cargo-cache reports wrongly disk usage

[tesuji]

cargo-cache reports wrongly disk usage

% ls -l /home/lzutao/.cargo/bin
total 74M
-rwxr-xr-x 12 lzutao lzutao 6.2M Oct 28 10:10 cargo*
-rwxr-xr-x 12 lzutao lzutao 6.2M Oct 28 10:10 cargo-clippy*
-rwxr-xr-x 12 lzutao lzutao 6.2M Oct 28 10:10 cargo-fmt*
-rwxr-xr-x 12 lzutao lzutao 6.2M Oct 28 10:10 cargo-miri*
-rwxr-xr-x 12 lzutao lzutao 6.2M Oct 28 10:10 clippy-driver*
-rwxr-xr-x 12 lzutao lzutao 6.2M Oct 28 10:10 rls*
-rwxr-xr-x 12 lzutao lzutao 6.2M Oct 28 10:10 rustc*
-rwxr-xr-x 12 lzutao lzutao 6.2M Oct 28 10:10 rustdoc*
-rwxr-xr-x 12 lzutao lzutao 6.2M Oct 28 10:10 rustfmt*
-rwxr-xr-x 12 lzutao lzutao 6.2M Oct 28 10:10 rust-gdb*
-rwxr-xr-x 12 lzutao lzutao 6.2M Oct 28 10:10 rust-lldb*
-rwxr-xr-x 12 lzutao lzutao 6.2M Oct 28 10:10 rustup*
% du -sch /home/lzutao/.cargo/bin
6.2M    /home/lzutao/.cargo/bin
6.2M    total
% cargo-cache 
Cargo cache '/home/lzutao/.cargo':

Total:                            987.55 MB
  12 installed binaries:           77.55 MB
  Registry:                       874.45 MB
    Registry index:               133.50 MB
    1411 crate archives:          169.27 MB
    867 crate source checkouts:   571.68 MB
  Git db:                          35.56 MB
    4 bare git repos:              35.56 MB
    0 git repo checkouts:              0  B

Meta

• cargo-cache 0.3.3 (813ad5b 2019-10-22)

[matthiaskrgr]

Can't look in detail right now but I'm wondering if the difference is related to symlinks.

[tesuji]

Yes, rustup hard-links those binaries shim like cargo, rustc.

[slonik-az]

As of cargo-cache version 0.5.1 the ~/.cargo/bin disk usage reported on MacOS is still incorrect. The reason seams to be that many toolchain binaries are hard-links of the same executable while cargo-cache counts their sizes separately.

$ cargo cache
Cargo cache '/Users/foobar/.cargo':

Total:                          106.62 MB
  14 installed binaries:        106.61 MB
  Registry:                          0  B
    0 registry indices:              0  B
    0 crate archives:                0  B
    0 crate source checkouts:        0  B
  Git db:                         9.53 KB
    2 bare git repos:             3.24 KB
    1 git repo checkouts:         6.29 KB

while the actual disk usage is 5 times smaller

$ du -hS ~/.cargo/bin
22M	/Users/foobar/.cargo/bin

[matthiaskrgr]

I added a quick filter to ignore symlinks when counting sizes in the dont_count_symlinks branch.
https://github.com/matthiaskrgr/cargo-cache/tree/dont_count_symlinks
does this work better for you?

[heftig]

Although ignoring symlinks isn't wrong, it wouldn't help in this case. The files aren't symlinks, they're hardlinked (see the 12 link count in the original listing).

You need maintain a set of the (dev, ino) of files you've already seen and ignore any file with the same metadata.