-Ztime-passes: print lint name

[matthiaskrgr]

It seems that for each lint that is run, rustc only prints pre_AST_expansion_lint_checks when invoked with -Ztime-passes:

fn pre_expansion_lint(sess: &Session, lint_store: &LintStore, krate: &ast::Crate) {
    sess.time("pre_AST_expansion_lint_checks", || {
        rustc_lint::check_ast_crate(
            sess,
            lint_store,
            &krate,
            true,
            None,
            rustc_lint::BuiltinCombinedPreExpansionLintPass::new(),
        );
    });
}

time: 0.000; rss: 62MB	plugin_registration
time: 0.000; rss: 62MB	pre_AST_expansion_lint_checks
time: 0.000; rss: 65MB	crate_injection
time: 0.000; rss: 65MB	pre_AST_expansion_lint_checks
time: 0.000; rss: 65MB	pre_AST_expansion_lint_checks
time: 0.001; rss: 68MB	pre_AST_expansion_lint_checks
time: 0.000; rss: 69MB	pre_AST_expansion_lint_checks
time: 0.000; rss: 69MB	pre_AST_expansion_lint_checks
....

It would be great if it could print the name of the lint here.

[jyn514]

Honestly I'd prefer it just not print the lint at all. They almost all take basically 0 time and they just clutter up the logs.

[matthiaskrgr]

Hm. I was hoping to be able to find out if clippy had some lints that took exceptionally long. There are some cases where linting a single file crate can take several minutes iirc.

[jyn514]

Another alternative is to make self-profile smarter and only prints passes that take more than x milliseconds (.005 seconds seems like a reasonable cutoff). Then most lines would be trimmed but long running lints would still show up.

[jyn514]

ah, @matthiaskrgr it turns out it's the other way around - it prints pre_AST_expansion_lint_checks for each AST node, not for each lint. So it doesn't have enough information here to tell you which lint is slow, you'd have to change the order of the loops for that.

[jyn514]

You can use -Zno-interleave-lints to get the lint name:

time:   0.000; rss:   88MB ->   88MB (   +0MB)  run_lint(UnusedParens)
time:   0.000; rss:   88MB ->   88MB (   +0MB)  run_lint(UnusedBraces)
time:   0.000; rss:   88MB ->   88MB (   +0MB)  run_lint(UnusedImportBraces)
time:   0.000; rss:   88MB ->   88MB (   +0MB)  run_lint(UnsafeCode)
time:   0.000; rss:   88MB ->   88MB (   +0MB)  run_lint(AnonymousParameters)
time:   0.000; rss:   88MB ->   88MB (   +0MB)  run_lint(EllipsisInclusiveRangePatterns)
time:   0.000; rss:   88MB ->   88MB (   +0MB)  run_lint(NonCamelCaseTypes)
time:   0.000; rss:   88MB ->   88MB (   +0MB)  run_lint(DeprecatedAttr)
time:   0.000; rss:   88MB ->   88MB (   +0MB)  run_lint(WhileTrue)
time:   0.000; rss:   88MB ->   88MB (   +0MB)  run_lint(NonAsciiIdents)
time:   0.000; rss:   88MB ->   88MB (   +0MB)  run_lint(HiddenUnicodeCodepoints)
time:   0.000; rss:   88MB ->   88MB (   +0MB)  run_lint(IncompleteFeatures)
time:   0.000; rss:   88MB ->   88MB (   +0MB)  run_lint(RedundantSemicolons)
time:   0.000; rss:   88MB ->   88MB (   +0MB)  run_lint(UnusedDocComment)

It's not particularly useful for -Ztime-lints because you end up with lots of duplicate names - I think summarize + -Zself-profile should be able to aggregate them, but I haven't yet found how. So this should probably be moved to https://github.com/rust-lang/rustc-perf.