Document and improve (a lot) lintcheck --perf (#14194)

In #14116 we added a benchmarking option for Lintcheck, this commit adds
a new chapter to the book AND improves that option into a more usable
state.

It's recommended to review one commit at a time.

- **Document how to benchmark with lintcheck --perf**
- **Several improvements on lintcheck perf (desc.)**

    - Now lintcheck perf deletes target directory after benchmarking,
benchmarking with a cache isn't very useful or telling of any
precise outcome.

    - Support for benchmarking several times without having to do
a cargo clean.

    - Compress perf.data

changelog: none

Author: Alexendoo

book/src/SUMMARY.md

@@ -30,6 +30,7 @@
         - [Updating the Changelog](development/infrastructure/changelog_update.md)
         - [Release a New Version](development/infrastructure/release.md)
         - [The Clippy Book](development/infrastructure/book.md)
+        - [Benchmarking Clippy](development/infrastructure/benchmarking.md)
     - [Proposals](development/proposals/README.md)
         - [Roadmap 2021](development/proposals/roadmap-2021.md)
         - [Syntax Tree Patterns](development/proposals/syntax-tree-patterns.md)

book/src/development/infrastructure/benchmarking.md

@@ -0,0 +1,55 @@
+# Benchmarking Clippy
+
+Benchmarking Clippy is similar to using our Lintcheck tool, in fact, it even
+uses the same tool! Just by adding a `--perf` flag it will transform Lintcheck
+into a very simple but powerful benchmarking tool!
+
+It requires having the [`perf` tool][perf] installed, as `perf` is what's actually
+profiling Clippy under the hood.
+
+The lintcheck `--perf` tool generates a series of `perf.data` in the
+`target/lintcheck/sources/<package>-<version>` directories. Each `perf.data`
+corresponds to the package which is contained.
+
+Lintcheck uses the `-g` flag, meaning that you can get stack traces for richer
+analysis, including with tools such as [flamegraph][flamegraph-perf]
+(or [`flamegraph-rs`][flamegraph-rs]).
+
+Currently, we only measure instruction count, as it's the most reproducible metric
+and [rustc-perf][rustc-perf] also considers it the main number to focus on.
+
+## Benchmarking a PR
+
+Having a benchmarking tool directly implemented into lintcheck gives us the
+ability to benchmark any given PR just by making a before and after 
+
+Here's the way you can get into any PR, benchmark it, and then benchmark
+`master`.
+
+The first `perf.data` will not have any numbers appended, but any subsequent
+benchmark will be written to `perf.data.number` with a number growing for 0.
+All benchmarks are compressed so that you can
+
+```bash
+git fetch upstream pull/<PR_NUMBER>/head:<BRANCH_NAME>
+git switch BRANCHNAME
+
+# Bench
+cargo lintcheck --perf
+
+# Get last common commit, checkout that
+LAST_COMMIT=$(git log BRANCHNAME..master --oneline | tail -1 | cut -c 1-11)
+git switch -c temporary $LAST_COMMIT
+
+# We're now on master
+
+# Bench
+cargo lintcheck --perf
+perf diff ./target/lintcheck/sources/CRATE/perf.data ./target/lintcheck/sources/CRATE/perf.data.0
+```
+
+
+[perf]: https://perfwiki.github.io/main/
+[flamegraph-perf]: https://github.com/brendangregg/FlameGraph
+[flamegraph-rs]: https://github.com/flamegraph-rs/flamegraph
+[rustc-perf]: https://github.com/rust-lang/rustc-perf

lintcheck/src/main.rs

@@ -120,14 +120,17 @@ impl Crate {
 
         if config.perf {
             cmd = Command::new("perf");
+            let perf_data_filename = get_perf_data_filename(&self.path);
             cmd.args(&[
                 "record",
                 "-e",
                 "instructions", // Only count instructions
                 "-g",           // Enable call-graph, useful for flamegraphs and produces richer reports
                 "--quiet",      // Do not tamper with lintcheck's normal output
+                "--compression-level=22",
+                "--freq=3000", // Slow down program to capture all events
                 "-o",
-                "perf.data",
+                &perf_data_filename,
                 "--",
                 "cargo",
             ]);
@@ -165,7 +168,7 @@ impl Crate {
             return Vec::new();
         }
 
-        if !config.fix {
+        if !config.fix && !config.perf {
             cmd.arg("--message-format=json");
         }
 
@@ -203,6 +206,11 @@ impl Crate {
             return Vec::new();
         }
 
+        // We don't want to keep target directories if benchmarking
+        if config.perf {
+            let _ = fs::remove_dir_all(&shared_target_dir);
+        }
+
         // get all clippy warnings and ICEs
         let mut entries: Vec<ClippyCheckOutput> = Message::parse_stream(stdout.as_bytes())
             .filter_map(|msg| match msg {
@@ -441,6 +449,35 @@ fn lintcheck(config: LintcheckConfig) {
     fs::write(&config.lintcheck_results_path, text).unwrap();
 }
 
+/// Traverse a directory looking for `perf.data.<number>` files, and adds one
+/// to the most recent of those files, returning the new most recent `perf.data`
+/// file name.
+fn get_perf_data_filename(source_path: &Path) -> String {
+    if source_path.join("perf.data").exists() {
+        let mut max_number = 0;
+        fs::read_dir(source_path)
+            .unwrap()
+            .filter_map(Result::ok)
+            .filter(|path| {
+                path.file_name()
+                    .as_os_str()
+                    .to_string_lossy() // We don't care about data loss, as we're checking for equality
+                    .starts_with("perf.data")
+            })
+            .for_each(|path| {
+                let file_name = path.file_name();
+                let file_name = file_name.as_os_str().to_str().unwrap().split('.').next_back().unwrap();
+                if let Ok(parsed_file_name) = file_name.parse::<usize>() {
+                    if parsed_file_name >= max_number {
+                        max_number = parsed_file_name + 1;
+                    }
+                }
+            });
+        return format!("perf.data.{max_number}");
+    }
+    String::from("perf.data")
+}
+
 /// Returns the path to the Clippy project directory
 #[must_use]
 fn clippy_project_root() -> &'static Path {