@Pietro Albini pointed out that the Crater results for 1.36 are pretty bad (700 regressions on the check only+rustdoc run) with a link to this comment on the run.
Here is a github query that covers unprioritized stable-to-beta regressions that are not already tagged with T-libs nor T-rustdoc
that yields 8 issues, lets walk through them.
(going to go oldest first)
first: "ICE in NLL borrow_check/universal_regions when bounding a trait object with nothing" #61315
looks like @Matthew Jasper is on top of it. I'll make it P-high just so I don't have to think about it.
next: "regression: failed to parse proc-macro input" #61470
petrochenkov has diagnosed this as expected
breakage fallout from a bug fix.
It seems like the best we can do here is try to update
agatan/squeeze-rs ? Its not really a bug we'd do anything in the compiler to address.
agatan/squeeze-rs looks like it might be abandonware; last update was in Nov 2017...)
closing #61470 as not-a-bug/wont-fix.
next: "regression: conflicting implementations of trait for Into<Box>" #61471
#61471 references two root crates that caused regressions downstream. However, the one that caused the most problems, the owner chimed in saying they had deprecated the crate and linked to the relevant issue. So the question is: Is crater supposed to be blacklisting such cases? (Can it, when there are problems downstream...?)
Okay after much investigation, I decided only one crate is actually affected here (the rest are abandoned, I'm pretty sure). I filed an issue against the affected crate and closed #61471.
next: "regression: overflow evaluating, Send/Sync?" #61472
(presumably due to PR #60444; the running hypothesis that niko and I share is that most instances of this are cases where the input code was unsound, and thus the fallout is expected. I don't think we can readily warning-cycle this change.)
I do however want to double check the note made by @Jake Goulding. I assume the discrepancy between behavior observed using
rustdocis due to some additional code paths being exercised by doc examples.