rust/crates/rust-analyzer/src/cli/analysis_stats.rs

//! Fully type-check project and print various stats, like the number of type
//! errors.

use std::{
    path::PathBuf,
    time::{SystemTime, UNIX_EPOCH},
};

use hir::{
    db::{AstDatabase, DefDatabase, HirDatabase},
    AssocItem, Crate, HasSource, HirDisplay, ModuleDef,
};
use hir_def::FunctionId;
use hir_ty::{Ty, TypeWalk};
use ide_db::base_db::{
    salsa::{self, ParallelDatabase},
    SourceDatabaseExt,
};
use itertools::Itertools;
use oorandom::Rand32;
use rayon::prelude::*;
use rustc_hash::FxHashSet;
use stdx::format_to;
use syntax::AstNode;

use crate::cli::{
    load_cargo::load_cargo, print_memory_usage, progress_report::ProgressReport, report_metric,
    Result, Verbosity,
};
use profile::StopWatch;

/// Need to wrap Snapshot to provide `Clone` impl for `map_with`
struct Snap<DB>(DB);
impl<DB: ParallelDatabase> Clone for Snap<salsa::Snapshot<DB>> {
    fn clone(&self) -> Snap<salsa::Snapshot<DB>> {
        Snap(self.0.snapshot())
    }
}

pub struct AnalysisStatsCmd {
    pub randomize: bool,
    pub parallel: bool,
    pub memory_usage: bool,
    pub only: Option<String>,
    pub with_deps: bool,
    pub path: PathBuf,
    pub load_output_dirs: bool,
    pub with_proc_macro: bool,
}

impl AnalysisStatsCmd {
    pub fn run(self, verbosity: Verbosity) -> Result<()> {
        let mut rng = {
            let seed = SystemTime::now().duration_since(UNIX_EPOCH).unwrap().as_millis() as u64;
            Rand32::new(seed)
        };

        let mut db_load_sw = self.stop_watch();
        let (host, vfs) = load_cargo(&self.path, self.load_output_dirs, self.with_proc_macro)?;
        let db = host.raw_database();
        eprintln!("{:<20} {}", "Database loaded:", db_load_sw.elapsed());

        let mut analysis_sw = self.stop_watch();
        let mut num_crates = 0;
        let mut visited_modules = FxHashSet::default();
        let mut visit_queue = Vec::new();

        let mut krates = Crate::all(db);
        if self.randomize {
            shuffle(&mut rng, &mut krates);
        }
        for krate in krates {
            let module = krate.root_module(db);
            let file_id = module.definition_source(db).file_id;
            let file_id = file_id.original_file(db);
            let source_root = db.file_source_root(file_id);
            let source_root = db.source_root(source_root);
            if !source_root.is_library || self.with_deps {
                num_crates += 1;
                visit_queue.push(module);
            }
        }

        if self.randomize {
            shuffle(&mut rng, &mut visit_queue);
        }

        eprint!("  crates: {}", num_crates);
        let mut num_decls = 0;
        let mut funcs = Vec::new();
        while let Some(module) = visit_queue.pop() {
            if visited_modules.insert(module) {
                visit_queue.extend(module.children(db));

                for decl in module.declarations(db) {
                    num_decls += 1;
                    if let ModuleDef::Function(f) = decl {
                        funcs.push(f);
                    }
                }

                for impl_def in module.impl_defs(db) {
                    for item in impl_def.items(db) {
                        num_decls += 1;
                        if let AssocItem::Function(f) = item {
                            funcs.push(f);
                        }
                    }
                }
            }
        }
        eprintln!(", mods: {}, decls: {}, fns: {}", visited_modules.len(), num_decls, funcs.len());
        eprintln!("{:<20} {}", "Item Collection:", analysis_sw.elapsed());

        if self.randomize {
            shuffle(&mut rng, &mut funcs);
        }

        let mut bar = match verbosity {
            Verbosity::Quiet | Verbosity::Spammy => ProgressReport::hidden(),
            _ if self.parallel => ProgressReport::hidden(),
            _ => ProgressReport::new(funcs.len() as u64),
        };

        if self.parallel {
            let mut inference_sw = self.stop_watch();
            let snap = Snap(db.snapshot());
            funcs
                .par_iter()
                .map_with(snap, |snap, &f| {
                    let f_id = FunctionId::from(f);
                    snap.0.body(f_id.into());
                    snap.0.infer(f_id.into());
                })
                .count();
            eprintln!("{:<20} {}", "Parallel Inference:", inference_sw.elapsed());
        }

        let mut inference_sw = self.stop_watch();
        bar.tick();
        let mut num_exprs = 0;
        let mut num_exprs_unknown = 0;
        let mut num_exprs_partially_unknown = 0;
        let mut num_type_mismatches = 0;
        for f in funcs {
            let name = f.name(db);
            let full_name = f
                .module(db)
                .path_to_root(db)
                .into_iter()
                .rev()
                .filter_map(|it| it.name(db))
                .chain(Some(f.name(db)))
                .join("::");
            if let Some(only_name) = self.only.as_deref() {
                if name.to_string() != only_name && full_name != only_name {
                    continue;
                }
            }
            let mut msg = format!("processing: {}", full_name);
            if verbosity.is_verbose() {
                if let Some(src) = f.source(db) {
                    let original_file = src.file_id.original_file(db);
                    let path = vfs.file_path(original_file);
                    let syntax_range = src.value.syntax().text_range();
                    format_to!(msg, " ({} {:?})", path, syntax_range);
                }
            }
            if verbosity.is_spammy() {
                bar.println(msg.to_string());
            }
            bar.set_message(&msg);
            let f_id = FunctionId::from(f);
            let body = db.body(f_id.into());
            let inference_result = db.infer(f_id.into());
            let (previous_exprs, previous_unknown, previous_partially_unknown) =
                (num_exprs, num_exprs_unknown, num_exprs_partially_unknown);
            for (expr_id, _) in body.exprs.iter() {
                let ty = &inference_result[expr_id];
                num_exprs += 1;
                if let Ty::Unknown = ty {
                    num_exprs_unknown += 1;
                } else {
                    let mut is_partially_unknown = false;
                    ty.walk(&mut |ty| {
                        if let Ty::Unknown = ty {
                            is_partially_unknown = true;
                        }
                    });
                    if is_partially_unknown {
                        num_exprs_partially_unknown += 1;
                    }
                }
                if self.only.is_some() && verbosity.is_spammy() {
                    // in super-verbose mode for just one function, we print every single expression
                    let (_, sm) = db.body_with_source_map(f_id.into());
                    let src = sm.expr_syntax(expr_id);
                    if let Ok(src) = src {
                        let node = {
                            let root = db.parse_or_expand(src.file_id).unwrap();
                            src.value.to_node(&root)
                        };
                        let original_file = src.file_id.original_file(db);
                        let line_index = host.analysis().file_line_index(original_file).unwrap();
                        let text_range = node.syntax().text_range();
                        let (start, end) = (
                            line_index.line_col(text_range.start()),
                            line_index.line_col(text_range.end()),
                        );
                        bar.println(format!(
                            "{}:{}-{}:{}: {}",
                            start.line + 1,
                            start.col_utf16,
                            end.line + 1,
                            end.col_utf16,
                            ty.display(db)
                        ));
                    } else {
                        bar.println(format!("unknown location: {}", ty.display(db)));
                    }
                }
                if let Some(mismatch) = inference_result.type_mismatch_for_expr(expr_id) {
                    num_type_mismatches += 1;
                    if verbosity.is_verbose() {
                        let (_, sm) = db.body_with_source_map(f_id.into());
                        let src = sm.expr_syntax(expr_id);
                        if let Ok(src) = src {
                            // FIXME: it might be nice to have a function (on Analysis?) that goes from Source<T> -> (LineCol, LineCol) directly
                            // But also, we should just turn the type mismatches into diagnostics and provide these
                            let root = db.parse_or_expand(src.file_id).unwrap();
                            let node = src.map(|e| e.to_node(&root).syntax().clone());
                            let original_range = node.as_ref().original_file_range(db);
                            let path = vfs.file_path(original_range.file_id);
                            let line_index =
                                host.analysis().file_line_index(original_range.file_id).unwrap();
                            let text_range = original_range.range;
                            let (start, end) = (
                                line_index.line_col(text_range.start()),
                                line_index.line_col(text_range.end()),
                            );
                            bar.println(format!(
                                "{} {}:{}-{}:{}: Expected {}, got {}",
                                path,
                                start.line + 1,
                                start.col_utf16,
                                end.line + 1,
                                end.col_utf16,
                                mismatch.expected.display(db),
                                mismatch.actual.display(db)
                            ));
                        } else {
                            bar.println(format!(
                                "{}: Expected {}, got {}",
                                name,
                                mismatch.expected.display(db),
                                mismatch.actual.display(db)
                            ));
                        }
                    }
                }
            }
            if verbosity.is_spammy() {
                bar.println(format!(
                    "In {}: {} exprs, {} unknown, {} partial",
                    full_name,
                    num_exprs - previous_exprs,
                    num_exprs_unknown - previous_unknown,
                    num_exprs_partially_unknown - previous_partially_unknown
                ));
            }
            bar.inc(1);
        }
        bar.finish_and_clear();
        eprintln!(
            "  exprs: {}, ??ty: {} ({}%), ?ty: {} ({}%), !ty: {}",
            num_exprs,
            num_exprs_unknown,
            percentage(num_exprs_unknown, num_exprs),
            num_exprs_partially_unknown,
            percentage(num_exprs_partially_unknown, num_exprs),
            num_type_mismatches
        );
        report_metric("unknown type", num_exprs_unknown, "#");
        report_metric("type mismatches", num_type_mismatches, "#");

        eprintln!("{:<20} {}", "Inference:", inference_sw.elapsed());

        let total_span = analysis_sw.elapsed();
        eprintln!("{:<20} {}", "Total:", total_span);
        report_metric("total time", total_span.time.as_millis() as u64, "ms");
        if let Some(instructions) = total_span.instructions {
            report_metric("total instructions", instructions, "#instr");
        }
        if let Some(memory) = total_span.memory {
            report_metric("total memory", memory.allocated.megabytes() as u64, "MB");
        }

        if self.memory_usage && verbosity.is_verbose() {
            print_memory_usage(host, vfs);
        }

        Ok(())
    }

    fn stop_watch(&self) -> StopWatch {
        StopWatch::start().memory(self.memory_usage)
    }
}

fn shuffle<T>(rng: &mut Rand32, slice: &mut [T]) {
    for i in 0..slice.len() {
        randomize_first(rng, &mut slice[i..]);
    }

    fn randomize_first<T>(rng: &mut Rand32, slice: &mut [T]) {
        assert!(!slice.is_empty());
        let idx = rng.rand_range(0..slice.len() as u32) as usize;
        slice.swap(0, idx);
    }
}

fn percentage(n: u64, total: u64) -> u64 {
    (n * 100).checked_div(total).unwrap_or(100)
}