Closed workingjubilee closed 6 months ago
All modified and coverable lines are covered by tests :white_check_mark:
Project coverage is 91.79%. Comparing base (
7d62045
) to head (c4e1957
).
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
It's not clear to me if this comment remains actionable: https://github.com/chronotope/chrono/blob/d79a0ee1aa2723dafdf4b8e9eab6a95016eec405/.github/workflows/test.yml#L164-L167
We should squash all these commits, since some of them cannot pass CI independently. Can just do it during merge in this csae.
Very nice, and glad to see this gone 😄.
I suppose there is not really a way to test this except by making a release. Since when does cargo's resolver have this ability? Is it supported by our MSRV 1.61? I can't find it, but according to archive.org it is documented this way for at least three years so 👍.
You can experiment with a Cargo project using cargo update --precise
and try first releasing 0.4.38-rc.0, if you like? But yes, the feature that will prevent this has essentially "always been there": it was part of the original resolver, so there was probably no release in which it was absent, and it has definitely been there since at least 1.56 (when they started using resolver = "2"
).
It is old but somewhat true again. Could you replace it with something like "We can't use
--all-features
because of the conflictingrkyv-*
features."?
Done.
Thank you!
Per Cargo's documentation on its resolver:
Thus, while it may be unusual to make this change on a SemVer minor release, it is de facto not a "breaking change" to dependents to remove the rustc-serialize feature. Dependents on it will simply not update to a later version of the chrono crate (i.e. 0.4.38 or later), as Cargo will correctly identify the incompatibility, and will keep giving them a maximum of
chrono = { version = "0.4.37", features = ["rustc-serialize"] }
instead. This may not be something one should do casually, but as the dependency has been deprecated for almost a year now, and Rust plans to remove the relevant code, it seems preferable.