Avoid double-caching when ccache is installed in PATH #2524

avikivity · 2025-12-20T20:12:31Z

On Linux, ccache is typically installed in $PATH as /usr/lib64/ccache/g++
or similar. If we keep it there, all compilations will be cached by both
ccache and sccache.

While the user could easily disable ccache with CCACHE_DISABLE, it's
reasonable to assume many will forget and will have their disk space
doubly consumed by both caches. Better to recognize this and disable
ccache under sccache.

This patch does this resolving the input compiler name to an absolute
path, with these rules:

if a directory in PATH contains a component ccache, it is skipped
if the compiler exists in a directory in PATH, but is a symlink
to a target ccache, it is skipped

These two tests will catch most ccache masquerade directories.

Fixes #2519

codecov-commenter · 2025-12-20T20:16:43Z

Codecov Report

❌ Patch coverage is 92.90780% with 10 lines in your changes missing coverage. Please review.
✅ Project coverage is 72.12%. Comparing base (e454f4e) to head (83fc260).

Files with missing lines	Patch %	Lines
src/compiler/nvcc.rs	52.94%	8 Missing ⚠️
src/util.rs	98.29%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2524      +/-   ##
==========================================
+ Coverage   72.04%   72.12%   +0.07%     
==========================================
  Files          66       66              
  Lines       36953    37085     +132     
==========================================
+ Hits        26623    26746     +123     
- Misses      10330    10339       +9

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

avikivity · 2025-12-20T21:00:10Z

update: fix rustfmt error

avikivity · 2026-01-07T12:07:06Z

@sylvestre please review

codspeed-hq · 2026-01-10T14:06:20Z

CodSpeed Performance Report

Merging this PR will not alter performance

_{Comparing avikivity:disable-ccache (83fc260) with main (e454f4e)}

Summary

✅ 54 untouched benchmarks
⏩ 2 skipped benchmarks¹

2 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

avikivity · 2026-01-20T13:50:22Z

@sylvestre can I help in any way to push this along?

sylvestre · 2026-01-22T13:02:10Z

src/util.rs

+    use std::env;
+    use std::path::Path;
+
+    const CCACHE_DIRS: &[&str] = &["/usr/lib/ccache", "/usr/lib64/ccache"];


The CCACHE_DIRS list only contains Linux paths. Consider adding macOS Homebrew paths (/usr/local/opt/ccache/libexec, /opt/homebrew/opt/ccache/libexec) or adding a comment explaining why they're omitted

also, it would be nice to be able to configure things somewhere

We can generalize it to any path that contains a component ccache.

About configuration, is it still needed if we just filter out directories with ccache components?

sylvestre · 2026-01-22T13:02:58Z

src/util.rs

+/// This prevents double-caching when ccache is installed and has
+/// /usr/lib/ccache or /usr/lib64/ccache in PATH, since those directories
+/// contain wrapper scripts that would call ccache.
+pub fn filter_ccache_from_path(env_vars: Vec<(OsString, OsString)>) -> Vec<(OsString, OsString)> {


Function allocates a new Vec even when no ccache dirs are present in PATH. Consider short-circuiting to return the input Vec unchanged when filtering is not needed

sylvestre · 2026-01-22T13:26:04Z

src/compiler/compiler.rs

        } = self;
+        // Filter out ccache directories from PATH to avoid double-caching
+        // when ccache is also installed on the system.
+        let env_vars = filter_ccache_from_path(env_vars.to_vec());


Filter is only applied to SingleCompileCommand but NvccCompileCommand (in nvcc.rs) also implements CompileCommandImpl and may need the same treatment it is intentional ?

Not intentional, I'll take a look.

sylvestre · 2026-01-22T13:26:28Z

sorry for the latency

avikivity · 2026-01-22T15:37:55Z

v2:

filter out any directory with a ccache component, so it also works on MacOS
avoid allocations if no ccache directories are present
apply to detect_c_compiler
apply to nvcc

avikivity · 2026-01-22T15:40:02Z

v3:

apply rustfmt

avikivity · 2026-01-22T16:42:31Z

v4:

use slice instead of Vec

avikivity · 2026-01-25T15:48:51Z

I believe the performance regression report is spurious, the code doesn't touch the macro finder code.

ahartmetz · 2026-01-27T12:19:50Z

So I'm not the maintainer here and this is merely a suggestion. I'm a bit uncomfortable with the brittleness of this approach to recognizing ccache. An additional, somewhat - perhaps less - brittle heuristic that could be added on POSIX platforms is to recognize when gcc, g++, clang, clang++ [...] is a symlink to a binary called "ccache". Assuming that ccache works like icecream and sccache (which btw supports symlinks compiler -> sccache now).

avikivity · 2026-01-27T12:24:05Z

So I'm not the maintainer here and this is merely a suggestion. I'm a bit uncomfortable with the brittleness of this approach to recognizing ccache. An additional, somewhat - perhaps less - brittle heuristic that could be added on POSIX platforms is to recognize when gcc, g++, clang, clang++ [...] is a symlink to a binary called "ccache". Assuming that ccache works like icecream and sccache (which btw supports symlinks compiler -> sccache now).

I believe it's unnecessary - it's unlikely people will have directories in path named ccache that host the real compilers instead of ccache masquerades.

I'm happy to make that change if the maintainers think it is necessary though.

ahartmetz · 2026-01-27T12:25:48Z

I believe it's unnecessary - it's unlikely people will have directories in path named ccache that host the real compilers instead of ccache masquerades.

I'm happy to make that change if the maintainers think it is necessary though.

I don't mean ANDing the two conditions, I mean ORing. The false positive rate of a compiler binary linking to a binary called ccache, which is actually not ccache, should be negligible.

avikivity · 2026-01-27T12:28:51Z

So I'm not the maintainer here and this is merely a suggestion. I'm a bit uncomfortable with the brittleness of this approach to recognizing ccache. An additional, somewhat - perhaps less - brittle heuristic that could be added on POSIX platforms is to recognize when gcc, g++, clang, clang++ [...] is a symlink to a binary called "ccache". Assuming that ccache works like icecream and sccache (which btw supports symlinks compiler -> sccache now).

I believe it's unnecessary - it's unlikely people will have directories in path named ccache that host the real compilers instead of ccache masquerades.
I'm happy to make that change if the maintainers think it is necessary though.

I don't mean ANDing the two conditions, I mean ORing

To clarify: IF the directory contains a path component ccache, OR if the looked-up compiler is a symlink to something named ccache, THEN skip this compiler executable?

Makes sense.

ahartmetz · 2026-01-27T12:30:16Z

I don't mean ANDing the two conditions, I mean ORing

To clarify: IF the directory contains a path component ccache, OR if the looked-up compiler is a symlink to something named ccache, THEN skip this compiler executable?

Makes sense.

That's what I meant, yes

ahartmetz · 2026-01-27T12:33:08Z

(Background: I MAY have had a distributed compiler, which I compiled myself, installed in [obscure]/local/bin at some point, but I have never renamed the distributed compiler binary to something else than its usual name - which would probably break something because it avoids calling itself by checking its binary name)
Edit: on second thought, I think such tools usually just eliminate their own binary name / path / inode, whatever it may be, from the compiler search - but still, the ccache / icecream / sccache binary is usually called exactly that, with the exception of sccache, which required hardlinks until recently.

On third thought, what the hell was I thinking, if it's not called gcc or g++ etc, it won't be found as a compiler at all, so you'd need to resolve the symlink to ccache, which is... doable but where this maybe gets a bit "too much".

avikivity · 2026-02-01T14:43:16Z

update:

rebased
changed approach - the utility function resolves relative compiler names to absolute paths, skipping over directories with ccache components or files that as symlinks that resolve to ccache.

On Linux, ccache is typically installed in $PATH as /usr/lib64/ccache/g++ or similar. If we keep it there, all compilations will be cached by both ccache and sccache. While the user could easily disable ccache with CCACHE_DISABLE, it's reasonable to assume many will forget and will have their disk space doubly consumed by both caches. Better to recognize this and disable ccache under sccache. This patch does this resolving the input compiler name to an absolute path, with these rules: - if a directory in PATH contains a component `ccache`, it is skipped - if the compiler exists in a directory in PATH, but is a symlink to a target `ccache`, it is skipped These two tests will catch most ccache masquerade directories. Fixes mozilla#2519

avikivity · 2026-02-01T14:44:11Z

update:

applied rustfmt

src/util.rs

avikivity force-pushed the disable-ccache branch from ec91704 to 608d5c6 Compare December 20, 2025 20:59

sylvestre force-pushed the disable-ccache branch 2 times, most recently from 1fe7bb9 to 4bb5579 Compare January 13, 2026 15:33

sylvestre reviewed Jan 22, 2026

View reviewed changes

avikivity force-pushed the disable-ccache branch from 4bb5579 to 2a7f25c Compare January 22, 2026 15:36

avikivity force-pushed the disable-ccache branch from 2a7f25c to ecce16b Compare January 22, 2026 15:39

avikivity force-pushed the disable-ccache branch from ecce16b to ce9f95b Compare January 22, 2026 16:42

avikivity force-pushed the disable-ccache branch 2 times, most recently from d8913d5 to 6681201 Compare February 1, 2026 14:42

avikivity force-pushed the disable-ccache branch from 6681201 to 83fc260 Compare February 1, 2026 14:44

ahartmetz reviewed Feb 2, 2026

View reviewed changes

src/util.rs Show resolved Hide resolved

Avoid double-caching when ccache is installed in PATH #2524

Are you sure you want to change the base?

Avoid double-caching when ccache is installed in PATH #2524

Conversation

avikivity commented Dec 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecov-commenter commented Dec 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

avikivity commented Dec 20, 2025

Uh oh!

avikivity commented Jan 7, 2026

Uh oh!

codspeed-hq bot commented Jan 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will not alter performance

Summary

Footnotes

Uh oh!

avikivity commented Jan 20, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

sylvestre commented Jan 22, 2026

Uh oh!

avikivity commented Jan 22, 2026

Uh oh!

avikivity commented Jan 22, 2026

Uh oh!

avikivity commented Jan 22, 2026

Uh oh!

avikivity commented Jan 25, 2026

Uh oh!

ahartmetz commented Jan 27, 2026

Uh oh!

avikivity commented Jan 27, 2026

Uh oh!

ahartmetz commented Jan 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

avikivity commented Jan 27, 2026

Uh oh!

ahartmetz commented Jan 27, 2026

Uh oh!

ahartmetz commented Jan 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

avikivity commented Feb 1, 2026

Uh oh!

avikivity commented Feb 1, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

avikivity commented Dec 20, 2025 •

edited

Loading

codecov-commenter commented Dec 20, 2025 •

edited

Loading

codspeed-hq bot commented Jan 10, 2026 •

edited

Loading

ahartmetz commented Jan 27, 2026 •

edited

Loading

ahartmetz commented Jan 27, 2026 •

edited

Loading