Conversation
|
The MSRV issues still need to be investigated. |
|
Ah yes: criterion depends on clap v4, the latest release of which depends on Rust 1.64 (2022-09-22) whereas our MSRV is 1.56 (2021-10-21). Bumping an MSRV in a patch release is a common enough policy. I guess our options are:
|
|
Attempting to reproduce now (rebased on master): 1.70.0 is now about six months old. While I wouldn't discount the option of pushing the MSRV to this for the next release of Rand, I'm not happy about usage of a dependency with a history of updating its MSRV to somewhat recent releases frequently. We could use a pinned version of Debian stable currently uses rustc 1.63.0, so we should probably support that. @newpavlov @vks should we do this? (Full test on MSRV with pinned |
|
... at least until Cargo is MSRV aware. |
|
I would prefer to exclude benchmarks (and possibly dev dependencies) from our MSRV guarantees. I think we could only rely on MSRV-aware Cargo once our MSRV supports it, so this would be quite far in the future. |
|
I agree with @vks. I also think that it could be worth to separate benchmarks, so they would not be built as part of tests. Especially considering that |
I forgot that we already do: #1275. So maybe all we need to do is update that (if needed) and use a separate CI test for MSRV which only builds? |
|
I've been trying to produce a (with various versions of proc-macro2, or indeed other crates). The error isn't particularly helpful. Google finds a couple of other cases, but no real answers. Other than this, At this point, I'm wondering if we simply can't support rustc 1.60.0? |
|
What about excluding our benchmarks from MSRV guarantees? It would be a problem if we want to compare performance across Rust versions, but I don't think benchmarks are core functionality that must work across all versions. |
I can't even get a build to work now: Possibly this |
|
I have a working build using rustc 1.61.0, but it fails on 1.60 as above. I'll post a new PR. |
|
@vks can you rebase this now? |
|
I rebased it, but some criterion dependencies are still causing trouble. |
|
I moved the benchmarks to their own crate, which should also help with compilation times for dev builds. I also fixed MSRV 1.61 compatibility by pinning Rayon. |
No description provided.