From the “Cargo Home” docs page (live, source), referring to what
directories should be cached in CI:
It should be sufficient to only cache the following directories across builds:
bin/registry/index/registry/cache/git/db/
This doesn’t seem to be sufficient when the job runs cargo install to
install tools, like Cargo subcommands (e.g., cargo-raze). If a job
installs a binary and only caches these directories, a re-run of the job
will fail at cargo install time with
error: binary `cargo-raze` already exists in destination
Add --force to overwrite
because the binary exists but Cargo has no record that it has been
installed. Sample failure:
https://github.com/tensorflow/tensorboard/runs/1367174061
It sufficed for me to also cache .crates.toml and .crates2.json.
Maybe only the former is needed; I included the latter to be safe.
Especially since these are dotfiles, it would be nice to document that
they, too, should be cached.
I would submit a patch, but I don’t know exactly what the right
recommendation is.
cargo version 1.47.0 on Linux (Debian/Ubuntu)
From the “Cargo Home” docs page (live, source), referring to what
directories should be cached in CI:
This doesn’t seem to be sufficient when the job runs
cargo installtoinstall tools, like Cargo subcommands (e.g.,
cargo-raze). If a jobinstalls a binary and only caches these directories, a re-run of the job
will fail at
cargo installtime withbecause the binary exists but Cargo has no record that it has been
installed. Sample failure:
https://github.com/tensorflow/tensorboard/runs/1367174061
It sufficed for me to also cache
.crates.tomland.crates2.json.Maybe only the former is needed; I included the latter to be safe.
Especially since these are dotfiles, it would be nice to document that
they, too, should be cached.
I would submit a patch, but I don’t know exactly what the right
recommendation is.
cargo version 1.47.0 on Linux (Debian/Ubuntu)