fix: Allow unknown guidance when try to coerce unsized

[Austaras]

A temporary workaround for #15984 and #11847.

[flodiebold]

I think it'd be good to compare the results of analysis-stats with this change on a few code bases (at least RA itself).

[Veykril]

Probably something other than r-a (or in addition to r-a), since r-a doesn't tend to exert the type system in a lot of ways.

[Austaras]

I think it'd be good to compare the results of analysis-stats with this change on a few code bases (at least RA itself).

Is there any guide how to do that?

[Veykril]

rust-analyzer analysis-stats project-to-analyze-path / cargo run --release -p rust-analyzer -- analysis-stats project-to-analyze-path

[Austaras]

I tested this PR again master branch for several repos

r-a

// master
Body lowering:       3.40s, 36ginstr, 311mb                                                                                                                                       
  exprs: 683668, ??ty: 35 (0%), ?ty: 111 (0%), !ty: 3                                                                                                                             
  pats: 160801, ??ty: 4 (0%), ?ty: 4 (0%), !ty: 0

// coerce
Body lowering:       3.30s, 36ginstr, 311mb                                                                                                                                       
  exprs: 683693, ??ty: 35 (0%), ?ty: 703 (0%), !ty: 3                                                                                                                             
  pats: 160826, ??ty: 4 (0%), ?ty: 6 (0%), !ty: 0

hyper

// master
Body lowering:       877.68ms, 10ginstr, 66mb                                                       
  exprs: 60549, ??ty: 3554 (5%), ?ty: 2920 (4%), !ty: 1                                             
  pats: 9081, ??ty: 973 (10%), ?ty: 912 (10%), !ty: 6

// coerce
Body lowering:       873.75ms, 10ginstr, 66mb                                                       
  exprs: 60549, ??ty: 3554 (5%), ?ty: 2924 (4%), !ty: 1                                             
  pats: 9081, ??ty: 975 (10%), ?ty: 912 (10%), !ty: 6

rust

// master
Body lowering:       14.90s, 158ginstr, 1303mb                                                                                                                                                                                      
  exprs: 3438690, ??ty: 137025 (3%), ?ty: 134094 (3%), !ty: 3565                                                                                                                                                                    
  pats: 620050, ??ty: 42969 (6%), ?ty: 35569 (5%), !ty: 896

// coerce
Body lowering:       14.39s, 158ginstr, 1303mb                                                                                                                                                                                      
  exprs: 3438690, ??ty: 137038 (3%), ?ty: 134518 (3%), !ty: 3547                                                                                                                                                                    
  pats: 620050, ??ty: 43178 (6%), ?ty: 35569 (5%), !ty: 896

I also run against webrender and huggingface/candle, but the results are identical.

Are the numbers good or bad?

[Veykril]

??ty means type contains {unknown}/{error}
?ty means type is unknown
!ty means type mismatch

The r-a case looks weird given ?ty, which base-line rust-analyzer did you use? the one before this PR or another?

Overall it seems to increase the contained unknowns a bit in some projects (regresses).

[Austaras]

I built 95e047 as base to test against.

[Veykril]

I can't find that commit. Anyways, the changes on rust-analyzer itself seem rather bad.

[Austaras]

Fine. Maybe this is not the right way to go.