old rand_core versions import from core::intrinsics module

[mejrs]

In rust-lang/rust#157082 (comment) we discovered that, through old rand_core versions, many crates are relying on the accidental stabilization of (in this case) core::intrinsics::transmute. We would like to be able to get rid of the hack required to support this, and the first step for that is fixing it in affected projects.

The affected rand_core versions are:

rand_core-0.1.0
rand_core-0.1.0-pre
rand_core-0.2.0
rand_core-0.2.1
rand_core-0.2.2
rand_core-0.3.0
rand_core-0.4.0
rand_core-0.4.2
rand_core-0.5.0 (but not 0.5.1)

The required change is in impls.rs

-use core::intrinsics::transmute;
+use core::mem::transmute;

Would you be willing to fix these and create new minor releases for these versions? I'd be willing to do it myself, but as I understand this, I'd need write access to create new branches.

Thanks in advance, mejrs on behalf of the Rust project.

[RalfJung]

Based on what we see in crater, rand_core-0.1 is not very relevant. But having semver-compatible fixes for 0.2, 0.3, and 0.4 would be great (in increasing order of importance, as one would expect 0.4 has the most use among these -- it alone accounts for more than half of the regressions across the entire crater run).

[newpavlov]

It's a bit surprising that those versions are still in use (I guess by mostly unmaintained crates).

The fix releases seems easy enough (though a bit annoying) to do. But it has to be done as part of https://github.com/rust-random/rand since the listed versions were developed there at the time. Publishing would also have to be manual.

@dhardy
Could you handle it?

[RalfJung]

In absolute numbers there were around 5200 regressions attributed to rand-core 0.4.x. That's a small fraction of the overall crater run, but it's still a lot.

Based on a tiny amount of random sampling, even crates released "just" 4 or 2 years ago seem to depend on ancient versions of criterion which then pulls in ancient versions of rand.

[dhardy]

It may just be a one-line change, but dealing with the test infrastructure for anything that old is quite painful in my experience (especially since this is before MSRV testing was properly supported and Cargo from that era struggles to deal with today's crates.io database). Oh, and we were using Travis for testing at the time.

I'll give it a go ...

[dhardy]

Done: 0.1.1, 0.2.3, 0.3.2, 0.4.3. Basic local testing only; it turns out these all have the same MSRV (1.22.0).

[RalfJung]

Thanks a lot ❤️

[thomcc]

Done: 0.1.1, 0.2.3, 0.3.2, 0.4.3. Basic local testing only; it turns out these all have the same MSRV (1.22.0).

It's very nice that you did this. I often complain about compiler breakage going unfixed in old versions of crates, and while I totally get why that is (it's pretty tedious to do), it's really nice to see it fixed. Thank you.