feat(report): cargo report timings HTML replay #16377

weihanglo · 2025-12-11T06:04:20Z

What does this PR try to resolve?

This implements cargo report timings HTML replay feature.

The help text looks like:

Reports the build timings of previous builds (unstable)

Usage: cargo report timings [OPTIONS]

Options:
      --open              Opens the timing report in a browser
      ...

Part of #15844

How to test and review this PR?

Open questions:

The replayed HTML report will be written to
- If in a Cargo workspace, write to <artifact-dir>/cargo-timings/
- Otherwise, write to a temp directory
The log discovery logic
- Currently look at the newest log file for the workspace.
- If not in a workspace, pick the newest log file in the log directory.
Snapshots HTML report and log files are added.
- They are huge in size. Should we not do this?
- Benefit: HTML snapshots help us know better when it is broken

Things left (but should do it follow-up PRs)

cargo help hasn't yet worked with with nested commands.
There are some more information required by HTML report, but these don't affect the major use of the report IMO.
- fresh/dirty/total units
- information of root units
- enabled features for each unit
- --target platform
- CPU usage metrics
- number of available parallelism fix(timing): more self-contained timing/log data #16378

rustbot · 2025-12-11T06:04:25Z

r? @epage

rustbot has assigned @epage.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

weihanglo · 2025-12-11T06:07:57Z

...argo_report_timings/html_snapshot/out/cargo-timing-20060724T012128000Z-b0fd440798ab3cfb.html

Preview link: https://refined-github-html-preview.kidonng.workers.dev/weihanglo/cargo/raw/ead4b9c2d5a86fab6a7849ecebee7d790be23e64/tests/testsuite/cargo_report_timings/log_file_specified/out/target/cargo-timings/cargo-timing-2006-07-24T012128000Z-b0fd440798ab3cfb.html

Are you running a mac and have that gatekeeper performance issue?

Interesting to see the impact of linking of build scripts

Yeah but not my personal machine 😞.
(technically possible but may not be allowed to)

src/cargo/core/compiler/build_config.rs

weihanglo · 2025-12-11T06:12:58Z

src/cargo/ops/cargo_report/timings.rs

+                // This is pretty similar to how the current `core::compiler::timings`
+                // renders `core::manifest::Target`. However, our target is
+                // a simplified type so we cannot reuse the same logic here.
+                let mut target_str = if target.kind == "lib" && mode == CompileMode::Build {
+                    // Special case for brevity, since most dependencies hit this path.
+                    "".to_string()
+                } else if target.kind == "build-script" {
+                    " build-script".to_string()
+                } else {
+                    format!(r#" {} "{}""#, target.name, target.kind)
+                };
+
+                match mode {
+                    CompileMode::Test => target_str.push_str(" (test)"),
+                    CompileMode::Build => {}
+                    CompileMode::Check { test: true } => target_str.push_str(" (check-test)"),
+                    CompileMode::Check { test: false } => target_str.push_str(" (check)"),
+                    CompileMode::Doc { .. } => target_str.push_str(" (doc)"),
+                    CompileMode::Doctest => target_str.push_str(" (doc test)"),
+                    CompileMode::Docscrape => target_str.push_str(" (doc scrape)"),
+                    CompileMode::RunCustomBuild => target_str.push_str(" (run)"),
+                }


We probably want to reconsider using the old cargo::core::Target struct instead of the new simplified one.

Or better, HTML from --timings should be generated by in-memory log message, so that we don't need to have two parallel implementations doing the same thing.

Or better, HTML from --timings should be generated by in-memory log message, so that we don't need to have two parallel implementations doing the same thing.

Would love to switch to something like that. Also, working to deprecate --timings

weihanglo · 2025-12-11T06:16:36Z

src/cargo/ops/cargo_report/timings.rs

+        TempDir::with_prefix("cargo-timings-")?.keep()
+    };
+
+    let timing_path = reports_dir.join(format!("cargo-timing-{run_id}.html"));


We might also want to change the output HTML path of --timings, though I am not sure if it is a part of stable interface. I guess not?

cargo-timing-20251211T043934.741441Z.html -> --timing timestamp-based file name

cargo-timing-2025-12-11T044138699Z-a2f1dccc7ac1667a.html -> RunId-based file name

..._specified/out/target/cargo-timings/cargo-timing-2006-07-24T012128000Z-b0fd440798ab3cfb.html

eth3lbert · 2025-12-11T13:52:54Z

I haven't taken a review yet, and this is not a blocker, but I'm wondering if it's possible to provide an HTML page that allows users to select a JSON(L) file for rendering (something similar to profiler.firefox.com or speedscope). This would be a more general solution that separates the presentation from the data, preventing the data from being embedded directly in the HTML. The implementation probably wouldn't need to be in Cargo itself. Some drawbacks are that it might require migrating some of the logic from Rust to frontend, and we'd likely need some setup to make it testable.

epage · 2025-12-11T15:17:02Z

I haven't taken a review yet, and this is not a blocker, but I'm wondering if it's possible to provide an HTML page that allows users to select a JSON(L) file for rendering (something similar to profiler.firefox.com or speedscope). This would be a more general solution that separates the presentation from the data, preventing the data from being embedded directly in the HTML. The implementation probably wouldn't need to be in Cargo itself. Some drawbacks are that it might require migrating some of the logic from Rust to frontend, and we'd likely need some setup to make it testable.

This would be dependent on stabilizing the json

weihanglo · 2025-12-11T15:34:57Z

…but I'm wondering if it's possible to provide an HTML page that allows users to select a JSON(L) file for rendering (something similar to profiler.firefox.com or speedscope)…

This will be nice and actually the entire prepare_context can be ported to JavaScript. Though like epage said, we need a stable interface that Cargo controls. I would love to see --timing entirely generated from in-memory log message on stable Cargo (this doesn't requires JSON schema stabilization, more a internal refactor #16377 (comment))

src/cargo/util/logger.rs

src/bin/cargo/commands/report.rs

src/cargo/ops/cargo_report/timings.rs

src/bin/cargo/commands/report.rs

src/cargo/ops/cargo_report/timings.rs

.ignore

### What does this PR try to resolve? Factored out of <#16377>. This makes data more self-contained. ### How to test and review this PR?

src/cargo/ops/cargo_report/timings.rs

tests/testsuite/cargo_report_timings/mod.rs

Currently look at the newest log file for the workspace. If not in a workspace, pick the newest log file in the log directory.

This implements `cargo report timings` HTML replay feature. The help text looks like: ``` Reports the build timings of previous builds (unstable) Usage: cargo report timings [OPTIONS] Options: --open Opens the timing report in a browser ... ``` The replayed HTML report will be written to * If in a Cargo workspace, write to `<artifact-dir>/cargo-timings/` * Otherwise, write to a temp directory The log discovery logic * Currently look at the newest log file for the workspace. * If not in a workspace, pick the newest log file in the log directory.

This adds a log file generated by ``` cargod build --timings -Zsection-timings -p cargo-util-schemas ``` With this we can track what changed in HTML report.

rustbot · 2025-12-12T22:43:36Z

This PR was rebased onto a different master commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

epage

Thanks!

View changes since this review

Update cargo submodule 27 commits in 2c283a9a5c5968eeb9a8f12313f04feb1ff8dfac..c46423de7351e3c4c734b2faa86088a9f5d1302b 2025-12-04 16:47:28 +0000 to 2025-12-12 23:16:12 +0000 - feat(report): cargo report timings HTML replay (rust-lang/cargo#16377) - feat: stabilize `-Zconfig-include` (rust-lang/cargo#16284) - fix(package): Don't verify registry for --list (rust-lang/cargo#16341) - Fixed incorrect locking logic when artifact-dir == build-dir (rust-lang/cargo#16385) - feat(log): make timing messages ready for HTML replay (rust-lang/cargo#16382) - chore(deps): update msrv (1 version) to v1.92 (rust-lang/cargo#16381) - Downgrade curl-sys to 0.4.83 (rust-lang/cargo#16379) - fix(timing): more self-contained timing/log data (rust-lang/cargo#16378) - test: update to `proc_macro::tracked::path` (rust-lang/cargo#16380) - refactor(lint): move lints to separate modules (rust-lang/cargo#16364) - fix(index): Apply feedback from Cargo team (rust-lang/cargo#16369) - fix(lints): handle lints separately at ws pkg level (rust-lang/cargo#16367) - feat(lint): new `implicit_minimum_version_req` lint (rust-lang/cargo#16321) - fix(info): default to local without explicit reg (rust-lang/cargo#16358) - Remove `[no-mentions]` handler in our triagebot config (rust-lang/cargo#16361) - Don't read the config file twice when $CARGO_HOME is a symlink (rust-lang/cargo#16325) - fix(timings): forgot to negate filter (rust-lang/cargo#16352) - fix(doctest): Include all search paths with new build layout (rust-lang/cargo#16348) - fix(layout): Remove hashes from bins in new layout (rust-lang/cargo#16351) - docs(faq): Include an entry on disk space (rust-lang/cargo#16349) - feat(timings): derive concurrency data from unit data (rust-lang/cargo#16350) - perf(layout): Use unit_id, not pkg hash, for bin/lib pkg_dirs for new layout (rust-lang/cargo#16345) - Validate target source paths before compilation with clearer errors (rust-lang/cargo#16338) - test(doc): Remove unused build script (rust-lang/cargo#16344) - refactor(timings): store UnitData in RenderContext instead (rust-lang/cargo#16346) - perf(clean): Optimize (legacy) clean with multiple -p specifiers (rust-lang/cargo#16264) - test: Adjust output for out-of-tree build-dir (rust-lang/cargo#16343)

Update cargo submodule 29 commits in 2c283a9a5c5968eeb9a8f12313f04feb1ff8dfac..e91b2baa632c0c7e84216c91ecfe107c37d887c1 2025-12-04 16:47:28 +0000 to 2025-12-13 16:29:21 +0000 - refactor(lints): move from cargo::util::lints to cargo::lints (rust-lang/cargo#16392) - test(lint): redact more due to line got omitted (rust-lang/cargo#16391) - feat(report): cargo report timings HTML replay (rust-lang/cargo#16377) - feat: stabilize `-Zconfig-include` (rust-lang/cargo#16284) - fix(package): Don't verify registry for --list (rust-lang/cargo#16341) - Fixed incorrect locking logic when artifact-dir == build-dir (rust-lang/cargo#16385) - feat(log): make timing messages ready for HTML replay (rust-lang/cargo#16382) - chore(deps): update msrv (1 version) to v1.92 (rust-lang/cargo#16381) - Downgrade curl-sys to 0.4.83 (rust-lang/cargo#16379) - fix(timing): more self-contained timing/log data (rust-lang/cargo#16378) - test: update to `proc_macro::tracked::path` (rust-lang/cargo#16380) - refactor(lint): move lints to separate modules (rust-lang/cargo#16364) - fix(index): Apply feedback from Cargo team (rust-lang/cargo#16369) - fix(lints): handle lints separately at ws pkg level (rust-lang/cargo#16367) - feat(lint): new `implicit_minimum_version_req` lint (rust-lang/cargo#16321) - fix(info): default to local without explicit reg (rust-lang/cargo#16358) - Remove `[no-mentions]` handler in our triagebot config (rust-lang/cargo#16361) - Don't read the config file twice when $CARGO_HOME is a symlink (rust-lang/cargo#16325) - fix(timings): forgot to negate filter (rust-lang/cargo#16352) - fix(doctest): Include all search paths with new build layout (rust-lang/cargo#16348) - fix(layout): Remove hashes from bins in new layout (rust-lang/cargo#16351) - docs(faq): Include an entry on disk space (rust-lang/cargo#16349) - feat(timings): derive concurrency data from unit data (rust-lang/cargo#16350) - perf(layout): Use unit_id, not pkg hash, for bin/lib pkg_dirs for new layout (rust-lang/cargo#16345) - Validate target source paths before compilation with clearer errors (rust-lang/cargo#16338) - test(doc): Remove unused build script (rust-lang/cargo#16344) - refactor(timings): store UnitData in RenderContext instead (rust-lang/cargo#16346) - perf(clean): Optimize (legacy) clean with multiple -p specifiers (rust-lang/cargo#16264) - test: Adjust output for out-of-tree build-dir (rust-lang/cargo#16343)

weihanglo added the Z-build-analysis Nightly: build-analysis label Dec 11, 2025

rustbot assigned epage Dec 11, 2025

weihanglo mentioned this pull request Dec 11, 2025

Tracking Issue for -Zbuild-analysis #15844

Open

23 tasks

weihanglo commented Dec 11, 2025

View reviewed changes

weihanglo force-pushed the report-timings branch from ead4b9c to 078ccc9 Compare December 11, 2025 06:22

weihanglo commented Dec 11, 2025

View reviewed changes

..._specified/out/target/cargo-timings/cargo-timing-2006-07-24T012128000Z-b0fd440798ab3cfb.html Outdated Show resolved Hide resolved

weihanglo force-pushed the report-timings branch from 078ccc9 to fb9da8d Compare December 11, 2025 15:12

weihanglo mentioned this pull request Dec 11, 2025

fix(timing): more self-contained timing/log data #16378

Merged