_since_epoch = files.1.created_date.duration_since(UNIX_EPOCH).expect("Invalid file date").as_secs();
|
|
if q_time == 0 || q_time > time_since_epoch {
|
|
q_time = time_since_epoch;
|
|
q_index = files.0;
|
|
}
|
|
}
|
|
for files in vector.iter().enumerate() {
|
|
if q_index != files.0 {
|
|
match fs::remove_file(vector[files.0].path.clone()) {
|
|
Ok(_) => {
|
|
removed_files += 1;
|
|
gained_space += vector[files.0].size;
|
|
}
|
|
Err(_) => {
|
|
failed_to_remove_files += 1;
|
|
warnings.push("Failed to delete".to_string() + &*vector[files.0].path);
|
|
}
|
|
};
|
|
}
|
|
}
|
|
}
|
|
DeleteMethod::AllExceptNewest => {
|
|
for files in vector.iter().enumerate() {
|
|
let time_since_epoch = files.1.created_date.duration_since(UNIX_EPOCH).expect("Invalid file date").as_secs();
|
|
if q_time == 0 || q_time < time_since_epoch {
|
|
q_time = time_since_epoch;
|
|
q_index = files.0;
|
|
}
|
|
}
|
|
for files in vector.iter().enumerate() {
|
|
if q_index != files.0 {
|
|
match fs::remove_file(vector[files.0].path.clone()) {
|
|
Ok(_) => {
|
|
removed_files += 1;
|
|
gained_space += vector[files.0].size;
|
|
}
|
|
Err(_) => {
|
|
failed_to_remove_files += 1;
|
|
warnings.push("Failed to delete".to_string() + &*vector[files.0].path);
|
|
}
|
|
};
|
|
}
|
|
}
|
|
}
|
|
DeleteMethod::None => {
|
|
panic!();
|
|
}
|
|
};
|
|
println!("{} {} {}", gained_space, removed_files, failed_to_remove_files);
|
|
(gained_space, removed_files, failed_to_remove_files)
|
|
}
|