ci: remove more unused files and directories in free runners

[marcoieni]

Before

From the mingw-check job in another PR:

After

From the mingw-check job in this PR:

try-job: aarch64-gnu
try-job: aarch64-gnu-debug

[rustbot]

r? @Mark-Simulacrum

rustbot has assigned @Mark-Simulacrum.
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

[marcoieni]

@bors try

[bors]

⌛ Trying commit 8c7d76c with merge 13ce81b...

[bors]

☀️ Try build successful - checks-actions
Build commit: 13ce81b (13ce81b18428322e8d4db038af49ad4526c60536)

[marcoieni]

@rustbot ready

[Mark-Simulacrum]

I'm wondering if we should be more aggressive here. For example, I could imagine that instead of listing out various binaries in /usr/local/bin and such, we'd instead try to delete everything except exactly what we need. On Linux at least presumably that "set of needed things" can be quite small, perhaps even not coming from GitHub at all. (e.g., use a minimal ubuntu docker image's contents as the base)

IIUC, it should be possible to create a mount point at (say) /mnt for /, rm -rf /, then (roughly) cp /mnt/minimal-root-base / or so.

I'm fine with playing whack a mole too, but it seems like a long journey...

r=me with nit fixed or not

[marcoieni]

I like the approach of specifying what files are needed instead of the unnecessary ones, but instead of applying to the whole ubuntu image, I would apply it to single directories where useful. E.g. we could list the binaries we need from the /usr/local/bin/ directory instead of listing the binaries we don't.

In theory this can also be applied to packages. E.g. list the packages you need and delete the ones that don't depend on them 💡

I'll merge this PR as it is, but I'll definitely keep in mind this principle for the next ones :)

[marcoieni]

@bors try

[bors]

⌛ Trying commit 7f38958 with merge b85d92a...

[jieyouxu]

I'm wondering if we should be more aggressive here. For example, I could imagine that instead of listing out various binaries in /usr/local/bin and such, we'd instead try to delete everything except exactly what we need. On Linux at least presumably that "set of needed things" can be quite small, perhaps even not coming from GitHub at all. (e.g., use a minimal ubuntu docker image's contents as the base)

That honestly might be a good approach to make GHA environmental changes less likely to influence build/test outcomes (e.g. accidentally relying on something from GHA env that can change between image updates)

[bors]

💔 Test failed - checks-actions

[marcoieni]

weird github error

@bors try

[bors]

⌛ Trying commit 7f38958 with merge 8496d5a...

[bors]

☀️ Try build successful - checks-actions
Build commit: 8496d5a (8496d5a75af53da3512bd9711768b91c2096ca76)

[marcoieni]

@bors try

[bors]

⌛ Trying commit 7f38958 with merge 72240b3...

[bors]

☀️ Try build successful - checks-actions
Build commit: 72240b3 (72240b3eee182202a6a46944a3c6a17880e639fd)

[marcoieni]

@bors r=Mark-Simulacrum rollup=never

[bors]

📌 Commit 7f38958 has been approved by Mark-Simulacrum

It is now in the queue for this repository.

[bors]

⌛ Testing commit 7f38958 with merge cfe9ffc...

[bors]

☀️ Test successful - checks-actions
Approved by: Mark-Simulacrum
Pushing cfe9ffc to master...

[rust-timer]

Finished benchmarking commit (cfe9ffc): comparison URL.

Overall result: no relevant changes - no action needed

@rustbot label: -perf-regression

Instruction count

This benchmark run did not return any relevant results for this metric.

Max RSS (memory usage)

This benchmark run did not return any relevant results for this metric.

Cycles

This benchmark run did not return any relevant results for this metric.

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 788.14s -> 788.303s (0.02%)
Artifact size: 347.85 MiB -> 347.81 MiB (-0.01%)