NLL: turn off migration mode

[pnkfelix]

Transition fully to NLL (instead of the migration mode currently used).

• one currently opts into this via #![feature(nll)] or -Z borrowck=mir
• It gets rid of AST-region inference errors, causing some code to be accepted that is rejected by migrate mode, and changing diagnostics in many cases.
• Potential blockers:
  • ~NLL-sound bugs that affect NLL mode without migration
    • NLL accepts higher-ranked subtype that non-NLL rejects #57642 (needs test for now)
  • Diagnostic differences
    • List: https://github.com/rust-lang/rust/issues?q=is%3Aopen+is%3Aissue+label%3ANLL-diagnostics
  • Migrate compare-mode=NLL test differences to revisions (to ease stabilization PR and meticulously check diagnostic differences)
  • Two-phase borrows (2PB) question: [NLL] prohibit "two-phase borrows" with existing borrows? #56254
  • Must have crater run showing no major crates fail in practice (crater run to estimate impact of full NLL transition #60680)
  • coercion becomes order dependent for higher-ranked functions #73154
    • Likely needs changes to the leak check
  • Investigate [NLL] Lifetimes errors in a function can mask later errors in same function #96331
    • Diagnostics-only change, should be fine

[pnkfelix]

Explicitly marking this P-medium for now, to reflect the fact that transitioning the 2015 edition to migrate mode is higher priority for the short term.

[nikomatsakis]

Created #60680 as an issue to do a first crater run to get a rough idea.

[jethrogb]

Is this the right place for comments/questions about the migration warning?

I got this warning:

   = warning: this error has been downgraded to a warning for backwards compatibility with previous releases
   = warning: this represents potential undefined behavior in your code and this warning will become a hard error in the future

The warning is good, my code definitely has UB. I found that I can't upgrade the warning to an error with deny/forbid(warnings). Why is that?

This code compiles and has UB on Rust 1.0, I thought NLL is only supposed to relax things? Are you still fixing any bugs in the old borrow checker?

[pnkfelix]

@jethrogb no, NLL was not meant to solely relax things. Such relaxation is the primary selling point of NLL, but we always knew it was going to introduce breakage.

The shift to NLL is coupled with a change in the underlying borrow-checker implementation to analyze MIR (instead of analyzing the higher-level AST, which is what the old borrow checker did). In order to avoid injecting breaking changes that people could not work around in the short term, we currently have NLL in a migration mode. That was added in PR #52681.

In hindsight, it may have been nice to leverage the lint system here to let people upgrade these warnings to errors. (One reason we didn't do that is that we didn't want people to #![allow(..)] the warnings here. For more on that, see also issue #34596 and PR #59658.)