fix: recursive type inference issue

See chalk-lab/Mooncake.jl#594 (comment)

Author: MilesCranmer

Project.toml

@@ -1,7 +1,7 @@
 name = "DynamicExpressions"
 uuid = "a40a106e-89c9-4ca8-8020-a735e8728b6b"
 authors = ["MilesCranmer <[email protected]>"]
-version = "2.0.0"
+version = "2.0.1"
 
 [deps]
 ChainRulesCore = "d360d2e6-b24c-11e9-a2a3-2a2ae2dbcce4"

src/Evaluate.jl

@@ -346,9 +346,8 @@ end
 
 # TODO: Hack to fix type instability in some branches that can't be inferred.
 # It does this using the other branches, which _can_ be inferred.
-@unstable function _get_return_type(tree, cX, operators, eval_options)
-    # public Julia API version of `Core.Compiler.return_type(_eval_tree_array, typeof((tree, cX, operators, eval_options)))`
-    return eltype([_eval_tree_array(tree, cX, operators, eval_options) for _ in 1:0])
+@unstable function _get_return_type(tree, cX, eval_options)
+    return Core.Compiler.return_type(deg0_eval, typeof((tree, cX, eval_options)))
 end
 
 # This basically forms an if statement over the operators for the degree.
@@ -361,7 +360,7 @@ function inner_dispatch_degn_eval(
 ) where {T,degree}
     return _inner_dispatch_degn_eval(
         tree, cX, Val(degree), operators, eval_options
-    )::(_get_return_type(tree, cX, operators, eval_options))
+    )::(_get_return_type(tree, cX, eval_options))
 end
 @generated function _inner_dispatch_degn_eval(
     tree::AbstractExpressionNode{T},