initial version of checksum based freshness #14137

Xaeroxe · 2024-06-25T05:18:22Z

Implementation for #14136 and resolves #6529

This PR implements the use of checksums in cargo fingerprints as an alternative to using mtimes. This is most useful on systems with poor mtime implementations.

This has a dependency on rust-lang/rust#126930. It's expected this will increase the time it takes to declare a build to be fresh. Still this loss in performance may be preferable to the issues the ecosystem has had with the use of mtimes for determining freshness.

rustbot · 2024-06-25T05:18:26Z

Thanks for the pull request, and welcome! The Rust team is excited to review your changes, and you should hear from @weihanglo (or someone else) some time within the next two weeks.

Please see the contribution instructions for more information. Namely, in order to ensure the minimum review times lag, PR authors and assigned reviewers should ensure that the review label (S-waiting-on-review and S-waiting-on-author) stays updated, invoking these commands when appropriate:

@rustbot author: the review is finished, PR author should check the comments and take action accordingly
@rustbot review: the author is ready for a review, this PR will be queued again in the reviewer's queue

Cargo.toml

Add unstable support for outputting file checksums for use in cargo Adds an unstable option that appends file checksums and expected lengths to the end of the dep-info file such that `cargo` can read and use these values as an alternative to file mtimes. This PR powers the changes made in this cargo PR rust-lang/cargo#14137 Here's the tracking issue for the cargo feature rust-lang/cargo#14136.

bors · 2024-07-26T22:03:01Z

☔ The latest upstream changes (presumably #13947) made this pull request unmergeable. Please resolve the merge conflicts.

Xaeroxe · 2024-07-26T22:23:39Z

Merge conflicts resolved.

src/cargo/core/features.rs

src/cargo/util/command_prelude.rs

src/cargo/util/context/mod.rs

epage · 2024-08-01T20:31:14Z

tests/testsuite/freshness_checksum.rs

+use cargo_test_support::{basic_lib_manifest, basic_manifest, project, rustc_host, rustc_host_env};
+
+#[cargo_test]
+fn checksum_actually_uses_checksum() {


At minimum, could you structure your PR so its

A commit with these tests without -Zchecksum-freshness

A commit with the checksum work that also updates the tests to pass -Zchecksum-freshness

A big benefit to this is it shows to reviewers / the community how this feature is comparing to what was being done before

(sometimes, I also break out "adding an unstable feature" into its own commit which is the flag + docs)

Note: I've not dug deep into the tests, waiting on this change

So it's worth noting that freshness_checksum.rs derives very heavily from freshness.rs. So a version of freshness_checksum.rs without the freshness flag would just be a subset of freshness.rs. I'm not sure how this provides new information. There are two tests which are truly unique to freshness_checksum.rs, which are same_size_different_content() and checksum_actually_uses_checksum().

One might debate the merit of duplicating the tests like that. If you really wanted to deduplicate the tests then this would likely require a special case be added to the test runner code.

tests/testsuite/freshness_checksum.rs

src/cargo/core/compiler/fingerprint/mod.rs

bors · 2024-09-10T14:51:28Z

☔ The latest upstream changes (presumably #14493) made this pull request unmergeable. Please resolve the merge conflicts.

weihanglo

Looks like we are almost there! Thank for yours efforts, epage and Xaeroxe. Thought it would be a tough review, but honestly it was a happy time :)

weihanglo · 2024-09-19T20:33:57Z

src/doc/src/reference/unstable.md

+## checksum-freshness
+* Tracking issue: [#14136](https://github.com/rust-lang/cargo/issues/14136)
+
+The `-Z checksum-freshness` flag will replace the use of file mtimes in cargo's


It may be worth noting that build script execution is not included in the current implementation.

weihanglo · 2024-09-19T20:38:10Z

Cargo.toml

@@ -23,6 +23,7 @@ anstream = "0.6.15"
 anstyle = "1.0.8"
 anyhow = "1.0.86"
 base64 = "0.22.1"
+blake3 = "1.5.2"


Note: need to check the compatibility of blake3 for Tier 1 with Host Tools and Tier 2 with Host Tools, as it contains some assembly code.

weihanglo · 2024-09-20T01:53:31Z

src/cargo/core/compiler/fingerprint/mod.rs

+                        });
+                    }
+                    let Ok(checksum) = Checksum::compute(prior_checksum.algo, file) else {
+                        return Some(StaleItem::MissingFile(path.to_path_buf()));


Why is checksum computation failure also a StaleItem::MissingFile?

weihanglo · 2024-09-20T02:03:19Z

src/cargo/core/compiler/fingerprint/mod.rs

+                    let Ok(file) = File::open(path) else {
+                        return Some(StaleItem::MissingFile(path.to_path_buf()));
+                    };
+                    let Ok(current_file_len) = file.metadata().map(|m| m.len()) else {
+                        return Some(StaleItem::FailedToReadMetadata(path.to_path_buf()));
+                    };


Since syscall stat is generally faster than open, should we reorder this part a bit that first compare file size then open it?

weihanglo · 2024-09-20T02:25:29Z

src/cargo/core/compiler/fingerprint/mod.rs

                let dep_info = target_root.join(dep_info);
+                let cargo_exe = cargo_exe;


Suggested change

let cargo_exe = cargo_exe;

weihanglo · 2024-09-20T03:51:47Z

src/cargo/core/compiler/fingerprint/mod.rs

@@ -2250,3 +2479,201 @@ pub fn parse_rustc_dep_info(rustc_dep_info: &Path) -> CargoResult<RustcDepInfo>
        Ok(ret)
    }
 }
+
+/// Some algorithms are here to ensure compatibility with possible rustc outputs.


nit: checksum part could potentially be split into a module. The LoC here in this file is already frightening.

(though I believe it will not help much 😓)

weihanglo · 2024-09-20T04:13:21Z

src/cargo/core/compiler/fingerprint/mod.rs

@@ -2102,13 +2271,14 @@ pub struct RustcDepInfo {
 struct EncodedDepInfo {
    files: Vec<(DepInfoPathType, PathBuf)>,
    env: Vec<(String, Option<String>)>,
+    checksum: Vec<(DepInfoPathType, PathBuf, u64, String)>,


(not a blocker)
I feel like files and checksum should eventually merge into one field.

weihanglo · 2024-09-20T04:14:30Z

src/cargo/core/compiler/fingerprint/dirty_reason.rs

+                                "the file `{}` has changed (checksum didn't match, {} != {})",
+                                file.display(),
+                                stored_checksum,
+                                new_checksum,


nit: arg capture whenver possible and making sense

Suggested change

"the file `{}` has changed (checksum didn't match, {} != {})",

file.display(),

stored_checksum,

new_checksum,

"the file `{}` has changed (checksum didn't match, {stored_checksum} != {new_checksum})",

file.display(),

weihanglo · 2024-09-20T04:16:00Z

src/cargo/core/compiler/fingerprint/dirty_reason.rs

@@ -183,6 +187,16 @@ impl DirtyReason {
            DirtyReason::PrecalculatedComponentsChanged { .. } => {
                s.dirty_because(unit, "the precalculated components changed")
            }
+            DirtyReason::ChecksumUseChanged { old, new: _ } => {


If we are not using new anywhere, should we just remove this field?

weihanglo · 2024-09-20T04:18:01Z

tests/testsuite/freshness_checksum.rs

Oops. The test module might need to be rewritten. See #14039

We can sort this out later as follow-ups. Don't worry

This file looks like a copy of freshness.rs with -Zchecksum-freshness and masquerade_as_nightly_cargo everywhere.

Could you point out which tests are new and worth a review? Guess they are

checksum_actually_uses_checksum

same_size_different_content

modifying_and_moving

Also could you add a test that verifying -Zchecksum-freshness is gated behind nightly? (A test without masquerade_as_nightly_cargo is sufficient.

rustbot assigned weihanglo Jun 25, 2024

This was referenced Jun 25, 2024

Tracking Issue for checksum freshness #14136

Open

Add unstable support for outputting file checksums for use in cargo rust-lang/rust#126930

Open

Xaeroxe force-pushed the checksum-freshness branch from 320f73c to 310cd79 Compare June 25, 2024 05:37

tgross35 reviewed Jun 25, 2024

View reviewed changes

Cargo.toml Outdated Show resolved Hide resolved

Xaeroxe force-pushed the checksum-freshness branch from cce62ba to 59441b6 Compare June 26, 2024 05:55

Xaeroxe mentioned this pull request Jul 1, 2024

MCP: Alternate cargo freshness algorithm, unstable flag to annotate depinfo file with checksums and file sizes rust-lang/compiler-team#765

Closed

3 tasks

rustbot added the A-infrastructure Area: infrastructure around the cargo repo, ci, releases, etc. label Jul 13, 2024

Xaeroxe force-pushed the checksum-freshness branch from 27e2a18 to c83be55 Compare July 13, 2024 14:58

Xaeroxe added 5 commits July 26, 2024 16:23

initial version of checksum based freshness

6f92c69

remove support for md5 and sha1, add support for xxhash

0f12264

switch to using twox-hash instead due to licensing issues

a264c26

update to use blake3 instead of xxhash per rustc MCP resolution

06c2024

add bsd 2 clause to allow list for blake3 dep arrayref

b8c21fa

Xaeroxe force-pushed the checksum-freshness branch from c83be55 to b8c21fa Compare July 26, 2024 22:23