Cache invalidation bug with rust-analyzer's rustc wrapper + Cargo interaction

[sunshowers]

Hi there!

There's a pretty bad cache invalidation bug currently present with rust-analyzer's rustc wrapper, where the wrapper skips over a cargo check but Cargo considers it up-to-date.

Steps to reproduce

1. Clone this repo and cd into it. The repo itself is trivial -- just what cargo init generates -- but I'm providing it just to be able to refer to it below.

2. Run cargo check -v -- for me it produces (paths might be different for you):

% cargo check -v
    Checking ra-rustc-wrapper-bug v0.1.0 (/home/rain/dev/ra-rustc-wrapper-bug)
   Running `/opt/rustup/toolchains/stable-x86_64-unknown-linux-gnu/bin/rustc --crate-name ra_rustc_wrapper_bug --edition=2024 src/main.rs --error-format=json --json=diagnostic-rendered-ansi,artifacts,future-incompat --diagnostic-width=180 --crate-type bin --emit=dep-info,metadata -C embed-bitcode=no -C debuginfo=2 --check-cfg 'cfg(docsrs,test)' --check-cfg 'cfg(feature, values())' -C metadata=ff2adbd4766bc495 -C extra-filename=-19dfec7f77f210bd --out-dir /home/rain/dev/ra-rustc-wrapper-bug/target/debug/deps -C incremental=/home/rain/dev/ra-rustc-wrapper-bug/target/debug/incremental -L dependency=/home/rain/dev/ra-rustc-wrapper-bug/target/debug/deps`
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.02s

3. Run cargo check -v again to ensure that the build is fresh:

% cargo check -v
       Fresh ra-rustc-wrapper-bug v0.1.0 (/home/rain/dev/ra-rustc-wrapper-bug)
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.00s

4. Check the .rmeta file (the path might be different for you):

% stat target/debug/deps/libra_rustc_wrapper_bug-19dfec7f77f210bd.rmeta
  File: target/debug/deps/libra_rustc_wrapper_bug-19dfec7f77f210bd.rmeta
  Size: 0               Blocks: 0          IO Block: 4096   regular empty file
Device: 259,4   Inode: 4823540     Links: 1
Access: (0644/-rw-r--r--)  Uid: ( 1000/    rain)   Gid: ( 1000/  ubuntu)
Access: 2025-07-21 19:38:17.245213032 +0000
Modify: 2025-07-21 19:38:17.245213032 +0000
Change: 2025-07-21 19:38:17.245213032 +0000
 Birth: -

5. Run touch src/main.rs -- this should cause Cargo's cache to be invalidated.
6. Run RA_RUSTC_WRAPPER=1 RUSTC_WRAPPER=~/dev/rust-analyzer/target/debug/rust-analyzer cargo check -v:

% RA_RUSTC_WRAPPER=1 RUSTC_WRAPPER=~/dev/rust-analyzer/target/debug/rust-analyzer cargo check -v
       Dirty ra-rustc-wrapper-bug v0.1.0 (/home/rain/dev/ra-rustc-wrapper-bug): the file `src/main.rs` has changed (1753126805.099594533s, 1m 48s after last build at 1753126697.231212978s)
    Checking ra-rustc-wrapper-bug v0.1.0 (/home/rain/dev/ra-rustc-wrapper-bug)
     Running `/home/rain/dev/rust-analyzer/target/debug/rust-analyzer /opt/rustup/toolchains/stable-x86_64-unknown-linux-gnu/bin/rustc --crate-name ra_rustc_wrapper_bug --edition=2024 src/main.rs --error-format=json --json=diagnostic-rendered-ansi,artifacts,future-incompat --diagnostic-width=180 --crate-type bin --emit=dep-info,metadata -C embed-bitcode=no -C debuginfo=2 --check-cfg 'cfg(docsrs,test)' --check-cfg 'cfg(feature, values())' -C metadata=ff2adbd4766bc495 -C extra-filename=-19dfec7f77f210bd --out-dir /home/rain/dev/ra-rustc-wrapper-bug/target/debug/deps -C incremental=/home/rain/dev/ra-rustc-wrapper-bug/target/debug/incremental -L dependency=/home/rain/dev/ra-rustc-wrapper-bug/target/debug/deps`
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.05s

Note that this build does not cause the output .rmeta to be updated.

7. Run cargo check -v again:

% cargo check -v
       Fresh ra-rustc-wrapper-bug v0.1.0 (/home/rain/dev/ra-rustc-wrapper-bug)
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.02s

Cargo now believes the build is fresh, but the .rmeta is still the same one from before!

% stat target/debug/deps/libra_rustc_wrapper_bug-19dfec7f77f210bd.rmeta
  File: target/debug/deps/libra_rustc_wrapper_bug-19dfec7f77f210bd.rmeta
  Size: 0               Blocks: 0          IO Block: 4096   regular empty file
Device: 259,4   Inode: 4823540     Links: 1
Access: (0644/-rw-r--r--)  Uid: ( 1000/    rain)   Gid: ( 1000/  ubuntu)
Access: 2025-07-21 19:38:17.245213032 +0000
Modify: 2025-07-21 19:38:17.245213032 +0000
Change: 2025-07-21 19:38:17.245213032 +0000
 Birth: -

Why this happens

Cargo uses mtime-based cache invalidation, through the following scheme (paths might be different for you):

• There's a fingerprint file: target/debug/.fingerprint/ra-rustc-wrapper-bug-19dfec7f77f210bd/dep-bin-ra-rustc-wrapper-bug. This fingerprint file contains the list of source files (in this case src/main.rs).
• Cargo compares the mtimes of each source file with the mtime of this fingerprint file. If any of the source files have an mtime newer than that of the fingerprint file, the build is considered dirty and is rebuilt, otherwise fresh
• How is the fingerprint file generated? rustc outputs its own fingerprint file ending in .d: target/debug/deps/ra_rustc_wrapper_bug-19dfec7f77f210bd.d. After a build completes, Cargo reads this file and updates its fingerprint file based on the contents of this .d file.

Now, when rust-analyzer's RUSTC_WRAPPER is invoked with a check build:

• The wrapper skips over the build and exits with code 0
• Cargo is unaware of this skip and believes the build is successful
• Cargo looks for the .d file output by rustc
• Cargo finds the file produced by the first cargo check run -- a file that is now outdated, but Cargo isn't aware of this
• Cargo updates its fingerprint file

As a result, the actual .rmeta output isn't up-to-date, but Cargo thinks it's up-to-date.

Possible solutions

• The rust-analyzer rustc wrapper could delete the .d file as part of skipping the build. The path to the .d file can be constructed through a combination of --out-dir, --crate-name, and -C metadata.
• Cargo could somehow be made aware that the build didn't happen.

---

rust-analyzer version: rust-analyzer 0.0.0 (9a1ee18e4d 2025-07-21)

rustc version: rustc 1.88.0 (6b00bc388 2025-06-23)

editor or extension: repros on the command line

relevant settings: none

repository link (if public, optional): https://github.com/sunshowers/ra-rustc-wrapper-bug

code snippet to reproduce: see repro

[sunshowers]

cc @davidbarsky who I've been chatting with about the bug

[Veykril]

Good news then, the wrapper is becoming unnecessary soon
#20047 #20101

That is, if you try a 1.89.0 toolchain or higher the wrapper shouldn't be used by rust-analyzer anymore.

(we have a couple of these bug reports, we should close them once 1.89 stabilizes)

[sunshowers]

Thanks @Veykril -- that's good to know. Do you think it's still worth fixing this bug? It's been hurting my workflow for several months, and I suspect I'll be using Rust 1.88 and below for some time. I think the fix to delete the .d file isn't too difficult overall.

[Veykril]

I don't see a reason not to if we can do this as outlined. Seems simple enough 👍