macrotest
macrotest copied to clipboard
MSRV violation
macrotest
no longer compiles on rust v1.34.2 because serde_json
v1.0.73 depends on itoa
v1.0.1 which doesn't compile on that version. See (duplicate build)[https://github.com/Emoun/duplicate/runs/4636793495?check_suite_focus=true].
As far as I can tell serde_json
has no MSRV, which means macrotest
should depend on a specific version to avoid violating MSRV.
I would suggest raising the MSRV of macrotest
to 1.36 which is the MSRV of serde_json
. This crate is a crate for testing and there doesn't seem to be much worth sticking around to support a very old compiler.
See also https://github.com/rust-lang/api-guidelines/pull/227#discussion_r527102913 and https://github.com/rust-lang/api-guidelines/pull/227/commits/38e610d13446dd81a09c81ec6121037706ee74d4:
Some crate bumps MSRV of the crate itself due to MSRV bump of the dev-dependency, but in some cases, the crate itself can be compiled with an older version, even if the dev-dependency requires a newer version. I'm not sure if it's really preferable to recommend
cargo test
instead ofcargo build
here. AFAIK there are many projects that do not runcargo test
on MSRV, but only runcargo build
. (e.g., serde, syn, regex, tokio, futures, etc...)
In practice, dev-deps often have much more lenient MSRV policy than then crate under test. Using
build
rather thantest
works around that.
macrotest should depend on a specific version to avoid violating MSRV.
I think this is far worse than violating the MSRV because if other dependencies requires the recent version of serde_json, this causes compile error.
I think the right place to do this is in the end-user's Cargo.toml or Cargo.lock (example).
serde_json
has no MSRV
No. serde_json v1.0.73's MSRV is Rust 1.36. See rust-version field in serde-json's Cargo.toml.
if other dependencies requires the recent version of serde_json, this causes compile error
Doesn't cargo just compile it twice?
serde_json v1.0.73's MSRV is Rust 1.36. See rust-version field in serde-json's Cargo.toml.
Thanks, i missed that. Though, does the project maintain the MSRV across versions?
Regardless, since serde_json
's MSRV is higher than macrotest
's, it seems to me fixing the version would be needed to comply with the current MSRV.
I'm not strictly against macrotest
upping it's MSRV, I've been thinking about doing it for duplicate
. However, I think it should be part of a major version bump.
Doesn't cargo just compile it twice?
It will be compiled twice if the major versions (or minor versions if the major version is 0) are different. In other cases, it will be conflict and cause compile error. See https://github.com/smol-rs/async-process/pull/4 for a specific example.
However, I think it should be part of a major version bump.
AFAIK, many cornerstone libraries (include serde_json) don't consider an MSRV increase as a breaking change. This comment by matklad is one of the explanations that I find particularly reasonable as to why such a policy should be adopted.
See also:
- https://github.com/rust-lang/api-guidelines/pull/227
- https://github.com/rust-lang/api-guidelines/discussions/231
- https://github.com/rust-lang/api-guidelines/discussions/123
AFAIK, many cornerstone libraries (include serde_json) don't consider an MSRV increase as a breaking change.
Yeah, my opinion is not mainstream sadly, and would therefore also fully respect if macrotest
decides to only bump the minor version.