Type aliases twice as slow to compile as closed type families.

Hello guys! This bug is awkard at last to me. So I notiiced today, that using type aliases in my code takes twice as long to copile than using closed type families with just single member. So to be most precise - when I replace the following line:

type XQ1 m a = Targets (Match m a)

with

type family XQ1 m a where XQ1 m a = Targets (Match m a)

the compilation time drops from 15s to 7s (this type alias is heavily used across the application). I think it could be somehow related to the famous #8095 bug, but I can of course be wrong here.

Unfortunetally the codebase is pretty big and its not so easy to make a minimal example, but I'll try to work on it during the upcoming weekend (please notify me if that would not be necessary).

And some statistics from -dshow-passes:

When using type alias:

*** Parser:
*** Renamer/typechecker:
*** Desugar:
Result size of Desugar (after optimization)
  = {terms: 10,303, types: 66,949, coercions: 16,383,834}
*** Simplifier:
Result size of Simplifier iteration=1
  = {terms: 10,722, types: 75,391, coercions: 16,507,338}
Result size of Simplifier iteration=2
  = {terms: 10,550, types: 74,437, coercions: 16,504,770}
Result size of Simplifier iteration=3
  = {terms: 10,543, types: 74,347, coercions: 16,504,734}
Result size of Simplifier
  = {terms: 10,543, types: 74,347, coercions: 16,504,630}
*** Tidy Core:
Result size of Tidy Core
  = {terms: 10,846, types: 75,522, coercions: 16,504,729}
*** CorePrep:
Result size of CorePrep
  = {terms: 14,823, types: 96,430, coercions: 16,504,729}
*** ByteCodeGen:
Upsweep completely successful.
*** Deleting temp files:
Warning: deleting non-existent /var/folders/y9/km24zv3d4yl711mwv85vwljc0000gp/T/ghc64965_0/ghc_95.c
Warning: deleting non-existent /var/folders/y9/km24zv3d4yl711mwv85vwljc0000gp/T/ghc64965_0/ghc_94.o
Ok, modules loaded: Prologue, Type.Bool, Type.Container, Type.List, Type.Set, Data.Cata, Type.Promotion, Data.Bits.Mask, Control.Monad.State.Dependent, Control.Monad.Poly, Control.Applicative.Poly, Type.Regex, Type.Either, Data.Reprx, Data.Impossible, Type.Cache.TH, Main, Control.Lens.Utils, Control.Lens.Wrapped.Utils, Data.Text.CodeBuilder, Data.Text.CodeBuilder.Tok, Data.Text.CodeBuilder.Builder, Data.Text.CodeBuilder.Doc, Data.Binary.Instances.Missing, Data.Default.Instances.Missing, Data.String.Class, Data.Text.Class, Data.Convert, Data.Layer, Data.Container.Class, Data.Container.List, Data.Functor.Utils, Type.Operators, Type.Show, Type.Wrapped, Data.Container.Opts, Data.Container.Poly, Data.Convert.Base, Data.Convert.Instances, Data.Convert.Instances.Map, Data.Convert.Instances.Num, Data.Convert.Instances.Text, Data.Convert.Instances.TH, Data.Convert.Bound, Data.Bits.Base.
*** Parser:
*** Desugar:
*** Simplify:
*** CorePrep:
*** ByteCodeGen:
...

and with type families:

*** Parser:
*** Renamer/typechecker:
*** Desugar:
Result size of Desugar (after optimization)
  = {terms: 10,303, types: 67,609, coercions: 162,249}
*** Simplifier:
Result size of Simplifier iteration=1
  = {terms: 10,722, types: 75,721, coercions: 323,027}
Result size of Simplifier iteration=2
  = {terms: 10,550, types: 74,767, coercions: 320,473}
Result size of Simplifier iteration=3
  = {terms: 10,543, types: 74,677, coercions: 320,437}
Result size of Simplifier
  = {terms: 10,543, types: 74,677, coercions: 320,333}
*** Tidy Core:
Result size of Tidy Core
  = {terms: 10,846, types: 75,852, coercions: 320,432}
*** CorePrep:
Result size of CorePrep
  = {terms: 14,823, types: 96,760, coercions: 320,432}
*** ByteCodeGen:
Upsweep completely successful.
*** Deleting temp files:
Warning: deleting non-existent /var/folders/y9/km24zv3d4yl711mwv85vwljc0000gp/T/ghc64863_0/ghc_95.c
Warning: deleting non-existent /var/folders/y9/km24zv3d4yl711mwv85vwljc0000gp/T/ghc64863_0/ghc_94.o
Ok, modules loaded: Prologue, Type.Bool, Type.Container, Type.List, Type.Set, Data.Cata, Type.Promotion, Data.Bits.Mask, Control.Monad.State.Dependent, Control.Monad.Poly, Control.Applicative.Poly, Type.Regex, Type.Either, Data.Reprx, Data.Impossible, Type.Cache.TH, Main, Control.Lens.Utils, Control.Lens.Wrapped.Utils, Data.Text.CodeBuilder, Data.Text.CodeBuilder.Tok, Data.Text.CodeBuilder.Builder, Data.Text.CodeBuilder.Doc, Data.Binary.Instances.Missing, Data.Default.Instances.Missing, Data.String.Class, Data.Text.Class, Data.Convert, Data.Layer, Data.Container.Class, Data.Container.List, Data.Functor.Utils, Type.Operators, Type.Show, Type.Wrapped, Data.Container.Opts, Data.Container.Poly, Data.Convert.Base, Data.Convert.Instances, Data.Convert.Instances.Map, Data.Convert.Instances.Num, Data.Convert.Instances.Text, Data.Convert.Instances.TH, Data.Convert.Bound, Data.Bits.Base.
*** Parser:
*** Desugar:
*** Simplify:
*** CorePrep:
*** ByteCodeGen:
...

As we can see the type aliase (used in one single place (!)) produces (100 x) more coertions than type families: 16,383,834 vs 162,249. This is really interesting thing and it seems related to the linked bug above.