This interacts with a lot of different issues. see rust-lang/rust#132325 for a summary of our candidate preference

• we avoid assembling impls shadowed by where-bounds #226

merging candidates allows more code to compile

duh! e.g. tests/ui/lifetimes/conflicting-bounds.rs

global where-bounds impl and detection

Keep 'static in the where-bounds, lazily normalize aliases in them: rust-lang/rust#140260

disabling normalization

The underlying reasoning here is "where-bounds should cover impls and alias-bounds if the where-bound is more general, as using the covered candidate may add undesirable constraints".

• normalization does not use impls covered by param env candidates #12
• using impl candidates shadowed by where-bound overflows by using impl with only super trait bounds #76
• alias-relate when normalizing in unnormalized env causes overflow #89
• Supertrait elaboration causes unnecessary param-env shadowing #100
• eagerly normalizing the param_env triggers impl shadowing in unavoidable ways #219

unnecessary region constraints by shadowing

• Solver prefers param-env bound over object bound #98
• higher-ranked goals in trait goal candidate selection #120 leak check

using impls for normalization results in unnecessary region constraints

• using shadowed impls may add unnecessary region constraints #125

implementation bugs

• discard_impls_shadowed_by_env forces unnecessary ambiguity #96