Fix concrete_solve deprecation

Author: ChrisRackauckas

src/solve.jl

@@ -230,10 +230,7 @@ function __solve(prob::DEProblem,args...;default_set=false,second_time=false,kwa
   end
 end
 
-################### Concrete Solve
-
-@deprecate concrete_solve(prob::DiffEqBase.DEProblem,alg::Union{DiffEqBase.DEAlgorithm,Nothing},
-                        u0=prob.u0,p=prob.p,args...;kwargs...) solve(prob,alg,args...;u0=u0,p=p,kwargs...)
+################### Differentiation
 
 struct SensitivityADPassThrough <: DiffEqBase.DEAlgorithm end
 
@@ -261,6 +258,26 @@ end
 ### Legacy Dispatches to be Non-Breaking
 ###
 
+@deprecate concrete_solve(prob::DiffEqBase.DEProblem,alg::Union{DiffEqBase.DEAlgorithm,Nothing},
+                        u0=prob.u0,p=prob.p,args...;kwargs...) solve(prob,alg,args...;u0=u0,p=p,kwargs...)
+                        
+ZygoteRules.@adjoint function concrete_solve(prob::DiffEqBase.DEProblem,
+                              alg::Union{DiffEqBase.DEAlgorithm,Nothing},
+                              u0=prob.u0,p=prob.p,args...;
+                              sensealg=nothing,
+                              kwargs...)
+  @warn("concrete_solve is deprecated. Only solve is required now.
+         Additionally, new keyword arguments have been added for
+         convenience of changing parameters and initial conditions, so
+         concrete_solve(prob,alg,u0,p,args...;kwargs...) is equivalent
+         to solve(prob,alg,args...;u0=u0,p=p,kwargs...). Note that
+         the `u0` and `p` keyword arguments are not required and will
+         be picked up automatically from the `DEProblem` if the `DEProblem`
+         is defined in the differentiation context. So carry about your
+         day using `solve`: nothing special is required for adjoints!")
+  _concrete_solve_adjoint(prob,alg,sensealg,u0,p,args...;kwargs...)
+end
+
 function _solve_adjoint(prob,sensealg,u0,p,args...;kwargs...)
   if isempty(args)
     _concrete_solve_adjoint(prob,nothing,sensealg,u0,p;kwargs...)