Closed dhardy closed 1 month ago
IIRC MSRV bump to 1.61 was caused by criterion. Now benchmarks reside in a separate crate outside of workspace, so it no longer applies. Is there another dev dependency which has similar MSRV?
Yes; rand
uses rayon
in an example, and that uses crossbeam
which requires 1.61. Meanwhile rand_distr
uses average
as a dev-dependency which also requires 1.61.
Hm... Personally, I think it's fine to bump MSRV to 1.61 then. The motivating request was a bit strange and they have still options to not update rand
or to ignore rust-version
.
It doesn't cost us much, so I don't see why we shouldn't do this.
It's something I dislike about the rust-version
tag actually: it should be set to the minimum version we actually want to test & support while allowing users to still try building (with a warning) on older versions.
This means that MSRV is not properly tested and we get weird exceptions "we declare 1.60 support, but, well, test only for 1.61" in different places. I guess as a middle ground we could release v0.9.0 with rust-version
equal to 1.60 and update it to 1.61 in v0.9.1.
while allowing users to still try building (with a warning) on older versions
They can do it by passing the --ignore-rust-version
flag.
TIL about --ignore-rust-version
.
Then I agree; there's no reason to merge this.
CHANGELOG.md
entrySummary
MSRV = 1.60 for build, 1.61 for tests
Motivation
This was requested.
Details
The lack of full tests on 1.60 isn't documented anywhere. Should it be?