[ConstraintSystem] Tweak the generic operator partition heuristic.

[hborla]

Before checking if an argument type to an applied disjunction conforms to a known protocol, grab its fixed type. This helps in cases where the argument type is a type variable, but the solver has already bound the type variable to a conforming type, e.g. Array. For example:

protocol P {}

struct S1: P {}
struct S2: P {}

func test(arr1: [S1], arr2: [S2]) -> [P] {
  return arr1.map { $0 as P } + arr2.map { $0 as P }
}

When the solver gets to the generic operator overload partition for +, it knows to attempt the Sequence overloads first because it already knows the result of .map is an array (and map is bound first because there are fewer overloads).

I haven't added a test because this case still doesn't scale well, but this fix mitigates the impact of changing overload binding order from #35025. Cases like this only happened to compile before because the solver happened to attempt Array.+ first.

Resolves: rdar://73892488

[hborla]

@swift-ci please test source compatibility

[hborla]

@swift-ci please smoke test compiler performance

[hborla]

@swift-ci please smoke test

[hborla]

@swift-ci please test compiler performance

[swift-ci]

Compilation-performance test failed

[swift-ci]

Compilation-performance test failed

[hborla]

@swift-ci please test source compatibility

[hborla]

@swift-ci please smoke test

[hborla]

@swift-ci please smoke test compiler performance

[swift-ci]

Summary for main smoketest

Unexpected test results, excluded stats for Kingfisher, SwiftyTextTable, SourceKittenFramework, Commandant, Yams, ReactiveMapKit, swiftlint, SWXMLHash, Result, ReactiveSwift, Alamofire, ReactiveCocoa, SwiftLintFramework

No regressions above thresholds

• Debug
  • debug brief
  • debug detailed
• Release
  • release brief
  • release detailed

Debug

debug brief

Regressed (0)

name	old	new	delta	delta_pct

Improved (0)

name	old	new	delta	delta_pct

Unchanged (delta < 1.0% or delta < 100.0ms) (1)

name	old	new	delta	delta_pct
time.swift-driver.wall	222.7s	223.9s	1.3s	0.57%

debug detailed

Regressed (0)

name	old	new	delta	delta_pct

Improved (0)

name	old	new	delta	delta_pct

Unchanged (delta < 1.0% or delta < 100.0ms) (4)

name	old	new	delta	delta_pct
file-search.NumFrameworkLookups	12,889	12,884	-5	-0.04%
file-search.NumIncluded	61,458	61,374	-84	-0.14%
file-search.NumMultiIncludeFileOptzn	231	231	0	0.0%
file-search.NumSubFrameworkLookups	10,310	10,310	0	0.0%

Release

release brief

Regressed (0)

name	old	new	delta	delta_pct

Improved (0)

name	old	new	delta	delta_pct

Unchanged (delta < 1.0% or delta < 100.0ms) (1)

name	old	new	delta	delta_pct
time.swift-driver.wall	291.2s	291.1s	-84.5ms	-0.03%

release detailed

Regressed (0)

name	old	new	delta	delta_pct

Improved (0)

name	old	new	delta	delta_pct

Unchanged (delta < 1.0% or delta < 100.0ms) (4)

name	old	new	delta	delta_pct
file-search.NumFrameworkLookups	12,884	12,884	0	0.0%
file-search.NumIncluded	61,374	61,374	0	0.0%
file-search.NumMultiIncludeFileOptzn	231	231	0	0.0%
file-search.NumSubFrameworkLookups	10,310	10,310	0	0.0%

[hborla]

Looks like a bunch of source compat projects are failing an assertion after the frontend, but no failures in the type checker