Download Evaluate/projects/charset-normalizer/rust/src/normalizer.rs from SYSUSELab/RustRepoTrans: direct link, hf CLI and curl.
- Browser
- Download file 6.6 kB
-
https://huggingface.co/datasets/SYSUSELab/RustRepoTrans/resolve/main/Evaluate/projects/charset-normalizer/rust/src/normalizer.rs
- Command line
-
hf download hf://datasets/SYSUSELab/RustRepoTrans/Evaluate/projects/charset-normalizer/rust/src/normalizer.rs
-
curl -L -o normalizer.rs https://huggingface.co/datasets/SYSUSELab/RustRepoTrans/resolve/main/Evaluate/projects/charset-normalizer/rust/src/normalizer.rs
6.6 kB
| use charset_normalizer_rs::entity::{CLINormalizerArgs, CLINormalizerResult, NormalizerSettings}; | |
| use charset_normalizer_rs::from_path; | |
| use clap::Parser; | |
| use dialoguer::Confirm; | |
| use env_logger::Env; | |
| use ordered_float::OrderedFloat; | |
| use std::fs::File; | |
| use std::io::Write; | |
| use std::{fs, process}; | |
| fn normalizer(args: &CLINormalizerArgs) -> Result<i32, String> { | |
| match (args.replace, args.normalize, args.force, args.threshold) { | |
| (true, false, _, _) => return Err("Use --replace in addition to --normalize only.".into()), | |
| (false, _, true, _) => return Err("Use --force in addition to --replace only.".into()), | |
| (_, _, _, threshold) if !(0.0..=1.0).contains(&threshold) => { | |
| return Err("--threshold VALUE should be between 0.0 and 1.0.".into()) | |
| } | |
| _ => {} | |
| } | |
| let mut results: Vec<CLINormalizerResult> = vec![]; | |
| let settings = NormalizerSettings { | |
| threshold: OrderedFloat(args.threshold), | |
| ..Default::default() | |
| }; | |
| // go through the files | |
| for path in &args.files { | |
| let full_path = &mut fs::canonicalize(path).map_err(|err| err.to_string())?; | |
| let matches = from_path(full_path, Some(settings.clone()))?; | |
| match matches.get_best() { | |
| None => { | |
| results.push(CLINormalizerResult { | |
| path: full_path.clone(), | |
| language: "Unknown".to_string(), | |
| chaos: format!("{:.1}", 1.0), | |
| coherence: format!("{:.1}", 0.0), | |
| is_preferred: true, | |
| ..Default::default() | |
| }); | |
| eprintln!( | |
| "Unable to identify originating encoding for {:?}. {}", | |
| full_path, | |
| if args.threshold < 1.0 { | |
| "Maybe try increasing maximum amount of chaos." | |
| } else { | |
| "" | |
| } | |
| ); | |
| } | |
| Some(best_guess) => { | |
| // add main result & alternative results | |
| for m in matches.iter() { | |
| let normalize_result = CLINormalizerResult { | |
| path: full_path.clone(), | |
| encoding: Some(m.encoding().to_string()), | |
| encoding_aliases: m | |
| .encoding_aliases() | |
| .iter() | |
| .map(|s| (*s).to_string()) | |
| .collect(), | |
| alternative_encodings: m | |
| .suitable_encodings() | |
| .iter() | |
| .filter(|&e| e != m.encoding()) | |
| .cloned() | |
| .collect(), | |
| language: format!("{}", m.most_probably_language()), | |
| alphabets: m.unicode_ranges(), | |
| has_sig_or_bom: m.bom(), | |
| chaos: format!("{:.1}", m.chaos_percents()), | |
| coherence: format!("{:.1}", m.coherence_percents()), | |
| unicode_path: None, | |
| is_preferred: true, | |
| }; | |
| if m == best_guess { | |
| results.insert(0, normalize_result); | |
| } else if args.alternatives { | |
| results.push(normalize_result); | |
| } else { | |
| break; | |
| } | |
| } | |
| // normalizing if need | |
| if args.normalize { | |
| if best_guess.encoding().starts_with("utf") { | |
| eprintln!( | |
| "{:?} file does not need to be normalized, as it already came from unicode.", | |
| full_path, | |
| ); | |
| continue; | |
| } | |
| // force or confirm of replacement | |
| if !args.replace { | |
| let filename = full_path.file_name().unwrap().to_str().unwrap(); | |
| let filename = match filename.rsplit_once('.') { | |
| None => format!("{}.{}", filename, best_guess.encoding()), | |
| Some(split) => { | |
| format!("{}.{}.{}", split.0, best_guess.encoding(), split.1) | |
| } | |
| }; | |
| full_path.set_file_name(filename); | |
| } else if !args.force | |
| && !Confirm::new() | |
| .with_prompt(format!( | |
| "Are you sure to normalize {:?} by replacing it?", | |
| full_path, | |
| )) | |
| .interact() | |
| .unwrap_or(false) | |
| { | |
| continue; | |
| } | |
| // save path to result | |
| results[0].unicode_path = Some(full_path.clone()); | |
| // replace file contents | |
| if let Err(err) = File::create(full_path).and_then(|mut file| { | |
| file.write_all(best_guess.decoded_payload().unwrap().as_bytes()) | |
| }) { | |
| return Err(err.to_string()); | |
| } | |
| } | |
| } | |
| } | |
| } | |
| // print out results | |
| if args.minimal { | |
| for path in &args.files { | |
| let full_path = fs::canonicalize(path).map_err(|err| err.to_string())?; | |
| println!( | |
| "{}", | |
| results | |
| .iter() | |
| .filter(|r| r.path == full_path) | |
| .map(|r| r.encoding.clone().unwrap_or("undefined".to_string())) | |
| .collect::<Vec<_>>() | |
| .join(", ") | |
| ); | |
| } | |
| } else { | |
| println!( | |
| "{}", | |
| if results.len() > 1 { | |
| serde_json::to_string_pretty(&results).unwrap() | |
| } else { | |
| serde_json::to_string_pretty(&results[0]).unwrap() | |
| } | |
| ); | |
| } | |
| Ok(0) | |
| } | |
| pub fn main() { | |
| let args = CLINormalizerArgs::parse(); | |
| // verbose mode | |
| if args.verbose { | |
| env_logger::Builder::from_env(Env::default().default_filter_or("trace")).init(); | |
| } | |
| // run normalizer | |
| match normalizer(&args) { | |
| Err(e) => panic!("{e}"), | |
| Ok(exit_code) => process::exit(exit_code), | |
| } | |
| } | |